Release Notes¶

Release notes for LimaCharlie platform components, organized by date.

Subscribe to these release notes

Every entry below is also published as a feed, so a new release can reach you instead of you checking this page. Entries that change an event schema or break an existing integration are additionally published on their own feed.

Everything: RSS or JSON Feed
Endpoint Agent only: RSS
Web App only: RSS
Adapters only: RSS
Extensions only: RSS
CLI only: RSS
Platform announcements only: RSS
Breaking and schema changes only: RSS

To route a feed into Slack, run /feed subscribe https://docs.limacharlie.io/10-release-notes/endpoint-agent.xml in the channel that should receive it.

For discussion and email notification of the same releases, set the Platform Updates category in the community forum to Watching. For service availability rather than releases, subscribe on the status page.

2026-09-16¶

Web App 6.4.0¶

A personal LimaCharlie Bots workspace, a Code Security overview backed by real scan coverage, and GitLab and Bitbucket source-control connections.

New Features¶

(Bots) LimaCharlie Bots workspace: a personal workspace at /sessions/workspace where you switch between named bots, build groups, and watch them hand work to one another. It carries a persistent roster, direct and group conversations, @mentions, saved drafts and read state, published files, questions and approvals, pause and stop, and explicit review and continuation. Each bot shows live progress: elapsed time, the tool it is running, what it is waiting on, and what is queued behind it. Handoffs between bots play as a compact bubble row above the composer, so an exchange is readable rather than gone before you see it. For users in the rollout, the AI Terminal button opens the workspace; the legacy terminal is one click away.
(Bots) Start a fresh chat and browse history: a New chat action gives a bot a clean runtime while keeping its configuration, saved memory, routines, and readable history. Past chats open read-only, with artifact downloads still available. Drafts and scroll positions stay separate per chat. Group conversations get the same New chat button, history picker, and read-only history view.
(Bots) Bot memories: a Memories browser in conversation details lists a bot's saved notes with names and timestamps. You can read, edit, and delete them. In a group you pick whose bank to open. Changes apply to new sessions and never touch chat history.
(Bots) Routines: scheduled routines move out of conversation details into their own destination at /sessions/workspace/routines, linked from the roster, the conversation header, and the quick switcher. Search every routine, filter by upcoming, paused, or needs attention, manage schedules, and inspect run history. A "Make this recurring" action on a bot's result builds the routine from the original request instead of asking you to retype it.
(Bots) Bot avatars: the free-text avatar field becomes a searchable gallery of 24 illustrations with live preview and a deterministic default. Existing initials, emoji, and icon names still render.
(Bots) Auto-approve is on by default: every eligible bot in a chat starts with auto-approve on. Per-bot opt-outs are kept, persist across a reload, and survive a change of group membership.
(Bots) A starter team that knows the platform: the generic Lead, Hunter, Response, and Reviewer roster is replaced by six LimaCharlie specialists — Sensor Fleet Operator, D&R Rulesmith, LCQL Hunter, CloudSec Cartographer, MailSec Investigator, and Automation Fabric Engineer. Each has focused operating instructions that route through the LimaCharlie AI skills.
(Cloud Security) Code Security overview: Code Security opened on a repository table, which hid most of what it does. A new default Overview tab leads with a prioritized fix queue — one dependency upgrade, and how many findings and repositories it clears — plus scanner coverage across active repositories, the capabilities your connected GitHub Apps actually hold, and a code to image to running-workload evidence strip. Repository, image, and registry inventories keep their tabs.
(Cloud Security) Code Security's controls follow detected App permissions: the UI used to infer what it could do from connection settings, and told orgs whose App already held the permission to go install a second one. It now reads the App's real capabilities. Available shows no notice, unavailable names the missing permissions and links to the install page, and unknown says the check could not be made and will be retried. Where an org has several connections, the ones that cannot publish are named. Capabilities are asked per repository, so a repository outside an App's selection is no longer told a check is available. Every answer states how fresh it is.
(Cloud Security) GitLab and Bitbucket Cloud connections: both can now be added and edited from the Cloud Security provider wizard like every other provider, with inline validation that matches the backend, the exact token scopes named on the permissions step, repository badges, and "Open on provider" links.
(Cloud Security) Pull-request checks from the Code page: Code Security can scan what a pull request introduces and publish the result as a check run a team can make required, but nothing reacted to a pull_request delivery, so that check never fired. Three webhook rules now cover it — rescan a repository after a push, check a pull request, and re-check one whose base branch moved. That last case GitHub reports without any push, so the check kept the conclusion it had and left a green gate on a diff nobody scanned. The Code page installs and removes the rules, with a badge per rule; previously they had to be written by hand with limacharlie hive set.
(Sensors) Event Collection reads its event types from the extension: the Event Types picker was built from a bundled list that had fallen behind and was missing 40 real events, so a recently added event could not be selected at all. It now reads the list from the Exfil extension, so new events appear as the platform adds them. A rule naming an event the extension no longer reports keeps that value instead of being written back truncated.
(Platform Logs) Audit log detail panel: selecting a log used to squeeze the table into a narrow split. The detail now opens as a drawer over the table, with the event type and time in its header, the payload below, and Escape to dismiss.
(Extensions) Varist scan paths: an "Online only" checkbox on the sensor picker. An offline sensor is dropped rather than queued, so listing every sensor in the org was inviting a scan that would never run.
(Outputs) Elastic output: the is_compress_request parameter is described, so the gzip toggle renders as an optional checkbox.
(Platform) Startup animation: app startup shows one of three security animations — radar, shield, or secure stream — in place of a generic spinner. It follows the theme and respects reduced motion.

Bug Fixes¶

(Cloud Security) Static analysis runs unless you turn it off, but the policy form only ever wrote engines that were on, so unticking it produced a record identical to one that never mentioned it and the engine kept running. The denial is now written, and the policy form, the Code setup checklist, and the repository drawer stop reporting the engine off on every record written before it was named.
(Cloud Security) A failed connection preflight blocks the save instead of decorating it. A required check that a source-control token is missing a scope is where an unusable connection is turned away, and "saved anyway" meant the refusal surfaced hours later in a scan status. Sync now in the edit flow is blocked on the same condition. An optional failure still saves, and a provider type with no tester is never blocked.
(Cloud Security) The push-rescan recipe keyed on routing and hostname alone, which any adapter in the org could reuse. It now requires a signature-verified webhook delivery.
(Cloud Security) Code Security's SBOM download accepts only https URLs, so a backend-supplied link can no longer carry another scheme into an anchor.
(Cloud Security) Scan Diagnostics showed the last pass's targets, which drop to zero whenever no repository is due. An org with hundreds of scanned repositories read as empty. It now shows standing tracked-repository coverage, with last-pass activity as its own labelled section.
(Cloud Security) The Code Actions setup step linked a file in a private repository. It now links the public setup guide.
(Bots) Routines render and edit correctly when the API omits zero clock fields, and a timezone alias such as the default Etc/UTC validates.
(Bots) A failed turn can start a new chat. New chat was disabled on a failed conversation with a tooltip asking you to finish work that had already released its runtime, and a provider-level failure such as a usage limit reproduced on every retry, so the conversation could never leave that state. A refused new chat now says why instead of showing the word "conflict".
(Bots) Avatars are visually distinct and color selection is its own control, a stale prompt closes when its task stops, the message composer grows to fit a multiline draft, chat follows new messages while you are at the bottom, and duplicate approvals, answers, and sends are prevented.
(Platform) The app no longer stays on an older deployment. "Update now" requested service-worker activation and reloaded immediately, which could interrupt the handoff and serve the old cache. The reload now waits for the new worker to take control, and unversioned files and SPA routes revalidate instead of sitting in the browser cache.
(Add-ons) Logged-out visitors opening an Add-ons marketplace page saw a blank screen. The public route now renders inside a layout, so the page has height and the marketplace sidebar appears.
(Extensions) Varist scan results showed a blank Status badge, and picking any Risk Level emptied the list, because the scan result often omits the rating flag. The flag is derived from the numeric rating when it is absent, and the Status column sorts on that value.
(Platform) Light-mode warning colors are darkened for readable contrast, including investigation lane indicators and event markers.
(Organizations) The Organizations list no longer shows a second spinner from the New Organization dialog while it loads plans and templates.

2026-09-08¶

Endpoint Agent 5.3.9¶

New Features¶

New dir_find command walks a directory tree and reports file metadata and hashes, with include and exclude patterns, size and age filters, and caps on how many files and bytes a single run may touch.
New file_grep command searches file contents for literal patterns, with optional case-insensitivity, a per-file match limit, surrounding context bytes, and an option to include binary files.
New container_list command inventories containers and images on a Linux host — runtime, image reference and digest, creation time, state and main process id — covering Docker and Podman, with containers under containerd and CRI-O identified from their process cgroups.
Artifact retrieval can now collect several files in one task: files are selected by pattern under a root directory, each paired with a caller-supplied payload id, and bounded by limits on file count and total bytes.
New repo_list command reports the git checkouts on a host along with their identity — worktree path, remotes and their URLs, the current branch and the commit it resolves to, whether HEAD is detached, the last fetch time, and optionally the declared submodules — so a repository can be tied to the hosts that have it cloned without a follow-up file read per candidate.

Bug Fixes¶

On Linux, fixed an upgrade leaving the sensor stopped. Stopping the service also terminated the upgrade process running inside it, so the new binary was installed but never started, and no restart policy brought it back.
On Linux, fixed a cloud-driven uninstall being terminated by the service stop it had just issued, which left the service definition, the sensor's data directory and the staged uninstaller behind.
On Linux and macOS, the sensor now shuts down cleanly when stopped. Termination signals were not handled, so every service stop, restart, reboot and upgrade was reported to the cloud as a crash — nearly the entire crash volume from those platforms. A stop now waits for the sensor's own teardown (measured at about 4 seconds with all modules loaded) rather than being immediate.
On Windows, fixed the sensor terminating itself when Windows briefly refused to return a process listing, which happens transiently while the process list is changing under it.
On CentOS and RHEL 7 (and other hosts with glibc 2.17), fixed a crash during a clean shutdown.
Fixed crash reports being lost on hosts that could not reach the cloud on their first connection attempt, real crashes then being reported as clean exits, and the sensor occasionally reporting a crash context it had manufactured itself.
The sensor's log file is now created readable only by its owner. It was world-readable, and it can contain DNS queries, file paths and command lines. A host that has already been logging keeps its existing file's permissions until that file is removed.
Hardened DNS collection against two out-of-bounds reads reachable from a crafted DNS packet the host merely observes: a datagram too short to hold a DNS header, and a compression pointer in the final byte of a packet.

Improvements¶

macOS and Linux crashes now produce a full crash report — stack trace, loaded modules and fault detail — where previously they left only a marker saying the last run had not exited cleanly. Debug information is published per release, so a report from either platform can be resolved to source without a follow-up on the affected host.
32-bit Windows crash reports now carry a complete stack trace instead of the two or three frames the optimized build allowed.
The Linux sensor download is substantially smaller — roughly 45% for the service binary and 30% for the Host Based Sensor module — because debug information is now published separately instead of being shipped to endpoints. Every upgrade transfers less.
The sample macOS MDM profile now marks the sensor's background service and system extension as managed, so they can no longer be switched off or removed by a user from System Settings.
Releases now include the Varist file scanning engine, so it can be turned on by configuration instead of requiring a custom build.

Web App 6.3.0¶

A full testing workspace for D&R and False Positive rules, endpoint agents as a vulnerability source in Cloud Security, and a Query Console that holds far less memory.

New Features¶

Rule testing workspace for D&R rules: the rule page's stored tests and its replay panel are now one surface. A fullscreen workspace replays every stored case against the rule as currently edited and reports pass or fail per case, with a summary in the section header. A replayed event can be saved onto the matching test side in one click, so an ad-hoc probe becomes a stored case. A second workspace runs the rule over stored telemetry as a historical scan and renders each hit as a row, with a "Keep as non-match test" button that stores the hit's matched payload on the suite. Each suite row also gets a play button beside Run all, so one case can be re-run without disturbing the other verdicts.
Rule logic tree: the detect block renders as its boolean tree with the replay evaluation drawn on it. Green nodes evaluated true, red nodes false, and dashed nodes were short-circuited. Clicking an operation shows what it read, what it expected, and what it saw in the event. Multi-event runs get a per-event switcher, a failed run pre-selects the operation that killed the match, and stateful rules render their with child and with descendant subtrees. Anything the view cannot reconstruct from the trace falls back to the raw lines rather than guessing.
Test coverage on the rules list: a sortable Tests column shows each rule's case count, and a coverage readout beside the filter controls reports over the rows on screen. Narrowing to Custom rules answers how much of what your own team wrote is tested, instead of being dragged down by managed and service rules.
A rule now tells you how much it matches: saving an enabled rule replays the last hour of stored telemetry in the background and reports how often the rule matched. A count at flood level (100 an hour or more) comes back as a warning. A rule that is too broad announces itself seconds after the save instead of hours later in Detections. False Positive rules report suppressions and never warn, because a busy suppressor is a working one. Metadata-only saves skip the check, and a failed check stays silent rather than blocking the save.
The save gate matches the server: saving a rule with failing tests always failed at the API, but the dialog offered "Save anyway". It now blocks the save and routes into the rule testing workspace. A report, add var or del var action with no name is also caught inline in the editor rather than as an opaque API error.
Edit rules in YAML or JSON, with import and export: the D&R editor gains a YAML/JSON toggle over the detect and respond editors, file export of the stored shape (detect, respond, tests) as .yaml or .json, drag-and-drop import validated before it reaches the editors, and smart paste on the detect editor. The editor sniffs each rule's format on open and warns before a conversion would drop YAML comments. False Positive rules get the same toolbar.
False Positive rules get the full testing experience: the testing workspace and the rule logic tree are now on the False Positive rule page as well.
Promote a Query Console query to a D&R rule: an LCQL query converts to a detect block with no AI involved. Sensor selectors, event types and the filter expression tree all map across. Time ranges and projections are dropped with an explicit note. Anything that would make the rule broader than the query, such as an unsupported operator or an incomplete query, stops the conversion instead.
Test a detection against its rule: a new action in the detection viewer opens the originating D&R rule's testing workspace with the triggering event preloaded. It appears when the detection carries a real sensor event and the viewer holds the replay permission.
Endpoint agents are a vulnerability source in Cloud Security: a new Policies tab turns the endpoint-agent vulnerability lane on, so CVEs a LimaCharlie agent reports become Cloud Security findings. The Coverage tab reports the lane beside cloud compute coverage, a host with no cloud provider behind it renders as a first-class resource with its own findings and pivots, and a finding's kill chain badges the hops a LimaCharlie agent covers.
More to go on when triaging a vulnerability: findings now carry the exploit band (KEV overdue, KEV due, exploit likely, exploit probable, elevated, baseline), whether a fix is available, which lane observed it (cloud, agent, or both), and CISA's KEV due date. Each of these is absent when the platform did not report it, and an absence is shown as unknown rather than as a negative answer.
VEX and SLA policy editors: two cloudsec_policy types had no authoring surface, and both sit behind a headline feature. Policies gains a VEX tab for the OpenVEX documents that move a finding to "not affected" with author provenance, including a client-side parse preview, and an SLA tab for the clock behind every finding's due date. A finding can now be marked not affected from the finding itself.
Coverage page: the Inventory tab's sensor coverage becomes a Coverage page that reports each lane separately, so a workload with an agent but no vulnerability source no longer reads as fully covered. The org-level pending count for endpoint-agent inventory is reported instead of being dropped.
Shared fixes: when one remediation resolves several findings, the finding shows the other findings it covers and drills through to the exact cause.
Integrations panel: Cloud Security accepts SARIF, CycloneDX, LimaCharlie scanner reports and third-party asset batches, and every one of those doors was API-only and named nowhere in the app. Settings gains an Integrations tab that documents them in-product, deep-linkable at ?tab=integrations.
Identity and asset filters: the Access and Assets screens gain server-side selectors, and sensitivity counts survive a cross-filter instead of resetting.
Compliance reason codes: a control that was not assessed now says why, rather than sitting blank. Azure authorization evidence has its own view, and a finding whose authorization coverage has a gap is classed as such.
Managed Kubernetes clusters are typed: the platform is retyping managed clusters from ComputeInstance to KubeCluster, with namespaces, workloads, services and roles alongside. Every Cloud Security list and map keyed on resource type now understands them, so a cluster keeps its place in the compute population and is classified the same way on every path that reaches it.
The Providers list names the provider: the source badge and the record name become one identity column with the brand glyph, the record name, and the provider's full name spelled out under it. A provider type this build has never seen degrades to a neutral badge carrying the raw token instead of rendering blank.
Elastic output: the is_create_action parameter is described, so the output editor renders it as an optional checkbox instead of an unlabelled required field.

Bug Fixes¶

The event tree could crash the app. An atom cycle in routing/parent and routing/this recursed until the stack blew and tripped the global error boundary. The tree and the ancestor walk now track the atoms they have visited, read a self-reference as "no parent", and truncate instead of spinning.
Atoms and sensor ids interpolated into Insight URLs are now escaped. Event data decides what an atom contains, and one carrying a / rerouted the request to a different endpoint that answers with a success and no event, so the ancestor chain silently truncated with no error. Detection ids are escaped on both the read and the delete.
The extension definition editor showed every permission unchecked no matter what the extension held, and saving from that state dropped every permission the user did not happen to re-tick.
Selecting a rule from halfway down the D&R sidebar jumped the list back to the top. The sidebar now keeps its scroll position while the editor still resets.
Pasting a second rule while a smart-paste suggestion was still showing kept the stale card and never offered the new suggestion's action.
The inline "mark as crown jewel" prefill seeded the resource family instead of the resource kind, which the platform rejected for data stores and identities and mis-seeded for compute.
Renaming an AI Session with a name over 200 characters failed with a raw API error. The input now shows an inline error and the mutation never fires. Length is counted in code points, so an emoji-heavy name the API accepts is not rejected.
A tooltip on truncated text now re-measures when its container resizes, so it appears and disappears as the column changes width instead of going stale.

Improvements¶

The Query Console holds far less memory. A loaded search now releases the rows on pages far from the viewport and fetches them again from the continuation token when you scroll back. What a search retains no longer grows with every row it has ever loaded, so a very large result set stays usable.
The Detections live feed holds up under a flood. Every incoming batch used to re-sort the whole accumulated list. Batches now merge into the already sorted rows.
Section collapsers can be dragged with touch as well as a mouse, and stop dragging cleanly when the tab is hidden or focus is lost.

2026-09-07¶

Extensions: Email Security bulk remediation, verdict revisions and the managed-detection switch¶

New Features¶

Bulk remediation over a selection you name. Any search result — a filtered page of the queue, a hunt, a list of ids in a file — can now be remediated at the provider in one operation, with the campaign sweep's preview-then-confirm discipline: POST /actions/bulk/preview reports each message's current state and the distinct-mailbox blast radius and mints a confirmation derived from that exact selection, POST /actions/bulk/execute returns a bulk_id immediately and paces the provider writes in the background, and GET /actions/bulk/{bulk_id} reports per-message outcomes and running counts. Capped at 500 per call, refused rather than truncated above it. limacharlie mailsec message bulk-action wraps the pair, reads its selection from --msg-uuids, --input-file or standard input, and carries the outcome in its exit code.
Verdict revisions are documented end to end. POST /messages/{msg_uuid}/verdict re-judges one message in the analyst or ai seat with a required rationale, and GET /messages/{msg_uuid}/revisions serves the whole append-only chain with the prior state each revision displaced. Re-recording a verdict a message already carries is a success reporting applied: false, not an error. Revising requires mailsec.act rather than mailsec.set, because it can promote a message's evidence to the 400-day lane and emits an EMAIL_VERDICT.
A resolved user report can be reopened. POST /reports/{report_id}/reopen (limacharlie mailsec report reopen) puts a report back in the queue and keeps its resolution columns, so the row still reads "previously resolved by …" rather than erasing what is being disputed. It is the escape hatch the abuse-mailbox auto-resolver depends on.
The managed detection pack has a documented switch. The managed_rules policy record, the console's Settings toggle, and the set_managed_rules / get_managed_rules extension actions all write the same thing. Absent means enabled; a record that states no value is refused; a change reaches the pipeline in seconds, with a five-minute backstop.
Link detonation is documented: what triggers it (a suspicious verdict with links, up to five URLs per message), the organization-wide claim that makes a campaign cost one fetch, every refusal reason, and what a mode: detonation verdict revision means — including that it never overrules a person or an agent.
Time to verdict is a real number. coverage's overview.processing_latency_p95 reports p50/p95/p99 with the basis it measures, the denominator it measured over, and — when there is no number — which of the two honest reasons applies. Backfilled and re-driven mail is excluded so a recovery cannot report a latency measured in days.
New Troubleshooting page: the entitlement gate (mail that arrives while unsubscribed is never read and is never backfilled), watch renewals, silent-mailbox stalls, EMAIL_INGEST_ERROR, connection handover during a platform update, and the 30-day deletion clock.
Custom rules now document scope versus the ? path wildcard — only scope correlates two conditions on the same link or attachment — and state plainly that a rule using lookup, or a post_verdict rule, cannot be backtested.

See Bulk Remediation, Detections & Verdicts, Policy Reference and Troubleshooting.

2026-09-04¶

Extensions: Email Security tenant purge¶

New Features¶

Tenant purge for Email Security: everything Email Security holds for an organization can now be deleted on request — the message index and the up-to-400-day evidence lane, campaigns, sender profiles, the product's action audit trail, user (abuse-mailbox) reports, stored raw messages and their parsed copies, link-detonation results, and the organization's provider connection and policy configuration. The purge also stops the mail connections at Microsoft 365 and Google Workspace, so the provider stops sending notifications instead of repopulating the index. It is irreversible, requires Owner-level authority (mailsec.act, billing.ctrl and user.ctrl together), and is recorded in the organization's audit log with the caller's identity and an optional reason.
The purge is two steps, and previews by default: limacharlie mailsec tenant purge prints the warning and mints a single-use confirmation token that expires in five minutes, and destroys nothing until that token is passed back with --confirm. A purge that reports complete: false did not finish and can simply be re-run with a fresh token. The Python SDK exposes the same pair as Mailsec.prepare_tenant_purge() and Mailsec.purge_tenant().
Email Security data is now also deleted automatically: 30 days after an organization unsubscribes from the extension — resubscribing at any point inside that window cancels the scheduled deletion — and immediately when the organization itself is deleted.

See Data retention and deletion, the CLI notes and DELETE /tenant.

2026-08-31¶

Endpoint Agent 5.3.8¶

New Features¶

New usb_list_devices command lists every USB device on the host — device class, manufacturer, product and serial number, and the interfaces each device exposes — so a composite device such as a flash drive that also presents a keyboard is reported as both.

Bug Fixes¶

On Windows, fixed a severe delay opening files on network shares. The sensor asked Windows for a fully resolved path on every file open, which a network redirector can only answer with one round trip per path component — measured at 6.4 seconds to open a single document over SMB. Network volumes now use the path the application opened with; local volumes are unchanged.
On Windows, fixed a possible system crash when non-cached asynchronous file I/O reached the file filter.
Fixed process events being attributed to the wrong parent on hosts that create processes rapidly. A process ID reused within roughly 70 seconds of its previous owner's exit could be linked to that dead process, which also prevented the real parent from ever being found.
On Windows kernel acquisition, fixed process and registry events being returned out of chronological order once their internal buffer filled.
On Windows kernel acquisition, fixed network connection and termination events being discarded a whole buffer at a time — nearly every event in a burst — instead of only the oldest, and with no record of the loss.
On Windows, the crash minidump is no longer written into C:\Windows\System32, where it accumulated and was never removed; it now goes to the sensor's data directory and is removed on uninstall.
Module images staged by a run that ended uncleanly are now cleaned up at startup instead of accumulating in the sensor's data directory.
Hardened process enumeration and termination tracking against memory-allocation failure and a malformed process list, either of which could leak memory or crash the sensor.
On Windows, fixed payload downloads and artifact uploads failing on hosts whose certificate store does not hold the server's root certificate; the sensor's embedded certificate chain was only consulted when Windows reported a partially trusted chain, not an untrusted root.
On Windows 7, fixed a crash while validating a server certificate for payload downloads and artifact uploads.

Improvements¶

Windows crash reports now carry the full fault detail — exception code, faulting address, the memory address accessed and whether it was read or written, registers, and a stack trace resolvable per module — instead of only an approximate location, so a crash can be diagnosed without a follow-up on the affected host.
Sensor modules delivered to a host now appear under a name that identifies the module rather than a generic temporary file name.
Updated embedded third-party libraries for HTTP, TLS certificate verification and archive handling.

2026-08-24¶

Endpoint Agent 5.3.7¶

New Features¶

On Linux, DNS events now report the process that sent the query or received the response, where the kernel supports it (verified on 5.15 and newer); older kernels keep delivering DNS events without a process ID. DNS events on Windows and macOS are unchanged.
On Linux, kernel telemetry — file, process, socket, network and DNS — now loads on kernel 5.4, where it previously failed to load entirely. Network isolation still requires a newer kernel.

Bug Fixes¶

On Linux, fixed all network connection telemetry being lost on kernels that cannot support network isolation; connection tracking now loads independently of isolation.
On Linux, fixed the DNS tracker staying on packet capture for the life of the sensor when it started before kernel acquisition was ready, leaving DNS events without a process ID on hosts where the kernel could provide one.
On Linux, fixed the process tracker staying on its netlink fallback under the same conditions, which could leave short-lived processes without a path or command line.
On Linux, hardened kernel DNS collection against packets whose declared payload length exceeds what they carry; such packets could corrupt the sensor's buffer and caused the DNS events behind them to be discarded.
On Linux, fixed sensor diagnostics reporting every kernel telemetry subprogram as inactive when the reporting component had simply not yet queried the kernel; the status is now omitted until it is known.
Fixed a possible deadlock when the sensor is stopped by a signal on Linux and macOS.

Improvements¶

On Linux, sensor diagnostics now report whether DNS events are being attributed to a process.

2026-08-20¶

Web App 6.2.0¶

Resizable columns on every table, more of LimaCharlie extensions moving onto first-class pages, multi-provider credentials for AI Sessions, better reporting when a search leaves events out, and keyboard navigation in the org pickers.

New Features¶

Resizable columns on every table: column resizing is now on by default for every table in the app, not just Search results.
The Query Console tells you when a search left events out: a search that sheds events used to report "Complete!" regardless, and a query whose only match was skipped rendered as "no results". A warning badge in the status row now carries the headline count ("16 events skipped"), and a notice card above the results — closed by default — carries the per-reason breakdown, an explanation of why the search moved past those events rather than stalling, and a sampled list of them.
Dedicated YARA Scanners page: ext-yara replaces the generic schema-driven renderer with a first-class page at /orgs/:oid/yara-scanners.
Dedicated File & Registry Integrity page: ext-integrity replaces the generic schema-driven renderer with a first-class page at /orgs/:oid/integrity.
Dedicated Sensor Cull page: ext-sensor-cull replaces the generic schema-driven renderer with a first-class page at /orgs/:oid/sensor-cull.
AI Sessions: multi-provider credentials: alongside Anthropic, you can now connect OpenAI, Google Gemini and OpenRouter, each with its own auth methods (API key, Azure OpenAI, Vertex for Gemini, Bedrock/Vertex for Anthropic). Session profiles gain a Provider select, session details show which provider a session ran on. See User Settings → AI Terminal.
Keyboard navigation in the org pickers: both the full-screen /orgs list and the bottom-left org switcher autofocus their search field, let Tab move into the results, arrow keys move the highlight and Enter selects. A single filtered match is preselected, so typing enough to narrow to one org and pressing Enter enters it.
Export affected hosts from Vulnerability Reporting: an Export hosts control (CSV or Excel) on the CVE detail page, the CVE drawer's Impacted Assets tab, and the package drawer's Affected hosts section. The export drains the entire listing rather than the pages the table happened to load, de-duplicates by sensor, and reports truncation instead of passing a partial file off as complete.
LOLRMM community rules render properly: LOLRMM rules are plain Sigma YAML and now open in the Sigma detail view instead of falling through to the "Unable to parse rule content" raw fallback.
Entra ID adapter on Microsoft national clouds: the Entra ID setup and edit forms gain the endpoint picker already offered by the Defender adapter, so a US Government tenant (GCC High, DoD) can be configured. The field is optional and omitted when unset, so existing commercial adapters are untouched.

Bug Fixes¶

Values loaded into a creatable multi-select rendered as empty — every existing pattern on an Artifact Collection or File & Registry Integrity rule was invisible in the edit modal. Adding one pattern to a field that looked empty silently replaced all the pre-existing ones on save.
AI Terminal state no longer bleeds between sessions: file upload and download cards stayed pinned to the session they happened in rather than appearing in every conversation, and switching sessions no longer carries over the composer draft, the retry prompt, the download bar or a stale "queued behind a blocked turn" notice. A freshly started session also no longer sits idle because an earlier session in the same tab already consumed an initial prompt.
The permissions table no longer overflows the addon and extension editors' modals, where the Write column sat off-screen until the user scrolled sideways.
Opening an app that has been deleted from a stale link now shows a "not found" zero state with a route back to the launcher instead of a load failure, and a malformed link no longer crashes the render.
Artifact Collection with no configuration saved yet reads as "no rules yet" instead of reporting an error.

Platform: Multi-Provider AI Sessions¶

New Features¶

AI provider choice for user sessions: user AI Sessions can now run on OpenAI (API key or Azure OpenAI), Google Gemini (AI Studio key or Vertex AI service account), and OpenRouter, in addition to Claude. Connect your own provider credentials under User Settings → AI Terminal (or via the API) and pick the provider on a session profile. Tool use and approvals, MCP servers, USD budget caps, hibernation, forking, and cost tracking work the same on every provider; a few capabilities stay Claude-only (plan mode, bypassPermissions, Task subagents, extended thinking). See AI Providers.

2026-08-14¶

Endpoint Agent 5.3.6¶

New Features¶

Linux hosts that cannot provide a cgroup2 mount — unprivileged containers, read-only root filesystems, kernels built without cgroup BPF support — now keep kernel-level file, process, socket and network telemetry instead of losing all of it.

Bug Fixes¶

Fixed the sensor silently discarding queued events when its outbound queue filled up; discarded events are now counted and reported.
The outbound event queue is now bounded by memory size as well as by event count, so a few very large events can no longer displace thousands of small ones.
Fixed every kernel-sourced DNS event on Linux reporting process ID 0 and being held for 10 seconds before delivery.
Fixed Linux network isolation being reported as available on hosts where the sensor could not exempt its own connection, where isolating the host would have cut off the connection needed to un-isolate it.
Fixed the sensor aborting under Wine while collecting Windows Event Log events; an incomplete Event Log bookmark implementation is now detected at the point of use and log resumption is disabled cleanly.
macOS file-access monitoring now matches file extensions case-insensitively, so a rule for .docx also matches Report.DOCX. Linux continues to match exactly.
Removing an adhoc exfil rule now returns a receipt reporting whether the rule was found and removed, matching the behavior of adding one.

Improvements¶

Sensor diagnostics now report how many events the outbound queue has discarded, their estimated size, and the current queue depth against its limits.
Sensor diagnostics now report which Linux kernel telemetry subprograms are active, so a partial kernel acquisition is visible rather than appearing healthy.
On Linux hosts without kernel DNS support, DNS collection now falls back to packet capture, and packet captures are opened only when that fallback is actually in use instead of on every Linux sensor.
The sensor now warns when file-access monitoring is enabled with no file patterns configured, instead of appearing healthy while unable to report anything.

2026-08-13¶

Web App 6.1.0¶

Several UX improvements, new remediation signal (SLA due dates, root-cause roll-ups, coverage honesty), updated UX for some extensions, and performance improvements for Query Console.

New Features¶

Cloud Security speaks one language: every remaining screen — Inventory (Resources and Third-party assets), Attack Surface, Access, the Query Console, Compliance, Policies, Settings and the MSSP fleet board — now opens on the shared tinted filter bar.
Compliance report layout: the Compliance screen becomes a charts-and-rail report — a compliance-score donut and a failures-by-severity bar chart side by side, the controls table beneath them, and the assignment cards as a rail running down beside all three.
Cloud Security fleet board modernized: /cloud-security adopts the UI language, adds two fleet rollup donuts, makes every column header a sort control, and splits freshness into sortable Last scan and Providers columns so a failing provider count is its own column rather than a badge in someone else's cell.
SLA due dates on the findings worklist: findings gain a Due column showing relative age ("in 6d", "12d ago", "today"), with a tooltip carrying the exact deadline, the SLA state, and the policy clause that set it.
Root-cause roll-ups on findings: when one finding's remediation already implies another, the child shows a "root cause" chip that links straight to its parent and the parent shows how many findings it implies.
Compliance score honesty: a framework that grades only a fraction of its controls no longer shows a bare green score. The score loses its good-result tint and states how much was actually assessed.
Vulnerability-source coverage: the Coverage tab adds a second dimension beside sensor coverage — whether anything is actually reporting CVEs on a workload — computed per scanner scope, so a workload with a sensor but no vulnerability source no longer reads as fully covered.
Real internet entry points in the kill chain: the first hop now renders the actual front door — App Service, Cloud Run URL, Kubernetes API server, function URL, load balancer, CDN, AI inference endpoint — labelled with its hostname, instead of a generic globe that said nothing about how an attacker got in.
Privilege escalation and cloud hierarchy in Topology: scope-escalation hops render as their own edge category with a distinct signature and priority on a crowded canvas, and cloud-hierarchy containers (organization, folder, project, account) draw as containers rather than workloads.
AWS boundary and trust evidence: permission boundaries, boundary-capped grants and trust-policy conditions on can-assume edges now have a screen, and a finding's cloud is read from the finding itself so vulnerability-coverage aggregates keep their cloud badge and readable resource name.
Dedicated Event Collection page: ext-exfil replaces the generic schema-driven form with a first-class page.
Dedicated Artifact Collection page: ext-artifact gains a custom made UI page with Collection Rules and PCAP Capture Rules tabs.
Dedicated Reliable Tasking page: ext-reliable-tasking moves off the generic schema renderer onto a custom page.
The shared filter bar reaches the rest of the app: the Platform Logs audit tab, False Positive Rules, and the sensor Vulnerabilities tab all move onto the same panel. The sensor Vulnerabilities tab's four bordered summary tiles become stats on the bar, and the Payloads list moves onto the standard table so dates and sizes stop wrapping.
New Org and Recent Org lists for everyone: the redesigned organization and recent-organization lists are now the default for all users.
Query Console performance: a search over a large result set now completes much faster, holds less memory and spends less time in garbage collection. Scrolling with a row selected is more performant, and result sets that previously exhausted the browser's heap now load.
Connect wizard permissions: the provider setup flow's "Permissions required" list now includes the AWS Organizations policy reads and the GCP Cloud Run / Cloud Functions viewer roles, so a customer following the least-privilege list verbatim gets the public-invoker exposure verdicts rather than silently missing them.

Bug Fixes¶

A single transient failure while polling for search results — a 502, a dropped connection, a laptop changing networks — permanently ended the search with no explanation; polling now survives it and picks the query back up.
Returning to the Query Console tab after a network reconnect silently re-ran a finished search as a brand new server-side scan, replacing the results on screen and billing the org a second time.
A crash affecting dropdown menus ("Maximum update depth exceeded") caused by an upstream react-select bug is fixed.
The Cloud Security "Failed" sync badge showed the collector error only as hover text that could not be selected or copied; the tooltip now stays open and the badge copies the raw error to the clipboard.
The Vulnerabilities tab and section no longer appear on a sensor's detail view for organizations not subscribed to vulnerability reporting, where they showed a misleading empty "0 findings" report.
The CAASM coverage-policy editor offered a capability no source could satisfy, which produced a permanent coverage gap on every device; it is removed from the picker, and policies that already contain it render as unsupported instead of blank.
SOPs and Organization Notes had no working "View Docs" link, and their editors called records "Sop" and "Org_notes"; both now link to the published documentation and use proper record type names.
The Inventory resource count rendered unformatted (373109 assets) while every other count on the screen used thousands separators, and a shared accessibility defect meant several dropdowns' labels were not associated with their control.
The "EDR everywhere" policy template button appeared to do nothing due to a read-after-write bug, and a full-screen empty state could swallow clicks on neighbouring controls.

2026-08-11¶

Endpoint Agent 5.3.5¶

New Features¶

Linux process events now identify the container a process belongs to, including its Kubernetes pod; the fields are absent for ordinary host processes.
New Windows script visibility through the Antimalware Scan Interface: scripts and .NET assemblies are reported as the host application submits them for scanning, already deobfuscated. Enabled from the cloud.
YARA memory detections now report the address of the match, so a detection points at the exact location in the target process.
YARA rules can now use the macho module to inspect Mach-O images.

Bug Fixes¶

Fixed cloud-driven sensor upgrades never taking effect on macOS. The new binary is now put in place atomically before the service is stopped, on both macOS and Linux.
Fixed the sensor sitting connected but idle for up to 10 minutes after its cloud connection was replaced, delaying collectors and telemetry; it now re-syncs immediately.
Fixed the sensor hanging indefinitely when a configured proxy accepts the connection but never answers the CONNECT request; the attempt is now bounded and retried.
Proxy replies split across packets, sent with LF-only headers, or answering with a 2xx status other than 200 are now handled correctly.
Fixed the sensor's log file being created world-writable on installed Linux sensors.
Uninstalling on Linux now removes the kernel acquisition mount point it used to leave behind, and macOS uninstall now deregisters its installer receipt.
Fixed the arm64 Debian package being refused by dpkg; it is now published as limacharlie_<version>_arm64.deb.
Payload downloads and artifact uploads are now strictly compliant HTTP, so filtering proxies and SSL inspection agents no longer drop them as malformed.
Fixed artifact uploads reporting success while delivering no data when a network device answered with a redirect.
Fixed certificate verification failures for payload downloads, artifact uploads and sensor upgrades on hosts with enterprise TLS inspection roots; verification now uses the operating system trust store alongside the built-in certificates.
Fixed the sensor deadlocking while listing process handles when Windows stopped answering a handle name query.
Fixed process hollowing detection rescanning every module in full instead of skipping the parts already matched on disk.
Fixed stateful detection tracking comparing only part of its key, which could route events to the wrong tracking group.
Fixed the sensor aborting under Wine when probing the Windows Event Log bookmark API.
Fixed a potential out-of-bounds read when matching short YARA rule namespaces.
Corrected swapped country and organization labels in the sensor's certificate issuer output.

Improvements¶

Payload download and artifact upload failures now report a readable description in the command receipt and in the local log, instead of a bare numeric code.
Transfer failures that previously all reported error 0 now report distinct codes: a sensor that has not yet received its artifact configuration, and a connection closed without a response.
Payload downloads and artifact uploads now use a full HTTP client, adding chunked response support, correct response framing and enforced timeouts.
The Linux x86_64 and i386 sensor now requires glibc 2.12 instead of 2.17, so it installs on older distributions; the optional eBPF module still requires 2.14.
A configuration update is no longer held up by an in-progress payload download.
The payload download size limit now applies to the whole transfer rather than to each resumed request.

2026-07-29¶

Web App 6.0.2¶

A rebuilt Cloud Security Risks triage table, finding ownership as a filter and column, a self-serve Cloud Security upgrade flow, and a reorganized sidebar.

New Features¶

Cloud Security Risks — dense triage worklist: the card-per-finding list is now a ruled, sortable table at roughly twice the row density, a sticky header, server-side sorting on Risk and Severity, and select-all in the header checkbox so the bulk triage bar only appears once you pick rows. Multiple UX improvements related to this change.
Owner facet and column on Risks: findings can now be filtered and scanned by who owns them. An Owner group in the facet rail lists Unassigned first, then your own account marked "(you)", then everyone else by count.
Data Security matches the Risks layout: the Data Security screen adopts the same tinted filter bar, flat facet rail and badge styling as the Risks worklist.
Self-serve Cloud Security upgrade: free-tier orgs can now upgrade without leaving the page. The trial band's CTA on the Cloud Security Overview opens an in-context upgrade modal that explains what upgrading changes, quotes the real monthly price for your org, collects payment details only if there's no card on file, and confirms on completion.
SOPs and Organization Notes move to Organization Settings: both sidebar entries move out of Automation — where they sat next to D&R rules, playbooks and lookups — into the Organization Settings group after Integrations, with matching breadcrumbs. Existing links and bookmarks to /sops and /org-notes still work.
Apps entry always visible in the sidebar: the Apps navigation item no longer disappears for users lacking the app.get permission; it renders for everyone and the page itself explains the missing permission when opened.

Bug Fixes¶

The "Cloud Security isn't enabled" screen sent its main button to the extension's configuration page, which has no way to subscribe and often showed a permission wall; it now goes to the marketplace entry where the extension can actually be subscribed, and the button reads "View extension".
Cloud Security still described itself as pre-GA and not publicly subscribable in the not-enabled gate and the feedback modal, even though it is generally available and self-serve.
The Risks lens picker showed an empty "Select..." whenever the active class filters didn't match a preset lens, making it look like no lens was applied while a filter was in fact active; it now reads "Custom".
Cloud resources were classified by their name, so a load balancer called "public-lb" was drawn as the internet node and a VM called "db-1" was drawn as a data store; classification now uses the resource's actual type. Similarly, an IAM principal named something like "public-reports@..." was labelled "Public - anyone on the internet".
The fleet board's top attack path card navigated to a retired URL and silently bounced users to the org Overview; it now opens the Attack Surface workspace, and old bookmarked path links redirect there too.
Third-party assets and reconciled inventory assets both displayed as "Third-party asset" in the graph type filter and node labels; inventory assets now read "Inventory asset".
The shared-fix strip and the "resolves N findings" count only considered one kind of root cause, so privilege-escalation causes never appeared and mixed causes were counted twice — the strip and the finding detail card could disagree on the same fix. The shared-fix card also described every cause as a firewall rule, even when the fix was an IAM role binding.
Sortable Risk and Severity column headers rendered a truncated "..." because the label plus the sort icon didn't fit their track.
In dark mode, Risks table rows swallowed their own dividers and the header band.

2026-07-28¶

Endpoint Agent 5.3.4¶

New Features¶

The registry listing command can now recursively list sub-keys down to an optional maximum depth, reporting fully-qualified key paths; oversized results are returned as partial replies flagged for follow-up.

Bug Fixes¶

Fixed corrupted process events from macOS kernel acquisition: invalid timestamps, missing file paths and command lines, and spurious phantom process records.
Fixed file transfers and the sensor's cloud connection aborting when a TLS 1.3 server sends a post-handshake session ticket, common behind SSL-inspection proxies.
Fixed repeated payload put commands failing on Windows: moving a file into place now overwrites an existing destination, matching other platforms.
Fixed payload downloads and log/artifact uploads failing when certificate verification is explicitly disabled.
Uninstalling the Windows MSI now fully removes the sensor's identity and data, so a later reinstall enrolls as a fresh sensor.
Fixed a rare crash caused by concurrent use of the sensor's random number generator, seen most often on macOS under heavy file activity.
Fixed a crash on sensor shutdown after a partially failed startup.
Fixed a potential crash when reading executable paths from the Windows registry under low-memory conditions.
Fixed assorted small latent bugs across the sensor uncovered by expanded static analysis.

Improvements¶

Windows Event Log collection now resumes from where it left off after a sensor restart, recovering events logged while the sensor was not running.
Sensor upgrade and uninstall operations now write warnings and errors to a dedicated log file in the sensor data directory.
Listing endpoint protection exclusions now reports an error code on failure, distinguishing a failed query from a host with no exclusions.
Windows WMI query failures now report a dedicated WMI error code instead of a generic failure, for example when Defender is disabled by policy.

Web App 6.0.1¶

A small fix to the Cloud Security onboarding sign-up flow.

Bug Fixes¶

Signing up from a Cloud Security onboarding link still asked which product you wanted before creating an account, even though the link had already committed you to one.

2026-07-27¶

Web App 6.0.0¶

Cloud Security goes generally available, a one-link self-serve onboarding flow, and a rebuilt Cloud Security Overview dashboard.

New Features¶

Cloud Security is generally available: the product is no longer gated behind an early-access flag, so every organization with the Cloud Security extension sees it in the sidebar.
Self-serve onboarding link: a single shareable URL - /onboard?purpose=cloud-security - now takes someone from no account to a working organization.
Cloud Security Overview rebuilt: the Overview now uses the same flat dashboard language as the Case Management dashboard.
Full-screen kill-chain view: the most-critical-path tile gains a "View full screen" action.
Trial status on the Overview: organizations on the Cloud Security trial now see a banner at the top of the Overview stating the provider-connection cap and how many are used, the trial length, and a "See plans" link — instead of only finding out when they hit the limit.
Provider-aware connect wizard diagram: the diagram in the provider setup wizard now draws what the provider you selected actually contributes — compute, network, data, identity, AI and vulnerability coverage for a cloud, or identities, apps and devices for a directory — and labels the credential path with how that provider authenticates (IAM roles, AssumeRole, GitHub App, and so on). The edge into your workspace now states the sync cadence you just chose on that step.

Bug Fixes¶

The Overview coverage bar read 100% in nearly every state, because it counted only sources that had already completed one sweep rather than the estate itself; it now reports accounts collecting, shows a real shortfall while a connected source has delivered nothing, and stays empty rather than showing 0/0 before the first scan lands.
Privilege-escalation findings were drawn with an invented "Internet -> public exposure" entry hop and a rationale claiming an open attack path, even for identity-only findings that are not internet-reachable; the exposure claim is now only made when the backend asserts it, and entitlement hops are labelled "can assume" instead of "reaches".
The generic rationale on some findings printed a bare category name such as "High ciem."; the three entitlement categories now spell out what they mean.
Verifying your email during sign-up sent you to the organization list and lost the onboarding flow you started; the verification link now returns you exactly where you left off, and the original tab no longer sits on a stale "verify your email" card after verification completes.
The choke-point sentence naming your crown-jewel resources appeared in English in every language; it is now translated across all locales.
Creating a secret inline in the provider wizard labelled the name field "New secret (required)", which read as if it were the credential itself; it now says "New secret name". Providers that take a raw key (LimaCharlie, OpenAI, Anthropic) also show a bare-key placeholder instead of suggesting a JSON wrapper.
The provider wizard and first-run artwork drew logos for internal resources on a provider LimaCharlie does not support; the illustrations no longer use them.

2026-07-26¶

Web App 5.14.0¶

A redesigned permissions editor, modernized Outputs and Artifacts experiences, and an "Overview" org diagram.

New Features¶

Permissions — grouped, read/write-aware editor: the flat ~120-row permission lists become collapsible resource groups (collapsed by default, search auto-expands matches) with separate Read and Write bulk toggles per group.
Outputs — modernized list and dry-run testers: the Outputs list moves to the modern sortable table. New "Test Transform" / "Test Template" modals dry-run expressions server-side against an editable sample event before saving, and a WebSocket output destination is now creatable from the UI.
Artifacts — modernized list and full-screen viewer: the Artifacts list becomes a Detections-style table with a new full-screen record viewer.
Org Overview diagram: new "Overview" page with high level diagram.
Search — full-data exports: the query console download menu splits into "Visible columns" (CSV/Excel/HTML/PDF, mirroring the table) and "All fields" (NDJSON plus new CSV/Excel that export every flattened field of every event), so nested event details are no longer dropped from spreadsheet exports.
Sign-up — Grid escape hatch: the sign-up flow asks whether users want the full LimaCharlie platform or the AI-assisted Grid before choosing an auth method.
Entra ID adapter: an optional streams parameter (risk detections / sign-ins / audit logs) is exposed in the cloud adapter form and CLI template, and USP sensor tiles gain per-type documentation links (starting with Entra ID / Azure AD / Office 365).
Extended localization: extension pages for 10+ more extensions are now fully localized across all nine languages.

Bug Fixes¶

Permission checkboxes in the API-key, add-on, and extension editors updated the form value but never visually toggled after the grouped-editor rewrite; they now toggle correctly.
Privileged permissions (apikey.ctrl, user.ctrl, billing.ctrl) could be bulk-granted in one click via select-all in the API-key/extension/add-on editors; they must now be granted individually.
A hidden HoverCard added a second window scrollbar on every page that used one, sliding the app under the sidebar when scrolled.
The modal close button floated over scrolled content in tall modals (e.g. the Data Classification rule editor); it now pins to the content and scrolls with the header.
The sidebar collapse caret was hidden and unclickable on tablet-width viewports because the fixed top navbar overran the sidebar rail.
Sortable table column headers wrapped their sort icon onto a second line in narrow columns (e.g. the Sensors "Type" column).
Long unbroken tokens (resource ids, base64 keys) overflowed the CAASM "What Changed" column.
Opening a Data Classification rule authored via IaC could crash the item detail when a hand-authored rule carried non-string values; such values are now coerced safely.
The Artifacts list sent Insight timestamps in milliseconds instead of seconds, breaking the feature once the API began rejecting out-of-range values.

2026-07-15¶

Web App 5.13.0¶

Expanded Vulnerability Reporting, a revamped audit log, a redesigned REST API section, and continued localization across the extension catalog.

New Features¶

Vulnerability Reporting: a new Vulnerable Packages tab (with a package drawer listing live affected hosts) and a "Group by application" toggle on per-host tables that collapses per-CVE rows to one line per application. Exports are stamped with org name and OID, gain a server-generated remediation-plan CSV and a vulnerable-packages CSV.
Vulnerabilities dashboard polish: the Packages table collapses like the other tabs, posture stats match the detail-panel strip, and the CVE and package detail sidebars cross-link to each other. The CISA KEV catalog now opens with the full filter parameters applied.
Audit log revamp: the Platform Logs audit tab adds sensor, date-range, and event-type filters plus client-side origin/identity search, bidirectional infinite scroll ("Load newer events"), and a row detail panel with a scrollable JSON viewer of the raw record.
REST API section redesign: API Details becomes a compact definition card with hover-to-copy on API Root, OID, and Org JWT and a Swagger reference link. The User-Generated, Service-Managed, and Ingestion key tables move behind a segmented pill switcher showing one table at a time, with a contextual Create button and one-line dates.
EPP status dashboard: the per-sensor EPP status page is redesigned as a status dashboard, with the metric tiles filled out and macOS status, product name, and not-subscribed states corrected on the Sensor Overview EPP card.
AI provider onboarding: a CLI escape hatch lets you configure AI providers not yet supported in the guided onboarding flow.
Internationalization: continued localization of the extension catalog - Artifact, binlib, reliable-tasking, EPP, dumper, integrity, feedback, exfil, atomic-red-team, govee, hayabusa, infrastructure, lookup-manager, and ~30 more extension pages are now translated across all nine locales.

Bug Fixes¶

Tooltips now render above modals instead of behind them.
Cloud adapter forms now enforce required secret and number fields before submit.
Sensor Overview hides cards and chips that don't apply to adapters and USP sensors, long tags no longer blow out the Overview grid, and EPP status resolves instead of stalling on "Waiting for sync".
A newly created hive record shows its GUID immediately, and User ID click-to-copy is restored on the account settings page.
Workload Scanning last-scan stats are corrected (wire tolerance) with a richer Last Scan detail, and the "Findings closed" column in Top remediations is fixed. The CVE sidebar count stat is labelled "Findings" rather than "Impacted hosts".

2026-07-14¶

Endpoint Agent 5.3.3¶

New Features¶

The sensor service can now be upgraded or uninstalled through cloud tasking.
Upgrade and uninstall commands reply with their result, including the OS error code on failure.
Sensor upgrades can be pinned to a specific version.
Sensor connections now report a unique per-boot identifier to the cloud.
Sensor connections now report the main network interface's MAC address.
Windows endpoint protection status now falls back to Security Center antivirus data when the Defender query fails.

Bug Fixes¶

Sensors that connect but receive no data from the cloud now retry with a fresh TLS handshake after 30 seconds.
Fixed connection failures under pinned-certificate trust: TLS hostname verification now applies only to public CA trust.
Fixed a TLS 1.2 handshake failure with servers using certain RSA-PSS signature schemes.
Fixed macOS notarization of the sensor binary.
Fixed a thread-synchronization race on Linux and macOS that could destabilize the sensor.
Fixed a small memory leak when kernel acquisition shuts down on macOS.
Fixed incorrect formatting in several sensor debug log messages.
Fixed assorted small bugs across the sensor on Linux and Windows.

Improvements¶

Updated the embedded TLS library (mbedtls) to 4.1.1.

2026-07-09¶

Web App 5.12.0¶

The console and Grid app are now fully localized in English, Español, 日本語, Français, Deutsch, Português, 한국어, Italiano, and Nederlands alongside redesigned sensor pages and per-sensor vulnerability report exports.

New Features¶

Internationalization: the console and Grid app are fully localized in English, Spanish (Español), Japanese (日本語), French (Français), German (Deutsch), Portuguese (Português), Korean (한국어), Italian (Italiano), and Dutch (Nederlands). A Language Switcher in User Settings sets the preferred language per device.
Sensor Analytics dashboard: the per-sensor Analytics page is rebuilt as a responsive card-grid dashboard with a single shared time-range selector, replacing the stack of full-width charts.
Sensor Overview redesign: identity, network, and system details are organized into grouped cards with inline status chips for connection, seal, network, and platform state. Seal and Isolate controls move into the page header, with the full pending/cancel state machine preserved.
Search progress indicator: the query console status line shows a live percent-scanned figure while a paginated query runs, with a hover tooltip listing batch, event, and data counts for the in-scope scan. Cancelled searches are now labelled "Cancelled" instead of "Complete!".
Per-sensor Vulnerability Report export: the Sensor Vulnerabilities tab adds an Export report menu with PDF, HTML, Markdown, CSV, and Excel formats. Reports include a KPI summary strip, top remediation recommendations aggregated by package and fix version, and the full findings table (CVE, application, version, severity, score, LC Risk, EPSS %, KEV, fix version, first detected date).
Builder Program promo: free-tier users periodically see a Builder Program offer (first 3 months free) in the console. The promo is limited to first-party LimaCharlie branding, shown at most once every 7 days, and suppressed permanently once a user expresses interest.
Vulnerability tables: First Detected and Version columns: both the sensor-level and org-level vulnerability tables add a sortable First Detected column (the date a vulnerability was first observed on the host, marking the start of the remediation clock) and a Version column showing the installed version of each vulnerable package. The version value also feeds the compliance CSV export, replacing a previously hardcoded blank.
Vulnerabilities dashboard and CVE detail redesign: the org Vulnerabilities page replaces the 2×4 KPI card grid with a compact posture strip, brings charts forward, and narrows the CVE list/detail gracefully on smaller viewports. The CVE detail gains a cleaner CVSS metric card and a full-width EPSS percentile meter with banded severity coloring.
Microsoft Defender adapter: an optional Endpoint field targets Enterprise (Commercial), GCC, GCC High (L4), or DoD (L5) Microsoft cloud environments. Existing adapters continue to use the enterprise endpoint by default.
SentinelOne adapter: optional site_ids and account_ids fields scope ingestion to a single tenant of an MSP/partner console, and a collect_agents toggle pulls all in-scope endpoints as individual sensors immediately rather than waiting for telemetry.
Sensor sub-page tables: Autoruns, Drivers, Event Collection, Packages, Users, Services, Integrity, and Processes migrate to the updated table component, gaining sortable column headers, truncation tooltips, and consistent viewport-fill height. Event Collection and Integrity rule chips move into per-row kebab menus, and several empty-state typos are fixed.
Analytics charts: all analytics charts across D&R rules, False-Positive rules, Outputs, Adapter Analytics, VibeRails, Billing quota, and the Grid app are updated. Outputs and Adapter Analytics charts now display side by side with a single shared time-range selector.

Bug Fixes¶

The "Create new App" button now works for orgs with zero apps; a transparent overlay was intercepting pointer events over the page header.
Opening the AI Terminal no longer crashes the app with "Maximum update depth exceeded".
Seal and isolate pending transition states now display correctly, so "Pending rejoin" and "Pending unseal" surface during the backend transition window instead of holding on "Isolated" or "Sealed".

2026-07-06¶

Endpoint Agent 5.3.2¶

Bug Fixes¶

Fixed connection failures under pinned-certificate trust: TLS hostname verification now applies only to public CA trust.
Fixed a TLS 1.2 handshake failure with servers using certain RSA-PSS signature schemes.

2026-07-02¶

Endpoint Agent 5.3.1¶

New Features¶

TLS 1.3 is now supported for the sensor's connection to the LimaCharlie cloud, with automatic fallback to TLS 1.2.
Shell commands can now be run as a specific user instead of always running with the sensor's own privileges.
USB Data Loss Prevention is now configuration-driven, so the enforcement mode and allowed-device list persist across restarts and can be managed centrally from the cloud.

Bug Fixes¶

Improved the reliability of running shell commands: fixed lost error output, incorrect exit codes, and commands being cut short on hosts whose clock jumps.
Endpoint protection status now reports specific errors when a query fails, instead of a single generic failure.
Fixed several memory-safety issues and memory leaks across the sensor, including at shutdown and on Windows.

Improvements¶

Linux kernel component failures are now easier to diagnose, with more detail reported in its status.

2026-06-29¶

Web App 5.11.0¶

Expanded AI cost tracking adds spend breakdowns, MSSP chargeback, and a savings trend to the AI Usage page, now available in the main web app alongside deeper Vulnerabilities filtering and functional app egress.

New Features¶

AI cost tracking: the AI Usage page now offers 7/30/90-day ranges, a KPI strip (spend, investigations, cost per investigation, tokens), spend breakdowns by model and detection rule with per-investigation unit cost, per-tenant re-bill markup for MSSP chargeback in the CSV export, anomaly and trend indicators, and a savings trend chart. Sub-cent costs display adaptive precision instead of rounding to $0.00.
AI Usage in the main web app: the AI Usage view, previously Grid-only, is now reachable from the main web app sidebar under the AI group, gated by the ai_agent.get permission.
Vulnerabilities subscription gating: orgs without the Vulnerability Reporting extension now see a subscribe call-to-action (for billing.ctrl users) or an admin-contact prompt, replacing the misleading empty state.
Vulnerabilities platform filtering: the Platform facet now scopes the CVEs tab, and dashboard charts and KPI tiles update to reflect active Severity and Platform filters, with a caveat label when other filters can't be represented in the rollup.
Vulnerabilities application grouping: the host vuln table splits the combined column into sortable Application and CVE columns, grouping all CVEs for a package together; the org drawer defaults to application sort and the sensor tab keeps score sort.
Vulnerabilities false-positive feedback: a per-finding "Report incorrect detection" action collects a structured reason and relays it to the product team, separate from the local mark-false-positive triage action.
Gmail adapter: setup and edit forms for single-mailbox OAuth and Workspace service-account flows, with per-feed capability toggles, subject scoping, and masked managed-secret storage for service-account credentials.
ThreatLocker adapter: Include Child Organizations scoping for parent API tokens and individual toggles for the Approval Requests, Unified Audit, and System Audit feeds.
Brand feature flags: branded deployments can disable fleet billing, case management, automation SOPs, mini apps, and the AI terminal per deployment.
Windows PowerShell installer: the Windows install wizard leads with a copy-paste PowerShell one-liner using LimaCharlie's hosted install.ps1, mirroring the Linux curl installer, with the manual EXE/MSI tab still available.
Search timing breakdown: the per-stage timing breakdown is now on by default for all users.
Sensor Connectivity: the Add Sensor panel now lists the org's webhook endpoint alongside the existing addresses, making firewall setup for cloud sensors and webhook adapters easier.
Apps launcher icons: apps now derive distinct icons from author emoji, required-permission prefixes, and other signals before falling back to the generic diamond.
Per-theme logos: runtime configs support a dedicated dark-theme logo alongside the standard logo.

Bug Fixes¶

Apps declaring allowed_origins can now make third-party fetch() calls; apps load from a real HTTP origin with their own permissive floor CSP instead of inheriting the console's strict policy. The brokered lc.api path and app isolation are unchanged.
Apps now open correctly on Grid, which previously failed with a sandbox handshake timeout.
Creating a false-positive rule from a detection now names the draft from the detection (category and detect ID) instead of "Untitled-1".
Deleting a REST API key now requires confirmation through a danger dialog, preventing accidental deletion that would break integrations.
Replaying a D&R rule with target: detection now runs against the detection stream instead of the event stream, which had matched nothing.
Projection queries that select ts without an alias now show the column, formatted as YYYY-MM-DD HHss in the table and exports.
Saved query text can now be edited in the Edit Query modal, not just renamed.
The saved-query size limit now matches the backend's 1024-byte ceiling, raised from a stricter 512-byte client cap.
CVE descriptions from the NVD feed now render as sanitized HTML instead of literal tag text on the CVE detail page and sidebar drawer.
The CVE detail page layout is now stable, with stacked tables sized to their actual row count instead of growing unboundedly or leaving large empty gaps.
The KEV and Total vulnerability tiles now use server-computed host-wide counts instead of page-limited values, fixing incorrect counts for sensors with many findings.
The create-case org picker now shows all cases-enabled orgs for accounts spanning more than 200 orgs, resolving names for up to 10,000 orgs.
Social share previews are fixed for Grid and the main web app, with OG and Twitter tags added and robots.txt updated to allow social crawlers while keeping the console out of search indexes.

2026-06-18¶

Web App 5.10.0¶

Fleet Billing arrives for MSSPs, alongside AI Terminal refinements and clearer session-state reporting.

New Features¶

Fleet Billing: a cross-tenant billing console for MSSPs.
AI Terminal: visual improvements and refined landing actions.
AI session statuses: Running, Waiting, and Ended states, with an indicator when a session is awaiting user input.
Access Management: added a Role column to the user table.
Case console: added a selectable page size.

Bug Fixes¶

Query console download now matches the visible table.
Added a Minimize button to the AI Session chat floating button.
Added the AI Session chat button to the sessions page.
Restored the "Download file from session" action in the chat menu.
Restored the app version and release notes in the profile menu, and fixed the Dark Mode font.
Adjusted the light-mode AI Terminal card fill color for better contrast.

2026-06-11¶

Web App 5.9.0¶

Notable improvements and fixes:

New Features¶

AI Terminal ShareCard — share sessions socially and invite users to tenants.
Docked corner chat — persistent AI chat launcher with pop-out, draft-new-session, minimize/maximize, and a live-session selector.
Session fork — fork an AI session with full lineage tracking.
Card-list session browser inside the chat layout with a unified action dropdown.
Onboarding demand-signal cards for unsupported AI providers.
Feedback rich card plus unsupported-request guidance for the AI FDE.
ThreatLocker platform and adapter support.
AI Workbench — agent usage moved into a dedicated Usage tab.
Search — scroll-to-top/bottom buttons, perceived + server timing on the status line, and query ID in the timing breakdown tooltip.
AI Sessions — rationalized session user state (Active + needs-attention), and lc-compliance as a selectable plugin.
Git Sync — added missing config hives to the UI.

Bug Fixes¶

Fixed dark-mode LimaCharlie logo visibility.
Included ai_agent.operate in FDE and worker-key permission lists.
Cases — batched bulk assign / tags / close-note to avoid rate-limit failures.
Restored the agent list table in the AI Agents tabs.
CVE detail now opens from the sensor vulnerability list.
Timestamp column hidden for aggregation search results.

2026-06-05¶

Endpoint Agent 5.3.0¶

New Features¶

New reg_get command fetches a single named Windows registry value by key and value name, for values too large or keys too crowded to retrieve with reg_list.
macOS installation is now a single guided window with a three-step checklist — system extension, network content filter, Full Disk Access — replacing the previous sequence of seven separate prompts.

Bug Fixes¶

Fixed macOS file-creation events reporting the parent directory instead of the new file's path, which stopped exfil watch rules from matching specific files.
Fixed a leak of registry key handles when listing Windows registry keys.

Improvements¶

File hashing is faster and allocates far less memory, lowering sensor overhead on hosts with heavy file activity.

2026-05-30¶

Endpoint Agent 5.2.6¶

Bug Fixes¶

Fixed very large artifact uploads never completing: each part is now retried up to ten times with backoff, so a brief network interruption no longer restarts the transfer from the beginning.
Fixed uploads of locked files larger than 4 GB on Windows looping indefinitely over the first 4 GB.
The sensor now shuts down promptly while an upload is retrying, instead of waiting out its backoff.

2026-05-09¶

Vulnerability Management Uplift¶

Major uplift to the Vulnerability Reporting extension and its surfaces.

Canonical asset-tag namespace: introduces the lc:asset:* tag convention (criticality, exposure, environment, owner, compliance) for cross-cutting asset metadata. The Vulnerability Reporting extension is the first consumer; the namespace is intended to be reused across LimaCharlie surfaces. See Asset Tag Namespace.
Vulnerability Reporting extension: new public-facing documentation covering setup, scan modes (scheduled / manual / all), criticality-tag overrides, KEV + EPSS enrichment, LC Risk scoring, and the full action surface. See Vulnerability Reporting.
Finding resolutions: documented the resolution model — every finding is implicitly open until an operator records mitigated, accepted, or false_positive. Accepted-exception expiries lapse back into the open count, and vuln_finding.* events (created, closed, kev_match, state_changed) can be routed via Outputs to Jira, Slack, Cases, etc.

2026-04-02¶

Endpoint Agent 5.1.0¶

New Features¶

The os_drivers command now works on Linux, listing the host's loaded kernel modules.
The sensor health check now downloads and verifies the latest health-check binary from the cloud, so diagnostics stay current without upgrading the sensor.

Bug Fixes¶

Fixed macOS and Windows sensors staying on a dead connection for hours after a network disruption; keepalive behavior is now consistent across all platforms and connections recover promptly.
Fixed occasional missing line breaks in the sensor's local log file.

Improvements¶

The sensor health check now reports details of the installed sensor binary on disk.
Sensor debug data now includes the sensor service version.
Sensor service details are now reported to the cloud on every sync, instead of only when debug data is requested.
Linux and macOS sensor binaries are smaller.

2026-02-08¶

Endpoint Agent 4.33.26¶

Feature: Windows ARM64 support
Fix cloud initialization issue on start and reboots
Fix an eBPF umount issue on Linux whenever LVM is used

Web App 5.5.5¶

Enhancements to the AI Sessions experience with granular permission controls and improved session management, adds Windows ARM64 sensor support for customers on newer Windows hardware, introduces search customization features including drag-and-drop column reordering and improved time range displays, and streamlines the sensor installation workflow with better Docker instructions and installation key visibility. The release also includes important security improvements for OAuth authentication and API key isolation between environments.

AI Sessions: Granular permissions, pagination, SOPs enabled
Windows ARM64: New sensor support for ARM-based Windows devices
Search: Drag-and-drop columns, column popover, improved time display
Investigation Viewer: Major overhaul with new sidebar and unified data handling
Sensor Installation: Better Docker instructions, fixed modal behavior
Security: Environment-specific API keys, GitHub OAuth trust, CSP fixes

2025-10-24¶

Endpoint Agent¶

Stable: 4.33.10.5¶

Fix an issue with the epp_list_exclusions command not showing all exclusions

Latest: 4.33.19¶

Fix an issue with the epp_list_exclusions command not showing all exclusions
Change the compression algorithm for Debian installer packages from xz to gzip for better compatibility with older distributions

Web App 5.2.1¶

A good batch of bug fixes.

2025-10-17¶

Endpoint Agent¶

New stable release 4.33.10.4 - fixed memory leak in file type tracker, upgrade recommended.

New release 4.33.18:

Number of fixes including:

A memory leak in the file type tracker
Potential crashes in cloud task processing and log tracker
Version reporting fixes on Windows 11, Linux, and macOS
Certificate updates to avoid solely relying on the platform keychain
Update the size of the file tracking buffer to avoid missing events on Linux

Web App 5.2.0¶

Features:

Show errors "in place" for Outputs, Detections, and Extensions to accelerate troubleshooting
Support text data types in extensions (e.g. for large inputs, per community request)

Bug fixes: Query console, D&R rule creation flow, Console, File system.

2025-10-13¶

Web App 5.1.0¶

Substantial update of Access Management for clarity and convenience of user management. Continued refinement of the UX for user and group management.

Bug fixes: Artifact list, file system, and more.

2025-09-30¶

Web App 5.0.0¶

The Query Console is now in the new UI theme. Feature-rich search and analytic capabilities integrated with the rest of the LimaCharlie platform marks a major revision.

Note

The Query Console remains in beta while we continue to improve performance and refine usability.

2025-09-12¶

Endpoint Agent v4.33.15¶

macOS system extension: fixes for potential install and upgrade issues, improved logging for troubleshooting
Fixes a macOS install issue when using the package installer (.pkg)

Web App v4.5.0¶

Many bug fixes and some ongoing UX improvements.

2025-08-28¶

Endpoint Agent 4.33.14¶

Reduce CPU usage of the OS tracker. The component is used for: new system service notifications, new driver notifications, and new autoruns/bootstrap notifications.

Web App 4.4.9¶

Improved UX for Access Management and Adapters pages
Add Wiz cloud sensor adapter integration
Many fixes and smaller improvements

Other Notable Updates¶

Org Templates: Re-introduction of org templates during org creation. Currently offering EDR Quick Start and Basic Browser Monitoring to get the new org configured on creation. You can always add configurations as code using the IaC Generator.
Parsing Unstructured Logs: With parsing_grok you can use OpenSearch Grok processor syntax, and tap into powerful ready-to-use Grok patterns and the vast knowledge of grokking data with Elastic.

2025-08-08¶

Endpoint Agent 4.33.13 and 4.33.10.3¶

Fix a Windows 2016 compatibility issue in the kernel driver for both lc:stable (4.33.10.3) and lc:latest (4.33.13) versions.

2025-07-18¶

Endpoint Agent 4.33.11¶

Bug Fixes:

Resolved event loss on high-traffic Windows systems
Fixed kernel upgrade failures that could occur during system updates
Addressed code signing compatibility issues on macOS

Breaking Changes:

Console logging is now opt-in via -v or --verbose flags. The previous -v flag for displaying installer version has been changed to -V. This improves default output cleanliness while maintaining debugging capabilities.

New stable version is now 4.33.10.

Web App 4.4.4¶

A patch release with minor bug fixes.

2025-07-10¶

Web App 4.4.3¶

Fixed regression with sensor timeline view.

LC Adapters v1.30.11¶

Integration with Cylance, Proofpoint Tap, and Wiz. Big and special thanks to community contributors shortstack and RagingRedRiot.

Note

These adapters are supported in the downloadable Adapter, but not yet rolled out to the web app as "cloud adapter."

2025-06-27¶

Endpoint Agent 4.33.9¶

Important fixes for Windows 7 and Windows 8 support.

Web App 4.1.2¶

Bug fixes for customers and community.

New Community Rule Sets¶

Two new rule sets from community partners added to the LimaCharlie Add-Ons collection:

SoteriaSec Commercial Ruleset: Google Workspace Rules
BLOKWORX Detection & Response: Rules covering detection of a collection of remote access services usage

2025-06-17¶

Web App 4.4.0¶

AI-powered community rules (Beta): Easy way to turn thousands of community rules into LimaCharlie detection & response rules.
New and improved Extensions page
Bug fixes, including a few around auto-generated Extensions UI for extension builders

2025-05-30¶

Endpoint Agent 4.33.8¶

Fix a potential deadlock on upgrade in the HBS component
Fix a reverse logic issue processing the LC_DISABLE_REVERSE_DNS_HOSTNAME environment variable. Accepted values: 1, true, 0, false (case insensitive)

Web App 4.3.3¶

AI-assisted detection read-out, navigation improvements, showing org selector consistently, and a number of bug fixes.

2025-05-22¶

Endpoint Agent 4.33.7¶

Linux:

Fix some Linux GLIBC compatibility issues. The minimum GLIBC supported version is now 2.17 (released 2012) for all 3 supported architectures (x86, x86_64, and ARM64)
Fix the Linux Alpine / musl libc binaries

macOS:

Standalone installer is now a universal binary (FAT) to prevent users from installing on the wrong architecture
Fix an issue where the host isolation command wouldn't terminate existing connections

Windows:

Added an environment variable (LC_LOCAL_CACHE_ONLY_REVOCATION_CHECK) to prevent the Windows WinTrust code signing library from updating its revocation cache from the internet. The default and recommended setting is to let WinTrust update its cache, but the sensor may connect to content delivery networks (CDNs) on port 80 to do so.

General:

The sensor troubleshooting tool (rphcp -H) was missing in the .deb, .msi, and .pkg installers

Web App 4.3.2¶

Fixes for a few edge-case crashes and recently reported bugs.

2025-05-20¶

LimaCharlie Endpoint Protection¶

Releasing LimaCharlie Endpoint Protection, which integrates with third-party EDR solutions to provide a better view of security operations and extend agent capabilities. This functionality comprises the EPP Extension, Web App, and a previously released Endpoint Agent v4.33.6.

Web App 4.3.1¶

UI support for Endpoint Protection solution, bug fixes.

Extensions¶

Endpoint Protection extension: a component of the EPP solution that codifies key configurations for Microsoft Defender.

2025-05-08¶

LC Adapter v1.30.1¶

Adding Sublime adapter. Audit logs from Sublime can be ingested cloud-to-cloud via the API.

Web App 4.2.8¶

A number of UI bug fixes.

2025-04-18¶

Endpoint Agent 4.33.6¶

Allow the sensor to drop the VDI file (delayed start) during the installation procedure via -t
Added a sensor troubleshooting utility: a standalone command and a command line option for the sensor (-H) to help diagnose common misconfigurations and connectivity problems

Web App 4.2.3¶

Fixing the artifact download broken in some cases, and other small bug fixes.

2025-04-11¶

Web App 4.2.1¶

AI co-writer for D&R: Use "ask AI" when creating a rule and it helps you write a detection and response based on your prompt. Currently uses Google's Gemini 2 Flash model tuned for LimaCharlie D&R rules. Standard AI disclaimer applies: "trust but verify."
Event Tree: Updated for usability and performance on giant trees. Enjoy collapsing and expanding groups of events, and traverse the tree with no strain on your browser.
Other performance optimizations and bug fixes

Endpoint Agent 4.33.5¶

Performance improvements for macOS
Infrastructure work to support Endpoint Protection Platforms (EPP), and added support for Microsoft Windows Defender

Note of Change

LC Detection Events are now immutable. One can no longer remove past events or modify them in any way, as detection events are a factual historic record and it's prudent to keep them as such.

2025-03-28¶

Web App 4.1.4¶

UI betterment: quick filters for common platforms on the Sensor list, reliable navigation from/to Detections, other small improvements and bug fixes.

Adapters¶

SentinelOne adapter: connects to SentinelOne MGMT API and sends alerts, threats, and other events of interest to LimaCharlie.

Endpoint Agent v4.33.4¶

Fix missing pipe event for Windows
Fix the kernel acquisition module for Linux ARM64 builds

Extensions and Adapters¶

Git-Sync: Take the best from LimaCharlie Infrastructure as Code by connecting with Git and syncing the desired sections of your configurations in an easy-to-use UI. See documentation.
ext-renigma v1.0.0: Initial release of integration with REnigma, an advanced malware analysis platform leveraging its unique Record and Replay technology.
Mimecast adapter: Connect to the Mimecast API to stream audit events as they happen.

Web App 4.1.1¶

Usability improvements on Detection page, ability to re-run command in sensor console, fix "copy array index," and numerous bug fixes.

2025-03-14¶

Web App v4.0.2¶

A long-awaited modernized UI is available (in preview). More work is on the way to further improve user experience.
In-product dashboards available (in preview): a bird's eye view on key detections and the flow of data.

This is not just a paint job: we made substantial internal changes and will continue to improve quality.

Note

On large orgs, the dashboards can take up to 15 seconds to load the very first time, and normalize after the first load. Optimizations are on the way.

The Query Console is not available in the Modern UI yet. We will bring it there in a much better shape. In the meantime, switch back to the Old Theme to access it.

Add-Ons and Adapters¶

PandaDoc adapter: Connect and fetch PandaDoc API logs.
CrowdStrike Falcon Cloud adapter: Connect to CrowdStrike Falcon Cloud to stream events as they happen in the CrowdStrike Falcon Console.
Cloud-CLI v1.4.8 Extension: Improved observability in CLI extensions such as ext-cloud-cli. Additionally, improved error handling and reporting around long-running CLI commands which may have gotten stuck or timed out.

2025-03-06¶

EDR Agent v4.33.2¶

Fixed a path expansion issue that would cause the cleanup command on Windows to leave configuration files after the uninstallation procedure.

Adapter v1.27.2¶

Added support for ZenDesk.

2025-02-28¶

LimaCharlie Labs¶

Introducing LimaCharlie Labs, where we share brave experiments and early prototypes of features and extensions that may or may not become production, based on your input and feedback. Check the LABS badge on the Web App.

Playbook Extension is now available in Labs.

Web App v3.10.1¶

Introduce Event Latency (routing/latency), and add latency metrics to Sensor Analytics, to help identify and troubleshoot event latency issues
Add "Search by Description" to the org list
Bug fixes
Report a Bug: Integrated tool to report bugs easily

2025-02-21¶

Web App v3.9.3¶

Bug fixes: handling edge-cases of org creation and adding users flows, fixing MS 365 sensor false status in certain rare conditions, other small fixes and internal instrumentation improvements.

CLI 4.9.12¶

Add users, simplified. Wrapping the new API, a new command limacharlie users invite makes it easy to add a user, or a batch of users, to the org without requiring them to create a LimaCharlie account.

EDR Endpoint Agent v4.33.1¶

Fix various directory and file permissions on macOS
Added a status file to help troubleshooting. The status file contains the sensor ID, organization ID, version, and the agent's service uptime.

Status file locations:

Platform Path

Linux /opt/limacharlie/hcp_hbs_status.json

macOS /Library/Application Support/limacharlie/hcp_hbs_status.json

Windows C:\ProgramData\limacharlie\hcp_hbs_status.json
Fix a missing package name for Microsoft Edge Update on Windows
Fix a pattern matching issue that was affecting file integrity notifications
Added the LC_DISABLE_REVERSE_DNS_HOSTNAME environment variable for customers wanting to use the local hostname instead of resolving it

2025-01-24¶

Web App v3.8.12¶

New Features:

New Australia Datacenter: Added a new datacenter in Australia to enhance performance and availability for users in the region.
Secrets Manager Integration: The SMTP password field now allows integration with the secrets manager, providing a more secure way to handle authentication credentials.
New Extension: ext-nims allows you to send detections from LimaCharlie to NIMS via the Notion API.

Bug Fixes and Enhancements:

Autofill OTP: The one-time password (OTP) field now properly auto-fills from password managers.
User Permissions Warning: A warning message has been implemented to notify users when revoking permissions for a user.

2025-01-09¶

Web App v3.8.10¶

Bug Fixes and Improvements:

Fixed a bug where creating a new secret in a secret manager and changing cloud adapter configuration at the same time would not update the cloud configuration with the new secret.

ext-usage-alerts v1.0.0¶

Newly released extension which allows you to create, maintain, and automatically refresh usage alert conditions for an organization.

2024-12-12¶

Web App v3.8.8¶

New Features:

Introduced user-level saved queries for improved data management.

Bug Fixes and Improvements:

Fixed the alignment of the "skip for now" text on the initial sensor onboarding screen during organization creation.
Resolved an error related to empty extension configurations.
Fixed a minor horizontal scroll issue on the sensors page.
Fixed an issue where the organization creation waiting room would display "missing permission errors" when opening the app.
Minor enhancement on the input field for adding a user to your organization, where it will now show an error if the "add user" button is clicked without an email filled in.
Updated various mentions of "Yara" to be all caps to reflect it being an acronym.

2024-10-28¶

New MITRE Report API¶

Added a new REST API and CLI for producing a MITRE report for a given organization based on the D&R rules in place (using their tags like attack..t1000.xxx).

CLI: limacharlie mitre-report

The resulting JSON report can be used with the ATT&CK Navigator. This capability makes it easier to track security coverage against the MITRE ATT&CK framework.

2024-10-19¶

EDR Sensor v4.31.1¶

Network connection stability enhancements on all platforms.

The enhancements are in both the cloud-triggered upgrade version of the sensor AND in the on-disk installation, but there is no requirement to deploy both simultaneously.

2024-10-17¶

New Sort and Bulk Actions for Tables¶

Added the ability to sort columns in the LimaCharlie web app. In addition, tables now support bulk actions (Enable/Disable and Delete). This applies to the following sections: Adapters, YARA Rules, Secrets, Lookups, False Positive Rules, and Detection and Response Rules.

Platform	Path
Linux	`/opt/limacharlie/hcp_hbs_status.json`
macOS	`/Library/Application Support/limacharlie/hcp_hbs_status.json`
Windows	`C:\ProgramData\limacharlie\hcp_hbs_status.json`