Added

• Acknowledgements window. The app now includes the licence texts of the open-source code it uses. You can open them from Help › Acknowledgements or from a link in the Settings footer.
• Licence status check. The app now checks once a day if your licence is still active. If your licence is stopped, Pro turns off.

Fixed

• Flush size value. The Dashboard now shows the full Flush size value in the database health card.
• Service colour. The third service colour in the trace waterfall is now yellow instead of pink, so it no longer looks like an error.
• Licence date after refund. When a renewal or Lifetime upgrade is refunded, the app now correctly reverts the updates date to the earlier date.
• New licence key field. The Plan card in Settings now shows a "New licence key" field for Pro users who are not on a Lifetime licence. You can paste a renewal or Lifetime upgrade key there.

Fixed

• Pro settings tags: On Free, the Pro settings now show a "Pro only" tag and the controls are locked. The upgrade note no longer shows prices or a "Load licence file" button, and the Plan tab shows a "Get Pro" row with a Buy button.

Changed

• Collector size: The collector no longer includes Prometheus service discovery plugins. The app download is 168 MB smaller, and the collector uses about 21 MB of memory instead of 35 MB.

Added

• MCP server status and last AI tool: The AI coding tools card now shows the MCP server status (Running, Starting, or Not running) and the address. After an AI tool uses Loggerhead, the card shows the tool name and the last tool call, and updates every second. The card warns when the port is outside the expected range, shows the reason when the endpoint does not start, and no longer shows the message "The server reads only".

Added

• Tabs in Settings. The Settings page now has four tabs: General, Data, AI, and Plan and updates. The General section is now two cards, and the AI coding tools have their own card.

Fixed

• AI settings and theme. AI settings and the theme now apply to the whole Mac. The General tab now says “All workspaces”, and switching between workspaces or environments no longer changes these settings.
• Trace span events and links. Events and links are no longer written for a group of spans that was not stored. This prevents duplicate events and links when the group is sent again.

Fixed

• Corrected page names The app now shows "Loggerhead receiver" instead of "Local OTLP receiver", "Linting page" instead of "Improvements page", and "Rules tab" instead of "Lint rules page". The AI analysis button and findings are on the AI analysis page.
• Corrected finding titles The finding for missing resource attributes now names what is absent: the release, the deployment, the host, or the deployment and the host. Group cards with no plural title now show the count, the noun, and the rule name.
• Fixed workspace port configuration The get_receiver_info tool, the health call, and the Collectors page now show the configured ports of the workspace the app shows, not the first workspace.
• Added Review button to single findings A single finding card now has a Review button when its action button does not open the review sheet. The Review sheet lists the series with their label values.
• Automatic refresh of Linting page The Linting page now refreshes its findings every 10 seconds while the page is open. A finding that resolves moves with no click.

Added

• Histogram percentiles and exemplars: The app now shows percentile series, such as p95, for histogram metrics. Metric points now include exemplars that link a metric to a specific trace.
• All resource attributes: The app now stores and displays every resource attribute, including service.instance.id, container.id, and cloud.*, instead of only three.
• Metric descriptions: The Metrics page and the query_metrics tool now show the description of each metric.
• Span detail panel: Clicking a span in the Traces page opens a detail panel. The panel shows the status message, the exception, the attributes, and the events of the span.
• Span events and links: The app now stores the events and the links of a span. The MCP tools return them.

Fixed

• Jump to old trace: Clicking a log line now opens its trace even when the trace is not among the newest 50. The waterfall shows the trace regardless of its age.
• UpDownCounter at zero: The "Broken metric" rule no longer flags an UpDownCounter, such as http.server.active_requests, when its value is zero.
• Unknown environment tag finding: The finding for records with an unknown environment tag now resolves one hour after the last record with that tag arrives.
• MCP lint engine for current workspace: The query_improvements and mark_finding_fixed tools now use the running lint engine for the current workspace. Findings no longer incorrectly resolve and reopen every 30 seconds.
• Delta metric series: The "Counter goes down" rule no longer flags a metric series that reports changes since the last point, such as a sum from the .NET SDK or the Prometheus receiver.
• OTLP data handling: The receiver now handles more OpenTelemetry fields correctly. Logs with no timestamp use the receive time. Structured log bodies store as JSON. Span status messages and exceptions store as attributes. Attribute values preserve zeros, false, and structured values. Metrics with the NO_RECORDED_VALUE flag are dropped instead of stored as zero.
• Histogram metrics: The app now stores histogram metrics, such as http.server.request.duration, as separate series for the count, sum, minimum, and maximum. The Metrics page shows them in their own group.
• Summary quantile type: The app now stores summary quantile samples, like go_gc_duration_seconds{quantile="0.99"}, as gauges. The "Counter goes down" rule no longer falsely flags them.
• OTLP/HTTP response encoding: The receiver now responds to a successful export in the same encoding as the request. The OpenTelemetry JS SDK no longer logs a deserialization warning.

Changed

• Mark as fixed: The "Mark as fixed" action now checks the telemetry that arrives after the fix. If the problem continues, the finding reopens. If the problem stops, the finding resolves.

Changed

• Redesigned Setup page. The Setup page now shows two ways to connect an app side by side, with "or" between them: "Do it yourself" and "Ask your AI agent". The AI key appears in its own card below the two ways. The footer shows progress and says that both ways finish when the first telemetry arrives. The sheet title is now "Get set up".

Added

• Ask your AI agent step: The Setup checklist now includes a new step, "Ask your AI agent", after "Send your first telemetry". The step has a "Copy prompt" button that copies a prompt for AI coding tools. The prompt contains the live OTLP ports of the current workspace. The checklist now uses the HTTP port that the receiver reports.

Fixed

• Workspace switching: Clicking the tile or name of a workspace in the switcher popover now switches to that workspace. The button opens the local environment of another workspace, or its first environment when it has no local environment.
• Workspace checklist steps: The Setup checklist now has four steps per workspace. The "Open Loggerhead at login" step is no longer shown in the workspace checklist; it remains in Settings › General.

Added

AI analysis API types The Settings now offer "Anthropic-compatible API" and "OpenAI-compatible API" as provider types. You can set a base URL for each type, and the OpenAI type requires a model name.

Fixed

AI analysis results persistence The AI analysis results now stay on the page after you navigate away and return. The spinner also continues to show if a run is still in progress.

Added

A log that should be a metric – The AI analysis now includes check L9. It finds log lines that only count an event or report a value. It suggests a metric name, type, unit, and attributes. It tells you to remove the log line or keep it at DEBUG. If a noisy_logs or chatty_service finding already names the same line or service, the AI analysis adds to that finding.

Chatty traces – A new lint rule groups traces by root service and root operation. It fires when an operation starts more than 60 traces per minute and 90 percent or more have no ERROR span and a root span faster than 1 second. The finding gives the traces per minute, spans per minute, and spans per trace. It recommends a tail sampler at the collector. The AI analysis now includes check T5, which looks for healthy, fast traces of low‑value work and suggests a drop rule or a sample rate.

Added

• Lint rule management – The Lint rules page now shows exceptions, ignored findings, and a change history with undo. AI coding tools can now modify lint rules.
• Resolved lint findings – The Improvements page now shows a “Recently resolved” list of findings that stopped firing in the last 24 hours. Each item shows the reason: fixed or aged.
• Review finding records – Clicking “Review” on a lint or AI finding now opens a sheet that shows the records that break the rule. Log rules show log records, trace rules show spans, and metric rules show one row per series. AI findings show evidence records.

Fixed

• False positive fixes – Eight lint rules no longer produce false positives for the CTE project: sensitive_data, metric_naming, counter_decrease, missing_resource_attributes, broken_metric, constant_metric, log_schema_drift, and oversized_trace.
• Version History button – The “Version History” button in the update window now opens the full changelog on the website.

Added

AI analysis with raw telemetry and check list – The AI analysis now reads the raw logs and spans of a 1‑hour or 24‑hour window. It sends a list of 20 checks to the model. The findings appear in their own “AI analysis” section above the lint‑rule findings. Each AI finding shows its pattern and the window it analyzed.

AI check settings per environment – You can now set AI checks for each environment. The Checks tab on the AI analysis page lets you turn built‑in checks on or off, and add your own custom checks. Each custom check has a name, prompt, and signal. The Findings tab names any custom checks that were left out.

Mask values before sending to a cloud model – Loggerhead now replaces values in log bodies, span names, and lint findings with placeholders before sending them to a cloud model. Built‑in masks cover email, JWT, UUID, IP, hex strings, and numbers. You can add your own masks in Settings under “Masks for cloud models”. Apple Intelligence runs on your Mac and gets the text unmasked.

Linting and AI analysis pages with tabs – The Linting page now has two tabs: Findings and Rules. The AI analysis page has two tabs: Findings and Checks. The Linting badge counts only lint findings. The AI analysis row has its own badge.

Equal card heights and copy prompt on findings – The findings page now lines up its cards with equal heights. Buttons stay at the bottom of each card. A group card shows 5 members and a “Show N more” row. The state tabs show the count in its own pill. Each finding card has a “Copy prompt” button that copies a prompt for a coding agent to fix the finding. The card menu also has “Copy the MCP prompt”.

Copy agent prompt in toolbar – The Linting page and the AI analysis page now have a “Copy agent prompt” button in the toolbar. It copies a prompt that tells any agent with the Loggerhead MCP server to read the lint findings and AI findings, fix the problems, and check each fix. An AI finding card also offers the MCP prompt.

MCP tool to read AI findings – The new MCP tool query_ai_findings lets an agent read the findings of the last AI analysis run. The result includes the status, time, window, model, and findings. You can filter by impact, source, workspace, and environment.

Fixed

Toolbar controls have consistent height – The AI analysis window picker now uses the same themed segmented control as the “Run AI analysis” button, so they have the same fill, font, and border. The “Run AI analysis” button no longer grows while the analysis runs; it hides its label and shows a small spinner. All toolbar buttons, segmented controls, and search fields now have a consistent height.

Apple Intelligence is the default AI provider – The app now accepts “apple” as the AI provider and uses it by default. On a Mac with Apple Intelligence, AI analysis runs with no key. On a Mac without it, the page tells you why and you can add a cloud provider key.

Lint rule filter labels stay on one line – The segment labels in the Lint rules filter bar no longer wrap to two lines. Each control gets only the width its text needs. When the window is too narrow, the state group moves to a second row.

Missing trace context rule reads one service at a time – The missing_trace_context rule now counts errors per service instead of across the whole workspace. The finding names the service that needs fixing.

Fixed

• Notarized MCP bridge: The MCP bridge is now notarized, so Gatekeeper accepts it when you download and unpack it. The bridge previously failed Apple's notarization check and Claude Desktop could not start it.
• Universal binaries for helpers: The collector and the MCP bridge now run on both Apple silicon and Intel Macs. They were previously built only for the Mac's own chip, despite the app itself supporting both.

Changed

• Eight MCP tools: The MCP server now lists all eight tools it provides. The manifest previously listed seven, so one tool was missing from the documentation.
• Lint rules expanded: The telemetry linting adds four new rules: missing_span_kind, oversized_trace, missing_resource_attributes, and deprecated_attributes. The catalogue now holds 41 rules in total, and every rule has a detector. The receiver also carries service.version, deployment.environment, and host.name onto every log record and span, and the missing-resource rule reads them from there.

Added

• Lint rules page: The new Lint rules page under QUALITY shows every rule for the current environment. Each row has a toggle, the rule's name, kind, a one-line description, threshold chips, and a chevron that opens the row. The page groups rules by signal and shows counts per group. You can reset all rules to their defaults.
• Edit rule thresholds: You can now edit a rule's thresholds from the Lint rules page. Each field validates before saving, and a refused value shows under its field with the accepted range.
• New lint rules: The first seven new lint rules are now active. The catalogue holds 41 rules in four groups: Logs 16, Metrics 12, Traces 10, and the new group. Each new rule carries a NEW chip until it ships.
• Rule descriptions: The Lint rules page now shows a description for every rule. Each description explains what the rule checks and what the thresholds mean.
• Finding links to rules: A finding now shows the rule's name as a link chip that opens the Lint rules page at that rule. You can get from a finding to the switch that turns it off.
• Finding states: A finding can now be marked as fixed or ignored. Marking a finding as fixed hides it until fresh data no longer trips the rule. The Lint rules page shows the ignored count per rule.
• MCP bundle: The bridge now ships as an MCP Bundle. Every tool carries a title and a read-only annotation.
• MCP port ladder: The app now binds its MCP endpoint on a fixed ladder of ports: 9696, 19696, and so on. The bridge probes the same three ports. No file permission stands between them.
• Admin page: The checkout now has an admin page with a sign-in. You can gift licences from it.
• Branded emails: Customer emails now carry a branded HTML part with a copy-the-key button. The button opens a page that shows the licence key.

Fixed

• MCP connection after first session: The app kept one SDK Server for the life of the process. The first Claude Code session after the app opened connected, and the next one did not. The app now keeps no state between clients, so each session connects the same way.
• MCP port file permission: The bridge found the app by reading a port file the app wrote. A current macOS refuses that file to every other process. The bridge now finds the app on a fixed port ladder, so it works on every Mac.
• Release notes heading check: The release notes guard now looks for an exact "## Added", "## Fixed", or "## Changed" heading. A stray title no longer satisfies it.
• Website noindex header: The noindex header now sits on the page itself, not on its redirect.

Changed

• Sidebar groups: The MANAGE sidebar group split into QUALITY, COLLECTORS, and MANAGE. The sidebar now has four groups: TELEMETRY, QUALITY, COLLECTORS, and MANAGE. A group whose every row the plan hides renders nothing.
• Duration thresholds: A duration threshold is now a plain seconds field. What you read is what you type. The hint under the field carries the human reading.
• Pro gate notes: Each Pro gate now carries two sentences: what you tried and what Pro gives. Every note ends with a pricing link that names the gate. No two gates share a sentence.
• Comparison pages: The website now has a Compare group with one page each for Grafana, Loki, and Jaeger. Each page opens with what the other tool does better, then what Loggerhead does better.
• Sender guides: The website now has seven sender guides, one per sender, from Next.js to Kubernetes. Each page has the same shape: what to install, where to point the endpoint, and the two closing steps.

Added

• Report a Problem – The Report a Problem sheet now sends your description, optional reply email, and a diagnostic report to support. The diagnostic report includes app version, macOS version, database details, and configuration, but no telemetry content or AI keys.
• Workspaces – The sidebar now has a Workspaces section where you can create and switch between workspaces and environments. Each workspace has its own receiver ports, and each environment has its own database.
• Workspace and environment sheets – New workspace and environment sheets let you set names, icons, colors, and ports. Production environments show a caution message.
• Database health card – The Dashboard now shows a Database health card with one row per environment, including size, writer load, and headroom.
• AI analysis on device – AI analysis now runs on your Mac using Apple Intelligence when supported. No data leaves your device unless you add a cloud provider key.
• Unknown environment rule – A new lint rule flags spans whose environment tag matches nothing. You can create the missing environment directly from the finding.
• Missing span name rule – A new lint rule flags spans with no operation name. The default threshold is one span.
• Login at startup – Loggerhead can open automatically at login so the receiver keeps running after a restart. Turn this on in Settings › General.
• Accessibility labels – Every control now has an accessibility label so VoiceOver can read them. Text contrast was improved to meet accessibility standards.

Fixed

• Environment switching – Switching environments now clears the previous environment's data from every page. The app no longer shows stale rows or badges after a switch.
• Alert cards – Multiple fired alerts from the same rule now appear as one card with an "Acknowledge all" button. Each alert still shows individually when expanded.
• Finding cards – Findings with the same rule and shape now group into one card with a member list, instead of separate cards.
• Font loading – Web fonts now preload, so pages no longer shift when the font loads.

Changed

• Settings layout – Settings are now split into three scopes: This Environment, This Workspace, and All Workspaces. Storage settings moved to This Environment.
• Workspace limits – Free plans get one workspace with one environment. Pro plans get unlimited workspaces and environments.