Jira & GitHub

• Connecting a Jira or GitHub account crashed LogStitch instead of opening the browser — the OAuth client IDs stopped reaching the build in v1.3.3 and never came back. Fixed, and a missing client ID now surfaces as a readable message rather than taking the app down.
• The Create Issue form is now driven by the project's own create metadata: issue types come from the selected project instead of a global list, priority options come from the project's scheme instead of five hardcoded values, and the priority and labels pickers are hidden entirely for projects that don't have those fields. A required field the form can't fill is called out before you write the issue.
• Rejections now say what happened — Jira authentication failed: Unexpected response (400) has been replaced by the API's own explanation, and a rejected request no longer burns a token refresh retrying something guaranteed to fail.

Memory Analysis

• The utilization chart gains a p99 memory line alongside p50 and p95, so the worst case is a shape you can see rather than a number in the table below.
• The chart now grows to fill the tab instead of sitting at a fixed height above dead space, config-change and right-size labels stay inside the plot instead of being clipped at the edges, and the shading was reworked so every legend swatch maps to exactly one thing on the plot.
• The hover tooltip no longer loses its date header to the top edge, and the staleness chip no longer reports No local data for a function that is fully synced.

Window & layout

• The main window stops resizing itself when you open or close a profile — one size, restored at launch, changed only when you drag it.
• The compact/expanded choice for the function and invocation lists is remembered across launches and stays in step across open profile tabs, instead of resetting to compact every time.

Reliability

• The configuration-change badge kept its flag but lost its before/after values on the next routine sync, so the popover could only say the values were unavailable. Memory, timeout and runtime changes now hold onto what they changed from, and a migration clears the badges that can no longer show a diff.
• The storage capacity banner now clears the moment you resolve it — raising the size limit in Settings, purging, or compacting — instead of staying on screen until the next relaunch.

Read on logstitch.app →

Hidden functions

• Hide any Lambda from the Function Navigator's context menu, or Hide these N straight from a filtered search, so a 60-function account can read like a 6-function one. Hiding is a view preference rather than a state change — hidden functions keep syncing, keep ingesting, and keep counting toward every analytics and cost rollup.
• Reveal them inline from the header's N of M functions pill, or bulk-unhide from the Manage Hidden sheet, which flags any hidden function that is still syncing so it can't burn API calls unnoticed.
• Every hide is undoable from a toast that also restores pins to their original order, and global search keeps returning hidden functions with a badge — so a search never comes up empty for a function you know exists.

MCP setup for any client

• The settings tab is now MCP Server, with a client picker covering Claude Code, Claude Desktop, Cursor, VS Code, Codex CLI, Zed, LM Studio, Windsurf and more — each with its own setup steps, a copyable config snippet, and Reveal in Finder wherever the config path actually resolves.
• Test Connection puts a real tools/list through to the server and reports refused, timed out, and unexpected responses separately, so a failed setup tells you what failed.
• Changing Server Port now updates the endpoint row and every snippet as you type, instead of quietly keeping the old port until you pressed Return.

Profiles

• Scan for New Profiles… now sits in the Profiles menu and the + toolbar menu, replacing an unlabelled toolbar button that read as a generic refresh — and results list only what is genuinely new or changed, with a real empty state when there is nothing to import.

Performance

• Searching a JSON field for a specific value — a request id, an order id — dropped from roughly 47 seconds to under a tenth of a second on a large database.
• LogStitch no longer animates while it is minimized or sitting behind another window, which had been quietly burning CPU in the background.

Read on logstitch.app →

The stitched request view

• Follow a request across every function it touched on a flow map — each hop shows duration with its cold-start share, retries, cost, and peak memory, under a plain-English verdict naming where it broke. Functions that were expected to run but never did appear as ghost hops, and inferred orderings are drawn dashed so they're never mistaken for proven causality.
• The view now presents full-bleed over the invocation list and detail pane instead of in a fixed 800×600 dialog, so the flow map, stream, and line reader all get real width.
• Click any line to open it in the inspector, pivot the whole chain on a JSON field from its context menu, or copy a Markdown summary or the visible log lines to the clipboard — the copied report names every active filter, so its scope is never overstated.

Keyboard navigation

• Move through invocations with the up and down arrow keys — each one loads as you land on it, no Return needed. Click a row or tab to the list to hand it focus; the arrows keep driving the function navigator otherwise.

Reliability

• Chain filters on indexed fields no longer empty the stream, and a report generated with a function filter applied now gives the request's real start time rather than the first visible hop's.
• Reopening a chain after switching invocations no longer re-presents the previous request.
• Log lines can no longer collide on identity — which could drop a row from the stream or scroll to the wrong one.

Read on logstitch.app →

Direct download

• Download LogStitch directly from logstitch.app with a 14-day free trial that keeps working offline after the first launch — no Mac App Store required.

One-time license

• Unlock with a one-time license — Personal ($49) covers up to 3 Macs, Commercial ($99) up to 5. Activate by pasting your key or clicking the link in your purchase email, and use Deactivate This Mac to free a seat when you move to a new machine.

Automatic updates

• The direct build keeps itself current — new releases install automatically, with a Check for Updates… menu item to check on demand.

Read on logstitch.app →

Pattern detection

• Functions whose structured loggers emit the stack as a JSON array now have their stack traces extracted and grouped, instead of being reported as "no stack trace" despite carrying a valid trace.
• Running pattern detection on a Lambda that exists under multiple AWS profiles now resolves to one consistent function record, so you no longer see duplicate pattern sets split across profiles.

Read on logstitch.app →

Errors you can trust

• The function list, Overview, and function details now split a function's errors into uncaught failures (the function crashed) and caught application errors (handled and logged). CloudWatch only ever reported the uncaught ones, so a function whose errors were all handled used to show a clean health dot while hundreds of real errors sat invisible — the health dot now keys on the uncaught rate, so "noisy but working" reads yellow and "actually down" reads red.
• A new Error Class filter (Uncaught / Caught / Unclassified) narrows the invocation list to exactly the kind of error you're chasing, and the selection is remembered in saved filter presets.
• Throttled invocations now appear as their own amber chip on each function row, separate from the error badge — infrastructure rejections are no longer mistaken for application errors.

MCP

• The MCP server moved to the stateless Streamable HTTP transport, so long-running sync_function calls no longer stall when the connection goes idle mid-fetch — results return on the same connection with keep-alive heartbeats while the work runs. Existing type: sse configs must re-copy the connection snippet as type: http.
• get_function_stats and get_errors now expose the same caught vs. uncaught breakdown, so Claude can reason about whether a function is failing or just chatty.

Reliability

• The error-level log filter no longer shows an empty list on the first click.
• Per-row tooltips are back.
• Operational-status filters now apply correctly.
• The Functions sidebar section now stays visible.

Read on logstitch.app →

Live metrics from CloudWatch

• The Overview tab now paints from CloudWatch within ~500ms instead of waiting for the local invocation table to fill — Duration Trend, Error Rate, Projected Cost, Invocation Volume, and Cold Starts all draw from CloudWatch first, with local data filling in only what CloudWatch can't derive.
• Function-list rows now show live invocation counts, error counts, and a health dot from CloudWatch over the last 7 days, replacing the all-time local aggregates that went stale the moment background sync paused — the most errors and most data sort orders use the live numbers too.

Know when your data is stale

• A staleness chip now appears on the Overview tab and every drilldown — Duration, Cold Starts, Memory, Cost, Error Rate, and Invocation Volume — whenever the local data behind a card lags the window you're viewing; tap it to sync the last 7 days and it clears itself when the fetch finishes.
• Selecting a function from the list now auto-syncs it when its data is stale, so the cards have usually caught up by the time you land on the Overview tab — a per-function cooldown keeps rapid clicks from burning CloudWatch quota.

MCP

• list_functions now reports live 7-day invocation and error counts plus health from CloudWatch, merged with your local data, and tags each function with its data_source (cloudwatch_live, local_sqlite, or hybrid) — pass skip_live to bypass the AWS call.
• list_functions and get_function_stats now include freshness metadata — how much local history actually backs the numbers — so Claude can reason about how complete an answer is.

Read on logstitch.app →

Syncs that survive navigation

• On-demand log syncs now run independently of the screen that started them — navigate away from the Invocations tab or switch to another function and the fetch keeps running, then picks up exactly where it left off when you come back.
• Newly synced invocations appear in the list as each page lands — whether from an on-demand or background sync — without resetting your scroll position or dropping the invocation you had selected; counts and sparklines tick up live.
• Cancelling a sync now keeps every page already fetched instead of discarding the whole run, so the next sync resumes from where you stopped rather than re-fetching from scratch.

Reliability

• Background and on-demand syncs no longer fetch the same function at the same time — background defers to an in-flight on-demand sync and resumes on a later cycle.
• Fixed background sync silently stalling on actively-running functions — showing "0 events" cycle after cycle — when AWS's reported stream timestamps lagged behind freshly ingested events.
• Background sync now honors your configured interval after a catch-up finishes, instead of immediately re-firing a redundant cycle.
• A sync that fails mid-flight (e.g. a dropped connection) now clears its spinner and surfaces the error, instead of spinning forever and blocking every later sync with "already in progress."

MCP

• The Claude Integration settings now list every available MCP tool, so you can see what Claude Code can do before enabling the local server.

Read on logstitch.app →

Async retries are now first-class

• Lambda async invocations (EventBridge, SNS, S3, etc.) retry failed attempts and reuse the same request ID — previously each retry quietly overwrote the earlier one and its logs, so a failure followed by a successful retry looked like a plain success.
• Every attempt is now persisted as a distinct invocation, with its own status, duration, and full log lines.
• A compact retry marker on every multi-attempt invocation in the list, mirroring the cold-start indicator.
• Click the Attempt N/M badge in the detail view to jump between sibling attempts, with status and duration shown per attempt.
• MCP get_invocation_logs returns every attempt's logs, with cumulative size budgeting so high-retry requests never blow the response.

Smarter correlation across profiles

• The same execution synced under two LogStitch profiles no longer appears "correlated" with a duplicate of itself — correlated views collapse rows by logical identity (function, region, request ID, timestamp).
• Two new filters in the Invocations popover: Correlation (All / Correlated / Standalone) and Attempts (All / Multiple / Single).
• The correlated-invocations popover now anchors correctly to its button and no longer clips the "View in Timeline" action.

Faster drill-downs from charts

• View Invocations shortcut on the Error Rate and Cold Starts cards (and their full-size views) opens the Invocations list pre-filtered to exactly the rows the chart represents, in the same time window.
• Opening a larger graph from an Overview card carries the Overview tab's selected window with it, instead of resetting to each report's default.
• Opening a report from the Invocations tab's Analytics Reports popover scopes it to the time range you're already looking at.

Reliability

• Patterns tab no longer falsely marks every pattern as Resolved after a large sync backlog — the incremental refresh now drains the full backlog before classifying lifecycle.
• Inline guidance on the SSO profile screens explaining the IPv4-only access portal URL vs. the Dual-stack URL, so the first sign-in doesn't fail on a confusing OIDC error.

Read on logstitch.app →

Stitching engine

• Groups CloudWatch log lines into AWS Lambda invocations by request ID.
• Cold-start detection, init duration, and billed duration surfaced inline on every invocation.
• Auto-parses JSON log lines into a navigable, collapsible tree.

Cross-account & cross-region

• Unlimited AWS profiles via the standard ~/.aws/config, including SSO and IAM role assumption.
• Correlates a single request across functions, accounts, and regions.
• No additional CloudWatch cross-account observability setup required.

Pattern & anomaly detection

• Auto-clusters similar error messages into named patterns.
• Tracks pattern frequency, first-seen, last-seen, and runtime impact.
• Z-score anomaly scoring on duration, error rate, cold-start rate, and projected cost.

Cost & performance analytics

• Per-function memory, duration p50/p95/p99, error rate, and cost trends.
• Cost projection and memory-tuning suggestions.
• 24h, 7d, and 30d windows — re-query for free after the initial fetch.

MCP server for Claude & AI tools

• Local Model Context Protocol server bound to 127.0.0.1 only (configurable port, default 47821) — never exposed to the network.
• Works with Claude Code and any MCP-aware client that can reach a local server. Claude Desktop is not supported because its custom connectors route through Anthropic's servers.
• 20+ tools including search_logs, get_invocation_logs, get_patterns_summary, get_cost_projection, get_correlated_invocations, and run_pattern_detection.
• Your AWS credentials never leave the macOS Keychain.

Issue-tracker integrations

• Link error patterns to Jira issues and GitHub issues/PRs.
• OAuth handled by auth.logstitch.app so no client secrets ship inside the app binary.

Local-first privacy

• All logs cached in a local SQLite database on your machine.
• No LogStitch cloud, no telemetry, no account required.
• Works offline after the initial fetch.

Platform

• Native macOS app built on AppKit.
• Universal binary — runs on Intel and Apple Silicon Macs. macOS 26.1 or later.
• Distributed via the Mac App Store.
• Free 14-day trial. $49 Personal for personal projects, $99 Business for commercial use at work. One-time purchase.

Read on logstitch.app →