Release Notes

v4.7.1latest2026-10-01

Added

A public product landing page at /welcome: the 90-second tour with chapters, every feature area, security, install snippets, pricing and an FAQ
GITDASH_LANDING_PAGE=true sends signed-out visitors at / to the landing page; Sign in leads to /login. Off by default for self-hosted instances

v4.7.02026-09-29

Added

Team insights v2: "What stands out" lists findings computed from the page's numbers, worst first, each linking to its section
One 30- or 90-day window for the whole Team insights page, kept in the URL
Reviewed by a human: merges counted only when someone other than the author and not a bot reviewed them; the reviewer bus factor counts human reviews only
Who reviews whom by pull request, Workload to watch with limit markers, and a People table with CSV export
Account links (organization mode, admins): count two GitHub logins of one person as one, from a suggestion on Team insights; manage them in Settings → Account links
Settings → Team insights: the organization's workday (time zone and hours) for after-hours and weekend figures

Fixed

Team insights' median time to merge is now the median over pull requests, not the median of each person's average

Improved

After-hours and weekend figures follow the organization's workday (default Asia/Saigon 08:00–19:00, was UTC 09:00–18:00), on Team insights and each repository's Team tab
Working habits on Team insights: commits over the limit by reason, the largest commit, and oversized commits grouped by pull request
Bots are recognised the same way everywhere and never count toward team numbers; an Include bots switch shows them

v4.6.22026-09-29

Fixed

A pull request with a pending review made the nightly sync fail for its whole repository; pending reviews are now ignored

v4.6.12026-09-29

Fixed

The nightly pull-request sync could time out every night with many repositories; it now stops at 240 s and continues the next night where it stopped
Repositories with more than 1,000 pull requests never finished their backfill; progress is now saved page by page
Every nightly run re-fetched every pull request; after the backfill only pull requests updated since the last run are fetched

Improved

Working habits says when none of an owner's repositories is synced yet

v4.6.02026-09-29

Added

Working habits on Team insights: per engineer, the share of commits over the size limit (more than 10 files or 200 lines) and pull requests with more than 20 commits, for one repository or the whole owner over 30 or 90 days
Commits are measured inside each merged pull request, so a squash merge is judged by its original commits, not the one large commit it leaves behind
Engineers see their own working-habits figures on their contributor profile, even without the feature
Settings → Working habits: admins change the three limits
Alert metric "Oversized commits" and a totals-only line in the Monday leadership digest
Nightly sync at 04:47 UTC (/api/cron/sync-commit-facts) that backfills 90 days over the first nights

Improved

Contributor profile: the person lookup runs alongside the searches, and responses carry a Server-Timing header

v4.5.32026-09-28

Added

Server cache can serve the last result while it refreshes in the background, so an expired entry no longer makes the next visitor wait

Fixed

Contributor profiles undercounted pull requests and reviews — only each repository's 30 most recently closed pull requests were scanned
Average pull request size on the contributor profile was always 0
The profile funnel's reviewed and approved steps were always 0; they now count reviews other people left on the person's pull requests

Improved

Contributor profiles load about 5× faster: three parallel GitHub searches instead of walking up to 30 repositories (measured 17.4 s → 3.5 s cold, 227 → 6 GitHub requests)
Contributor profiles stay cached for 30 minutes and are refreshed in the background for up to 6 hours after that; Refresh always fetches fresh data

v4.5.22026-09-27

Added

Quick start, Access control and Caching & rate limits sections in these docs
Feature pages for the repository Workflows and Pull requests tabs; every feature page names the switch that controls it

Fixed

Change failure rate levels were documented as 5/10/15% — the app uses ≤5% Elite, ≤15% High, ≤30% Medium
Deploy frequency, lead time and hotfix detection are described as the app measures them (last 30 releases / 60 pull requests; hotfix, revert, fix-prod or emergency branches)
The Docker image name (dinhdobathi/gitdash), nightly history sync, demo mode requiring sign-in and fine-grained tokens for organizations were documented wrongly
The API reference was missing the admin endpoints and issue creation, and listed a parameter /api/github/repos does not read

Improved

Every screenshot retaken on a live instance with the 4.5 interface
Documentation search covers every section

v4.5.12026-09-27

Fixed

The FAQ raised a hydration error in the browser because one answer put a list inside a paragraph
The Helm chart still declared appVersion 4.2.8 after 4.5.0 shipped, so a chart install without an explicit image tag pulled the old image. Chart 0.7.0 declares 4.5.1

Improved

The waiting-for-access page shows where you are in the process — signed in, waiting for an admin, dashboards next — and gives you your GitHub login and numeric id to send to an admin with one click. It checks again every 15 seconds, has a Check now button, and still moves on by itself once a group is granted
Version badges in the sidebar, sign-in and waiting pages show the full version instead of major.minor, so a patch release is visible
Documentation now covers the two sign-in failures seen in practice with GITDASH_ALLOWED_ORGS: orgs that restrict OAuth Apps, and fine-grained tokens created under the user rather than the org

v4.5.02026-09-27

Added

Organization-mode permissions: fixed groups (devops, security, dev, pm, admin), per-group feature grants, an admin area with users, a permission matrix and an audit log, and a waiting page for people without a group. Enforced on the server, rolled out behind GITDASH_RBAC_ENFORCE
GitHub API caching: every GitHub read is cached per token and replicas share results through a Postgres table, keeping dashboards well inside the rate limit. Refresh buttons bypass every cache
Personal access token sign-in in organization mode, with GITDASH_ALLOWED_ORGS to limit sign-in to members of your organizations
A redesigned interface following the Graphite console design contract, and an intro video behind Explore the demo on the sign-in page

Improved

Organization mode refuses to start without DATABASE_URL and GITDASH_ADMIN_GITHUB_IDS, and /api/health answers 503, so a misconfigured rollout never goes live half-working
Alert destinations, AI and email settings and manual database sync are admin-only in organization mode

v4.2.92026-09-04

Added

Pull-request data now syncs on its own schedule, so the four people-metrics alert rules that read it can actually fire
The weekly Leadership Digest can be delivered to Slack as well as email
Dashboard metrics can be exported as CSV or JSON
An anomaly on a workflow can be filed as a GitHub issue from a confirmation dialog, using your own token. Off by default behind a feature flag

Improved

Features that write to GitHub default to off and are excluded from Enable all, so one click can never arm them

v4.2.82026-08-16

Fixed

Pages left a large band of empty space down the right-hand side on wide screens. Every page declared its own width cap — max-w-5xl on Security, max-w-6xl on Issues and Org Health, max-w-7xl on Team Analytics, none at all on Alerts and Reports — and none of them centred, so all the unused width collected in one block on the right rather than splitting evenly. On a 2000px display the Security page ended around 1010px with almost as much blank beside it as content
The app now has a single centred content container, so every page agrees on width and stays centred at any viewport. The cap is deliberately generous because this is a dashboard — the PR leaderboard alone is nine columns, and narrowing those tables to a reading measure would trade one layout problem for another. The contributor print brief keeps its narrow measure, and the documentation keeps its prose width

v4.2.72026-08-16

Fixed

The security panel blamed your token for problems it had not caused. GitHub answers with 403 both when a token is insufficient and when a security feature is simply switched off for a repository — and every 403 was mapped to "Token lacks permission", which told people to regenerate a PAT that was already sufficient. A 403 is now classified by its response body, so "Code Security must be enabled for this repository" is reported as "Not enabled for this repo" and never prompts a token change that could not have helped. Anything genuinely ambiguous still reports a permission problem, because that fix is at least in your hands
The documented claim that reading security alerts requires the security_events scope was wrong. Measured against the API, a classic PAT with repo reads Dependabot, code scanning and secret scanning alerts. The panel warning and the Security documentation now say what is actually needed, and the fine-grained PAT permissions — which genuinely are separate — are named on their own

v4.2.62026-08-15

Added

The deployments panel now separates "no deployments recorded in the last 30 days" from "none ever recorded". A repository with deployment history whose newest record is older than the window is flagged "Recording stopped", with the total on record and how long ago the last one was — a gap that usually means the pipeline was replaced or its deploy step removed, which the previous wording hid entirely

Fixed

"This repository has no deployments" read as "you do not deploy", which was misleading on repositories that deploy daily through GitHub Actions. The panel now states the actual mechanism: GitHub only writes a deployment record when a job declares an environment: key or something calls the Deployments API, so a deploy workflow without one is invisible to these metrics. The one-line fix is named directly

v4.2.52026-08-15

Added

The deployment environment driving the headline figures is now selectable — click any row under "By environment". Auto-detection can only ever guess when a repository deploys to several production targets, so choosing beats guessing better. The choice is remembered per repository
Every environment now carries its own deploys/day, change failure rate and MTTR, so switching is instant and environments can be compared directly

Fixed

The collapsible card header built for Team Analytics in v4.0.11 was left as a local function in that one page, so the DORA drill-down and PR lifecycle toggles on the repository overview kept their original bare "› Show …" text style. Extracted to a shared component and applied to both — the inconsistency was obvious once polished cards sat directly beneath a plain text link

v4.2.42026-08-15

Fixed

Deployment metrics showed 0.00 deploys/day with a null failure rate on repositories that deploy constantly. Status lookups are capped at 40 per request, and were taken from the most recent deployments overall — so on a busy repo none of them belonged to the chosen production environment, leaving every headline figure empty. The production environment is now selected first and its statuses resolved first, with the remaining budget covering other environments
Environment matching used exact names, so a platform-labelled environment like "Production — my-app" was skipped in favour of a bare, stale "Production". Matching is now word-boundary based, and within a tier the busiest environment wins. Word boundaries also stop "main" matching names like "domain-staging"
Deploy frequency reported 0.00 when no production status could be resolved. 0.00 asserts that a team ships nothing; the honest answer when statuses are unknown is that we cannot say, so it now returns null like the change failure rate and MTTR already did

Improved

The partial-sample note now explains that production is resolved first, so headline figures are complete even when other environments are sampled

v4.2.32026-08-15

Added

Command palette on ⌘K (Ctrl+K on Windows) — jump to any repository or page from anywhere. Repo-scoped destinations like Team, Issues and Security appear only while you are inside a repository, since they are meaningless without knowing which one
Fuzzy matching over both the repository name and owner/name, so "acme/api" finds what you expect. The repository list is fetched lazily on first open and shares its SWR key with the repositories page, so it costs nothing when that data is already cached

Improved

All 11 outstanding lint warnings resolved — the codebase is now completely clean. Worth stating what they were NOT: every one was an authentication transition (sign-out, PAT change, session expiry, first login) where a full page reload is deliberate, because it discards the SWR cache populated under the old token
Converting those to router.push() would have been a regression: the previous session's cached data would remain in memory and could render after sign-out. They are now documented in place with the reason, rather than silently suppressed or wrongly "fixed"

v4.2.22026-08-15

Added

Issue & Triage Health — a new page at /repos/[owner]/[repo]/issues covering the half of engineering GitDash could not see. PRs, CI and deployments measure supply; issues are where work arrives, and a backlog growing faster than it drains is a leading indicator no delivery metric will show you
Backlog direction (opened vs closed in the window), median and p90 time to close, an age distribution of the open backlog, label and assignee breakdowns, and the oldest open issue
Two triage-debt signals: unlabelled open issues (work that arrived but was never routed) and issues open 14+ days with not a single comment (someone reported it and heard nothing back)

Improved

Pull requests are excluded from every issue metric. GitHub's REST API returns PRs from the issues endpoint, so counting them would report delivery throughput as triage throughput — wrong in a direction that looks entirely plausible. There is a dedicated test for it
Costs three paginated calls with no per-issue fan-out. Time-to-first-response would need one call per issue, so "issues nobody has commented on" is used instead — a cheaper signal for the same worry
Counts are labelled as a floor rather than a total when the sample cap is hit, since the list is ordered by recent activity and quiet old issues fall outside it

v4.2.12026-08-15

Added

Deployments panel on the repository overview, using GitHub's Deployments API — real deploy frequency, change failure rate and MTTR, plus a per-environment breakdown and recent rollout history
MTTR is measured from the first failure of a streak to the next successful deploy on the same environment, which is when things actually broke rather than the last symptom before the fix

Improved

DORA figures now state their provenance. Until now deploy frequency came from Releases (falling back to counting every merged PR), and change failure rate from PRs whose branch matched /hotfix|revert|emergency/ — so a team that does not tag releases had its merge rate reported as its deploy rate, and a team that does not name branches "hotfix" got a change failure rate of exactly 0%, which reads as excellence rather than as no signal
Measured figures never silently replace estimated ones. When a repo has no deployments the panel explains what the existing numbers actually mean instead of hiding, so you always know which you are reading
Only successful production deploys count toward frequency — a failed rollout is not a delivery. Pending and in-progress deploys are excluded from the failure rate entirely, and a rate with nothing conclusive reports null rather than a flattering 0%

v4.2.02026-08-15

Added

GitHub Security Alerts on the Security page — vulnerable dependencies (Dependabot), code scanning findings, and exposed secrets, with severity counts, age of the oldest open alert, and mean time to remediate per source. Until now the page only lint-checked workflow YAML and showed none of GitHub's own findings
Exposed secrets are always ranked critical: those alerts carry no severity field, and a live credential needs no triage debate

Improved

An unreadable source can never look like a clean one. Each source reports its own status, and a permission failure renders as a loud actionable warning rather than an empty list — on a security page, conflating "we couldn't look" with "nothing found" would let a token gap read as safety
Reading these alerts needs the security_events scope, which GitDash has never requested. When a source returns 403 the panel says exactly that and how to fix it, for both classic and fine-grained tokens
The page is now titled "Security" rather than "Workflow Security", since it covers more than workflow files

v4.1.52026-08-15

Added

Bring-your-own AI provider in organization mode — Settings → AI Provider lets a team choose its own provider (Bailian, Gemini or Qwen), model, API key and optional base URL, instead of using whatever the server operator configured. Stored in the database (migration 6)
Standalone deployments are deliberately unchanged: the section does not appear and the environment defaults apply, so a self-hosted instance still works out of the box with no setup

Improved

A configured org key is used EXCLUSIVELY — the server's own keys are never tried as a fallback behind it. Falling back would silently bill the deployment owner for an organization's traffic, which is a surprise best discovered before an invoice rather than after
The API key is write-only and encrypted at rest with the same AES-256-GCM helper as email credentials; the UI shows a masked hint, and a blank field keeps the stored key. Base URLs must be https, since that URL carries the key
aiEnabled() and configuredProviders() are now async so they can account for a stored override; /api/ai/status reports which source is actually in effect

v4.1.42026-08-14

Added

AI executive summary in the Weekly Leadership Digest — 4-6 sentences of plain prose at the top of the Monday email, written for a reader who will not open the dashboard. Costs zero extra GitHub calls: it reuses the scorecard the digest already computed
The rule-based narrative is passed to the model as an anchor it may reprioritise and rephrase but never contradict, so the two halves of the email cannot disagree in front of a CTO. The AI section is explicitly labelled so a reader always knows which half a machine wrote

Fixed

The leadership digest email interpolated repository names and narrative text into an HTML body without escaping. With model-generated text now in that body, all fields are escaped — the plain-text alternative is left as-is, since it is not markup

Improved

The digest sends regardless of AI. Missing keys, provider errors, timeouts, an exhausted budget, unparseable output, and an outright throw all degrade to the rule-based narrative alone — covered by a test per failure mode
The prompt forbids week-over-week claims. Because the email is weekly, models reached for "unchanged from last week" phrasing the data cannot support: the scorecard has no previous week, and its trend field compares halves of a 30-day window

v4.1.32026-08-14

Added

Email Delivery settings — configure Resend or SendGrid from Settings instead of environment variables, so enabling alert emails, daily digests and the Weekly Leadership Digest no longer needs a redeploy. Stored in the database (migration 5)
"Send test email" button. Email was previously the one feature whose misconfiguration was invisible — an alert or a Monday digest simply never arrived, and the only trace was a server log
The API key is write-only: it is encrypted at rest with AES-256-GCM (keyed from SESSION_SECRET) and never returned to the browser. The UI shows a masked hint like "••••4f2a", and leaving the field blank keeps the stored key

Fixed

The SMTP_HOST documentation was actively misleading. That path has never spoken the SMTP protocol — it POSTs to SendGrid's HTTP API at ${SMTP_HOST}/v3/mail/send — so the documented example smtp.yourprovider.com could never work, and SMTP_PORT was never read at all. The example is corrected and the limitation stated plainly
Provider selection was duplicated across three send paths with three copies of the same env-var branching. Now resolved once in resolveEmailProvider()

Improved

Existing env-var setups are untouched: resolution is database-first with an environment fallback, so an instance already using RESEND_API_KEY keeps working after upgrading and only switches over when someone explicitly enables email in Settings
Email config is instance-wide, like alert rules — any authenticated user can change it, so the settings record and display who last changed it and when

v4.1.22026-08-14

Added

AI Failure Hypotheses — when a workflow has 3+ recent failures, the Reliability tab offers ranked likely causes with the specific evidence behind each and a confidence level. Inferred from failed job and step names, timing shifts, trigger/branch clustering and workflow-file change dates — never from run logs, which GitDash does not fetch for any feature
Confidence is validated against a literal allowlist (high/medium/low) rather than coerced, and rank is re-derived from position because models routinely emit duplicate or out-of-order ranks

Improved

Job detail is fetched only for runs that actually failed (capped at 10) instead of every run in the window — roughly 10 GitHub calls rather than 30-50, keeping the expensive part proportional to the problem rather than the window size
The GitHub fan-out is cached separately from the LLM call on this route. Elsewhere the cache is fingerprinted on the snapshot, so a hit still pays the fan-out; here the fan-out is the expensive part and gets its own key

v4.1.12026-08-14

Added

AI Anomaly Explanations — each flagged metric on the Workflow Detail reliability tab gains a "Why did duration spike?" button that explains the outliers from surrounding metadata: baseline statistics, when the workflow file last changed, and the trigger mix. Lazy by design — opening the tab costs nothing until you ask
Bailian (Alibaba Cloud) added as a third provider and tried first. It speaks the Anthropic Messages API rather than the OpenAI shape, so the provider layer now models the wire format explicitly instead of assuming one protocol

Improved

Extended thinking is disabled on Bailian requests. Qwen models there enable it by default, costing roughly 10x the output tokens for no benefit on structured extraction — measured 799 vs 85 output tokens on an identical request. The response parser still picks the text block explicitly, in case a model ignores the flag

v4.1.02026-08-14

Added

AI Insights — an optional card on the Repository Overview and Team Health Scorecard pages that turns the metrics already on screen into plain-English analysis: what changed, why it matters, and what to do next. Gemini primary with Qwen fallback, both via their OpenAI-compatible endpoints, so no vendor SDK is installed
Entirely opt-in: with no provider key configured on the server, every AI surface is hidden and GitDash behaves exactly as it did in v4.0.11. New "AI Insights" toggle in Settings → Feature Flags, and a new /api/ai/status capability probe
Privacy is enforced at the type level: prompts are built server-side from an allowlisted snapshot, and the test suite walks the serialized output to fail the build if a token, log, commit message, file path or URL ever appears. Metrics, names, logins and dates are sent; code, logs, YAML and message bodies never are
Cost controls: 15-minute snapshot-fingerprinted caching, 20 requests/minute per token, a per-attempt and total wall-clock timeout, and a daily token budget (AI_DAILY_TOKEN_BUDGET). All in-process, so they bound cost per instance rather than globally — documented as a damage-limiter, not a hard spend cap

v4.0.112026-08-13

Improved

Team Analytics page redesign, implemented from a second Claude Design import ("Section Headers"): PR Leaderboard, Reviewer Load Matrix, Review Bottleneck, Workload Risk Radar, Bus Factor Map, and Runner Utilization now use a polished card-style collapsible header with a live data count badge, instead of a plain chevron-and-text toggle. Content components unchanged — same API, same data

v4.0.102026-08-13

Improved

Team Health Scorecard loading state: a spinner + "Collecting data across up to N repositories…" status line now sits above the skeleton tiles, so a slow fetch (or one hitting GitHub's rate limits) visibly reads as working rather than frozen

v4.0.92026-08-13

Fixed

Every page logged "Error with Permissions-Policy header: Unrecognized feature: 'interest-cohort'" — a stale FLoC-related directive no browser recognizes anymore. Removed; camera/microphone/geolocation stay disabled

v4.0.82026-08-13

Improved

Team Health Scorecard redesign, implemented from a Claude Design import: estate-distribution card + stat tiles in the header, a proper grouped table (column headers, risk-band sections with count chips), interactive filter/sort pills, and bus-factor pip indicators. Presentational only — same API, same data

v4.0.72026-08-13

Fixed

Team Health Scorecard had no visible entry point from the main dashboard — the only labeled link lived in the /org/[orgName] header, reachable solely via an unlabeled icon-only "Org overview" button. The main dashboard header now shows a labeled "Team Health Scorecard" button directly whenever an org is selected

v4.0.62026-08-13

Fixed

The four v4.0.0–v4.0.3 features (Team Health Scorecard, Workload Risk Radar, 1:1 Prep Sheet, Leadership Digest) were only mentioned in changelog prose — no dedicated Features doc page, no sidebar entry, so they were effectively undiscoverable outside this release-notes tab. Each now has a full feature page with a "New in vX.Y" badge
Feature Overview index still labeled Team Insights "(In Development)" and Contributor Profile "(Coming Soon)" even though both have been fully built for several releases

Improved

New VersionBadge convention: any doc content added for a new feature (or a meaningful addition to an existing one) now carries a "New in vX.Y" pill, so what's new stays visible outside the changelog

v4.0.52026-08-13

Fixed

Contributor profile failed with "Failed to fetch contributor profile" whenever the repo owner is a personal GitHub account rather than an organization — the route always called GET /orgs/{org}/repos, which 404s for personal accounts. Now falls back to GET /users/{username}/repos on 404

v4.0.42026-08-13

Added

Org switcher: "Go to org by name" input, always available — reaches any org directly regardless of what GitHub's org-discovery API returns for your token

Fixed

Org switcher showed only "Personal repos" with no explanation or workaround when GitHub's orgs.listForAuthenticatedUser API returned an empty list (stale OAuth scope, fine-grained PAT, or org visibility settings) — now explains why and offers the manual org input as a fallback
Navigating to an org not in the auto-discovered list incorrectly showed "Personal Repos" as the page title even though the repo list was correctly filtered to that org

v4.0.32026-08-13

Added

Weekly Leadership Digest — narrative org-wide summary emailed every Monday. New leadership_digest alert metric in the Alerts page (pick an org + email, no threshold or window needed). Reuses the Team Health Scorecard's computation; no new database table, no "last sent" state — the daily cron just checks if today is Monday

v4.0.22026-08-13

Added

1:1 Prep Sheet — one-click manager brief at /contributor/[login]/brief: period-over-period PR/review/cycle-time comparison plus auto-generated talking points (conversation prompts, not verdicts). Print-friendly. Zero extra GitHub API calls — reuses the contributor profile's existing data and SWR cache

v4.0.12026-08-13

Added

Workload Risk Radar — team-wide people-risk signals: sustained after-hours/weekend work, activity cliffs ("went quiet"), and concurrent-PR overload. New Team Analytics section, framed as a conversation-starter rather than a verdict

v4.0.02026-08-13

Added

Team Health Scorecard — org-wide ranked view (Healthy / Watch / At Risk) combining DORA tier and bus-factor risk per repo, worst-first, with a throughput trend. New page at /org/[orgName]/health

Improved

Bus-factor calculation extracted into a reusable lib function (src/lib/bus-factor.ts) so the scorecard can compute it per-repo without an internal HTTP round trip; the existing /api/github/bus-factor route is unchanged in behavior

v3.2.02026-08-13

Added

Custom domain: production now serves from gitdash.info
Scheduled background sync — daily Vercel Cron re-syncs every tracked repo so Reports and Alerts stay current without a manual click
Rate-limit budget widget in the sidebar — live GitHub API quota, free to check (GET /rate_limit is excluded from rate-limit accounting)
Runner Utilization — per-runner job counts, durations, and failure rates in Team Analytics
Partial-data indicators — fan-out routes (bus-factor, open-pr-health, repo-contributors, contributor-profile, repo-dora) now report when some sub-requests were rate-limited and show an amber banner instead of an undercounted number
Anomaly-driven alerts — new anomaly_count metric fires on statistical outliers using the same detector the workflow detail page already uses
Digest alert channel — bundle matching alerts into one daily summary email instead of a notification per event
Review Bottleneck — flags overloaded reviewers and single-point-of-failure review dependencies in Team Analytics

Fixed

RecentFailuresWidget on the repo dashboard never showed newly-loaded failures — its cache-read memo depended on an identity-stable SWR cache object; now driven by a tick counter

Improved

Consolidated all Recharts imports behind one shared module — eliminates duplicated ~388 KB chunks across routes
DORA drill-down and PR lifecycle sections (collapsed by default on the repo page) now lazy-load via next/dynamic
RepoRow on the home page is memoized — search keystrokes no longer re-render the full visible table
Fan-out routes' batch-of-N loops converted to the bounded worker pool (pLimitSettled) — no more waiting on the slowest member of each batch

v3.1.32026-03-13

Added

3.x release line — responsive mobile navigation, redesigned settings, global footer, demo mode, and test infrastructure (Vitest)
DB-backed reporting and alert rules with browser and email delivery (Resend / SendGrid / SMTP)
Cost analytics via the GitHub Enhanced Billing API
Reverse-proxy / Zscaler support — redirects use the public-facing origin from x-forwarded headers; dedicated /api/health endpoint for Kubernetes probes

Fixed

Middleware activation and reverse-proxy redirect fixes (issues #3, #4); ALWAYS_PUBLIC check now runs before the HTTPS redirect so probes and static assets are never redirected

Improved

API Reference rebuilt to cover all REST endpoints; docs corrected (Neon Postgres only — no SQLite; middleware path is src/middleware.ts)
See CHANGELOG.md for the full 3.0.0 → 3.1.3 history

v2.10.12026-03-03

Added

Feature Overview index page — 13 clickable feature cards replacing the long scroll, each navigates to a dedicated feature sub-page
13 individual feature detail pages in docs (one per page in the app): Repositories, Repository Overview, Workflow Detail, Audit Trail, Security Scan, Repo Team Stats, Team Insights, Contributor Profile, Cost Analytics, Reports, Alerts, Settings, Org Overview
Workflow Detail tabs section — each of the 5 tabs (Overview, Performance, Reliability, Triggers, Runs) now has its own card with description and screenshot slots; Performance shows a 2-column screenshot grid
Screenshot slots in every feature page — drop a PNG into public/screenshots/<name>.png and it renders automatically; absent files show a placeholder with the expected filename
Screenshots for Repositories, Workflow Detail (Overview, Performance, Reliability, Triggers, Runs), and Cost Analytics now pre-loaded from docs/screenshots/
Sidebar sub-items for all 13 feature pages — visually indented, smaller text, lighter colour when inactive
⌘K search index expanded with all 13 feature sub-pages (individually searchable with accurate excerpts)
FeaturePageHeader shared component — consistent icon + name + path badge + chip row across all feature pages

Improved

Full feature audit — API Reference and feature pages refreshed to match live code
DocSearch section labels now distinguish Features / Reference / Support groups

v2.9.02026-03-03

Added

DORA 4 Keys at repository level — Deploy Frequency, Lead Time, CFR, MTTR computed from real merged PRs and GitHub Releases
DORA drill-down charts: PR Cycle Time Breakdown (segmented bar), PR Size vs Velocity (scatter + regression line), PR Throughput (12-week bar), Workflow Stability (30d line with Elite/High reference lines)
Team Insights page (/team) — global view with repo picker, sortable contributor leaderboard (8 metrics), and reviewer load heatmap
Contributor Profile page (/contributor/[login]) — KPI cards, 52-week GitHub-style activity heatmap, weekly commit chart, PR lifecycle funnel, commit hour distribution, languages touched, recent PRs table
Bus factor analysis — per-module contributor Herfindahl index, flags modules with <2 active contributors
People-based alert metrics: PR throughput drop, review response P90, after-hours commit %, PR abandon rate, unreviewed PR age
Reviewer Load Matrix component (author × reviewer heatmap) in both /team and repo team pages
/api/github/repo-dora — new endpoint returning full RepoDoraSummary (5-min cache)
/api/github/repo-contributors — new endpoint: per-contributor stats + reviewer matrix + bus factor
/api/github/contributor-profile — new endpoint: full contributor data from PR, review, and commit APIs
/api/github/bus-factor — new endpoint: per-module concentration analysis
Docs redesigned to single-section navigation (opencode.ai style) — clicking nav replaces content, no more full-page scroll
Prev / Next navigation at the bottom of each docs section

Improved

Action Duration Trend (renamed from Duration Trend) on repo overview page
Docs search (⌘K) now switches to the selected section instead of scrolling
Version badge in sidebar reads from package.json via NEXT_PUBLIC_APP_VERSION at build time

v2.3.02026-03-01

Added

Modernized /docs page with component library, search, collapsible sidebar, IntersectionObserver ToC
GitHub webhook receiver at /api/webhooks/github — workflow_run events auto-sync to Neon DB
Alert rule evaluation wired into POST /api/db/sync — rules checked after every sync
Slack webhook delivery for alert rules (channel=slack)
/docs publicly accessible before authentication (no login required)

Fixed

Sidebar no longer renders on /docs for unauthenticated visitors
SWR global 401 handler no longer redirects away from /docs

Improved

upsertRuns() replaced N+1 per-row SQL loop with single Neon HTTP transaction — up to 500× fewer round-trips per sync

v2.2.02026-02-20

Added

Reports page — DB-backed historical reporting with daily area chart and quarterly breakdown
Alert rules CRUD UI at /alerts with per-repo and per-org scopes
Neon PostgreSQL integration with idempotent schema migration (ensureSchema)
POST /api/db/sync — incremental GitHub → DB sync with cursor tracking

Improved

Sync cursor prevents re-fetching already-stored runs on repeated syncs

v2.1.02026-02-10

Added

Cost Analytics page at /cost-analytics — GitHub Actions billing breakdown by SKU/runner type
Monthly burn rate progress bar with warning/critical thresholds
Org dashboard at /org/[orgName] — reliability heatmap and sortable repo table
Audit tab, Security tab, Team stats tab at /repos/[owner]/[repo]/*

Fixed

OAuth state now expires after 5 minutes to prevent stale CSRF tokens

Improved

Repo overview fetches up to 10 workflows in parallel batches of 5

v2.0.02026-01-15

Added

Organization mode — GitHub OAuth App login with isolated per-user sessions
Multi-arch Docker image (linux/amd64 + linux/arm64)
iron-session v8 with AES-256-GCM cookie encryption
HTTP security headers: CSP, HSTS, X-Frame-Options, X-Content-Type-Options
Rate limiting on /api/auth/setup (5 req/min) and /api/auth/login (10 req/min)

Improved

SESSION_SECRET minimum length enforced at startup in production
Lazy Neon DB singleton — no build-time crash when DATABASE_URL is absent

v1.0.02025-12-01

Added

Initial release — standalone mode with PAT-based authentication
Repositories list with fuzzy search and keyboard navigation
Workflow dashboard with 5 tabs: Overview, Performance, Reliability, Triggers, Runs
Auto-refresh every 30 seconds while runs are in-progress
Browser notifications for new workflow failures (opt-in)
CSV export from the Runs tab

Release Notes

v4.7.1latest2026-10-01

Added

A public product landing page at /welcome: the 90-second tour with chapters, every feature area, security, install snippets, pricing and an FAQ
GITDASH_LANDING_PAGE=true sends signed-out visitors at / to the landing page; Sign in leads to /login. Off by default for self-hosted instances

v4.7.02026-09-29

Added

Team insights v2: "What stands out" lists findings computed from the page's numbers, worst first, each linking to its section
One 30- or 90-day window for the whole Team insights page, kept in the URL
Reviewed by a human: merges counted only when someone other than the author and not a bot reviewed them; the reviewer bus factor counts human reviews only
Who reviews whom by pull request, Workload to watch with limit markers, and a People table with CSV export
Account links (organization mode, admins): count two GitHub logins of one person as one, from a suggestion on Team insights; manage them in Settings → Account links
Settings → Team insights: the organization's workday (time zone and hours) for after-hours and weekend figures

Fixed

Team insights' median time to merge is now the median over pull requests, not the median of each person's average

Improved

After-hours and weekend figures follow the organization's workday (default Asia/Saigon 08:00–19:00, was UTC 09:00–18:00), on Team insights and each repository's Team tab
Working habits on Team insights: commits over the limit by reason, the largest commit, and oversized commits grouped by pull request
Bots are recognised the same way everywhere and never count toward team numbers; an Include bots switch shows them

v4.6.22026-09-29

Fixed

A pull request with a pending review made the nightly sync fail for its whole repository; pending reviews are now ignored

v4.6.12026-09-29

Fixed

The nightly pull-request sync could time out every night with many repositories; it now stops at 240 s and continues the next night where it stopped
Repositories with more than 1,000 pull requests never finished their backfill; progress is now saved page by page
Every nightly run re-fetched every pull request; after the backfill only pull requests updated since the last run are fetched

Improved

Working habits says when none of an owner's repositories is synced yet

v4.6.02026-09-29

Added

Working habits on Team insights: per engineer, the share of commits over the size limit (more than 10 files or 200 lines) and pull requests with more than 20 commits, for one repository or the whole owner over 30 or 90 days
Commits are measured inside each merged pull request, so a squash merge is judged by its original commits, not the one large commit it leaves behind
Engineers see their own working-habits figures on their contributor profile, even without the feature
Settings → Working habits: admins change the three limits
Alert metric "Oversized commits" and a totals-only line in the Monday leadership digest
Nightly sync at 04:47 UTC (/api/cron/sync-commit-facts) that backfills 90 days over the first nights

Improved

Contributor profile: the person lookup runs alongside the searches, and responses carry a Server-Timing header

v4.5.32026-09-28

Added

Server cache can serve the last result while it refreshes in the background, so an expired entry no longer makes the next visitor wait

Fixed

Contributor profiles undercounted pull requests and reviews — only each repository's 30 most recently closed pull requests were scanned
Average pull request size on the contributor profile was always 0
The profile funnel's reviewed and approved steps were always 0; they now count reviews other people left on the person's pull requests

Improved

Contributor profiles load about 5× faster: three parallel GitHub searches instead of walking up to 30 repositories (measured 17.4 s → 3.5 s cold, 227 → 6 GitHub requests)
Contributor profiles stay cached for 30 minutes and are refreshed in the background for up to 6 hours after that; Refresh always fetches fresh data

v4.5.22026-09-27

Added

Quick start, Access control and Caching & rate limits sections in these docs
Feature pages for the repository Workflows and Pull requests tabs; every feature page names the switch that controls it

Fixed

Change failure rate levels were documented as 5/10/15% — the app uses ≤5% Elite, ≤15% High, ≤30% Medium
Deploy frequency, lead time and hotfix detection are described as the app measures them (last 30 releases / 60 pull requests; hotfix, revert, fix-prod or emergency branches)
The Docker image name (dinhdobathi/gitdash), nightly history sync, demo mode requiring sign-in and fine-grained tokens for organizations were documented wrongly
The API reference was missing the admin endpoints and issue creation, and listed a parameter /api/github/repos does not read

Improved

Every screenshot retaken on a live instance with the 4.5 interface
Documentation search covers every section

v4.5.12026-09-27

Fixed

The FAQ raised a hydration error in the browser because one answer put a list inside a paragraph
The Helm chart still declared appVersion 4.2.8 after 4.5.0 shipped, so a chart install without an explicit image tag pulled the old image. Chart 0.7.0 declares 4.5.1

Improved

The waiting-for-access page shows where you are in the process — signed in, waiting for an admin, dashboards next — and gives you your GitHub login and numeric id to send to an admin with one click. It checks again every 15 seconds, has a Check now button, and still moves on by itself once a group is granted
Version badges in the sidebar, sign-in and waiting pages show the full version instead of major.minor, so a patch release is visible
Documentation now covers the two sign-in failures seen in practice with GITDASH_ALLOWED_ORGS: orgs that restrict OAuth Apps, and fine-grained tokens created under the user rather than the org

v4.5.02026-09-27

Added

Organization-mode permissions: fixed groups (devops, security, dev, pm, admin), per-group feature grants, an admin area with users, a permission matrix and an audit log, and a waiting page for people without a group. Enforced on the server, rolled out behind GITDASH_RBAC_ENFORCE
GitHub API caching: every GitHub read is cached per token and replicas share results through a Postgres table, keeping dashboards well inside the rate limit. Refresh buttons bypass every cache
Personal access token sign-in in organization mode, with GITDASH_ALLOWED_ORGS to limit sign-in to members of your organizations
A redesigned interface following the Graphite console design contract, and an intro video behind Explore the demo on the sign-in page

Improved

Organization mode refuses to start without DATABASE_URL and GITDASH_ADMIN_GITHUB_IDS, and /api/health answers 503, so a misconfigured rollout never goes live half-working
Alert destinations, AI and email settings and manual database sync are admin-only in organization mode

v4.2.92026-09-04

Added

Pull-request data now syncs on its own schedule, so the four people-metrics alert rules that read it can actually fire
The weekly Leadership Digest can be delivered to Slack as well as email
Dashboard metrics can be exported as CSV or JSON
An anomaly on a workflow can be filed as a GitHub issue from a confirmation dialog, using your own token. Off by default behind a feature flag

Improved

Features that write to GitHub default to off and are excluded from Enable all, so one click can never arm them

v4.2.82026-08-16

Fixed

Pages left a large band of empty space down the right-hand side on wide screens. Every page declared its own width cap — max-w-5xl on Security, max-w-6xl on Issues and Org Health, max-w-7xl on Team Analytics, none at all on Alerts and Reports — and none of them centred, so all the unused width collected in one block on the right rather than splitting evenly. On a 2000px display the Security page ended around 1010px with almost as much blank beside it as content
The app now has a single centred content container, so every page agrees on width and stays centred at any viewport. The cap is deliberately generous because this is a dashboard — the PR leaderboard alone is nine columns, and narrowing those tables to a reading measure would trade one layout problem for another. The contributor print brief keeps its narrow measure, and the documentation keeps its prose width

v4.2.72026-08-16

Fixed

The security panel blamed your token for problems it had not caused. GitHub answers with 403 both when a token is insufficient and when a security feature is simply switched off for a repository — and every 403 was mapped to "Token lacks permission", which told people to regenerate a PAT that was already sufficient. A 403 is now classified by its response body, so "Code Security must be enabled for this repository" is reported as "Not enabled for this repo" and never prompts a token change that could not have helped. Anything genuinely ambiguous still reports a permission problem, because that fix is at least in your hands
The documented claim that reading security alerts requires the security_events scope was wrong. Measured against the API, a classic PAT with repo reads Dependabot, code scanning and secret scanning alerts. The panel warning and the Security documentation now say what is actually needed, and the fine-grained PAT permissions — which genuinely are separate — are named on their own

v4.2.62026-08-15

Added

The deployments panel now separates "no deployments recorded in the last 30 days" from "none ever recorded". A repository with deployment history whose newest record is older than the window is flagged "Recording stopped", with the total on record and how long ago the last one was — a gap that usually means the pipeline was replaced or its deploy step removed, which the previous wording hid entirely

Fixed

"This repository has no deployments" read as "you do not deploy", which was misleading on repositories that deploy daily through GitHub Actions. The panel now states the actual mechanism: GitHub only writes a deployment record when a job declares an environment: key or something calls the Deployments API, so a deploy workflow without one is invisible to these metrics. The one-line fix is named directly

v4.2.52026-08-15

Added

The deployment environment driving the headline figures is now selectable — click any row under "By environment". Auto-detection can only ever guess when a repository deploys to several production targets, so choosing beats guessing better. The choice is remembered per repository
Every environment now carries its own deploys/day, change failure rate and MTTR, so switching is instant and environments can be compared directly

Fixed

The collapsible card header built for Team Analytics in v4.0.11 was left as a local function in that one page, so the DORA drill-down and PR lifecycle toggles on the repository overview kept their original bare "› Show …" text style. Extracted to a shared component and applied to both — the inconsistency was obvious once polished cards sat directly beneath a plain text link

v4.2.42026-08-15

Fixed

Deployment metrics showed 0.00 deploys/day with a null failure rate on repositories that deploy constantly. Status lookups are capped at 40 per request, and were taken from the most recent deployments overall — so on a busy repo none of them belonged to the chosen production environment, leaving every headline figure empty. The production environment is now selected first and its statuses resolved first, with the remaining budget covering other environments
Environment matching used exact names, so a platform-labelled environment like "Production — my-app" was skipped in favour of a bare, stale "Production". Matching is now word-boundary based, and within a tier the busiest environment wins. Word boundaries also stop "main" matching names like "domain-staging"
Deploy frequency reported 0.00 when no production status could be resolved. 0.00 asserts that a team ships nothing; the honest answer when statuses are unknown is that we cannot say, so it now returns null like the change failure rate and MTTR already did

Improved

The partial-sample note now explains that production is resolved first, so headline figures are complete even when other environments are sampled

v4.2.32026-08-15

Added

Command palette on ⌘K (Ctrl+K on Windows) — jump to any repository or page from anywhere. Repo-scoped destinations like Team, Issues and Security appear only while you are inside a repository, since they are meaningless without knowing which one
Fuzzy matching over both the repository name and owner/name, so "acme/api" finds what you expect. The repository list is fetched lazily on first open and shares its SWR key with the repositories page, so it costs nothing when that data is already cached

Improved

All 11 outstanding lint warnings resolved — the codebase is now completely clean. Worth stating what they were NOT: every one was an authentication transition (sign-out, PAT change, session expiry, first login) where a full page reload is deliberate, because it discards the SWR cache populated under the old token
Converting those to router.push() would have been a regression: the previous session's cached data would remain in memory and could render after sign-out. They are now documented in place with the reason, rather than silently suppressed or wrongly "fixed"

v4.2.22026-08-15

Added

Issue & Triage Health — a new page at /repos/[owner]/[repo]/issues covering the half of engineering GitDash could not see. PRs, CI and deployments measure supply; issues are where work arrives, and a backlog growing faster than it drains is a leading indicator no delivery metric will show you
Backlog direction (opened vs closed in the window), median and p90 time to close, an age distribution of the open backlog, label and assignee breakdowns, and the oldest open issue
Two triage-debt signals: unlabelled open issues (work that arrived but was never routed) and issues open 14+ days with not a single comment (someone reported it and heard nothing back)

Improved

Pull requests are excluded from every issue metric. GitHub's REST API returns PRs from the issues endpoint, so counting them would report delivery throughput as triage throughput — wrong in a direction that looks entirely plausible. There is a dedicated test for it
Costs three paginated calls with no per-issue fan-out. Time-to-first-response would need one call per issue, so "issues nobody has commented on" is used instead — a cheaper signal for the same worry
Counts are labelled as a floor rather than a total when the sample cap is hit, since the list is ordered by recent activity and quiet old issues fall outside it

v4.2.12026-08-15

Added

Deployments panel on the repository overview, using GitHub's Deployments API — real deploy frequency, change failure rate and MTTR, plus a per-environment breakdown and recent rollout history
MTTR is measured from the first failure of a streak to the next successful deploy on the same environment, which is when things actually broke rather than the last symptom before the fix

Improved

DORA figures now state their provenance. Until now deploy frequency came from Releases (falling back to counting every merged PR), and change failure rate from PRs whose branch matched /hotfix|revert|emergency/ — so a team that does not tag releases had its merge rate reported as its deploy rate, and a team that does not name branches "hotfix" got a change failure rate of exactly 0%, which reads as excellence rather than as no signal
Measured figures never silently replace estimated ones. When a repo has no deployments the panel explains what the existing numbers actually mean instead of hiding, so you always know which you are reading
Only successful production deploys count toward frequency — a failed rollout is not a delivery. Pending and in-progress deploys are excluded from the failure rate entirely, and a rate with nothing conclusive reports null rather than a flattering 0%

v4.2.02026-08-15

Added

GitHub Security Alerts on the Security page — vulnerable dependencies (Dependabot), code scanning findings, and exposed secrets, with severity counts, age of the oldest open alert, and mean time to remediate per source. Until now the page only lint-checked workflow YAML and showed none of GitHub's own findings
Exposed secrets are always ranked critical: those alerts carry no severity field, and a live credential needs no triage debate

Improved

An unreadable source can never look like a clean one. Each source reports its own status, and a permission failure renders as a loud actionable warning rather than an empty list — on a security page, conflating "we couldn't look" with "nothing found" would let a token gap read as safety
Reading these alerts needs the security_events scope, which GitDash has never requested. When a source returns 403 the panel says exactly that and how to fix it, for both classic and fine-grained tokens
The page is now titled "Security" rather than "Workflow Security", since it covers more than workflow files

v4.1.52026-08-15

Added

Bring-your-own AI provider in organization mode — Settings → AI Provider lets a team choose its own provider (Bailian, Gemini or Qwen), model, API key and optional base URL, instead of using whatever the server operator configured. Stored in the database (migration 6)
Standalone deployments are deliberately unchanged: the section does not appear and the environment defaults apply, so a self-hosted instance still works out of the box with no setup

Improved

A configured org key is used EXCLUSIVELY — the server's own keys are never tried as a fallback behind it. Falling back would silently bill the deployment owner for an organization's traffic, which is a surprise best discovered before an invoice rather than after
The API key is write-only and encrypted at rest with the same AES-256-GCM helper as email credentials; the UI shows a masked hint, and a blank field keeps the stored key. Base URLs must be https, since that URL carries the key
aiEnabled() and configuredProviders() are now async so they can account for a stored override; /api/ai/status reports which source is actually in effect

v4.1.42026-08-14

Added

AI executive summary in the Weekly Leadership Digest — 4-6 sentences of plain prose at the top of the Monday email, written for a reader who will not open the dashboard. Costs zero extra GitHub calls: it reuses the scorecard the digest already computed
The rule-based narrative is passed to the model as an anchor it may reprioritise and rephrase but never contradict, so the two halves of the email cannot disagree in front of a CTO. The AI section is explicitly labelled so a reader always knows which half a machine wrote

Fixed

The leadership digest email interpolated repository names and narrative text into an HTML body without escaping. With model-generated text now in that body, all fields are escaped — the plain-text alternative is left as-is, since it is not markup

Improved

The digest sends regardless of AI. Missing keys, provider errors, timeouts, an exhausted budget, unparseable output, and an outright throw all degrade to the rule-based narrative alone — covered by a test per failure mode
The prompt forbids week-over-week claims. Because the email is weekly, models reached for "unchanged from last week" phrasing the data cannot support: the scorecard has no previous week, and its trend field compares halves of a 30-day window

v4.1.32026-08-14

Added

Email Delivery settings — configure Resend or SendGrid from Settings instead of environment variables, so enabling alert emails, daily digests and the Weekly Leadership Digest no longer needs a redeploy. Stored in the database (migration 5)
"Send test email" button. Email was previously the one feature whose misconfiguration was invisible — an alert or a Monday digest simply never arrived, and the only trace was a server log
The API key is write-only: it is encrypted at rest with AES-256-GCM (keyed from SESSION_SECRET) and never returned to the browser. The UI shows a masked hint like "••••4f2a", and leaving the field blank keeps the stored key

Fixed

The SMTP_HOST documentation was actively misleading. That path has never spoken the SMTP protocol — it POSTs to SendGrid's HTTP API at ${SMTP_HOST}/v3/mail/send — so the documented example smtp.yourprovider.com could never work, and SMTP_PORT was never read at all. The example is corrected and the limitation stated plainly
Provider selection was duplicated across three send paths with three copies of the same env-var branching. Now resolved once in resolveEmailProvider()

Improved

Existing env-var setups are untouched: resolution is database-first with an environment fallback, so an instance already using RESEND_API_KEY keeps working after upgrading and only switches over when someone explicitly enables email in Settings
Email config is instance-wide, like alert rules — any authenticated user can change it, so the settings record and display who last changed it and when

v4.1.22026-08-14

Added

AI Failure Hypotheses — when a workflow has 3+ recent failures, the Reliability tab offers ranked likely causes with the specific evidence behind each and a confidence level. Inferred from failed job and step names, timing shifts, trigger/branch clustering and workflow-file change dates — never from run logs, which GitDash does not fetch for any feature
Confidence is validated against a literal allowlist (high/medium/low) rather than coerced, and rank is re-derived from position because models routinely emit duplicate or out-of-order ranks

Improved

Job detail is fetched only for runs that actually failed (capped at 10) instead of every run in the window — roughly 10 GitHub calls rather than 30-50, keeping the expensive part proportional to the problem rather than the window size
The GitHub fan-out is cached separately from the LLM call on this route. Elsewhere the cache is fingerprinted on the snapshot, so a hit still pays the fan-out; here the fan-out is the expensive part and gets its own key

v4.1.12026-08-14

Added

AI Anomaly Explanations — each flagged metric on the Workflow Detail reliability tab gains a "Why did duration spike?" button that explains the outliers from surrounding metadata: baseline statistics, when the workflow file last changed, and the trigger mix. Lazy by design — opening the tab costs nothing until you ask
Bailian (Alibaba Cloud) added as a third provider and tried first. It speaks the Anthropic Messages API rather than the OpenAI shape, so the provider layer now models the wire format explicitly instead of assuming one protocol

Improved

Extended thinking is disabled on Bailian requests. Qwen models there enable it by default, costing roughly 10x the output tokens for no benefit on structured extraction — measured 799 vs 85 output tokens on an identical request. The response parser still picks the text block explicitly, in case a model ignores the flag

v4.1.02026-08-14

Added

AI Insights — an optional card on the Repository Overview and Team Health Scorecard pages that turns the metrics already on screen into plain-English analysis: what changed, why it matters, and what to do next. Gemini primary with Qwen fallback, both via their OpenAI-compatible endpoints, so no vendor SDK is installed
Entirely opt-in: with no provider key configured on the server, every AI surface is hidden and GitDash behaves exactly as it did in v4.0.11. New "AI Insights" toggle in Settings → Feature Flags, and a new /api/ai/status capability probe
Privacy is enforced at the type level: prompts are built server-side from an allowlisted snapshot, and the test suite walks the serialized output to fail the build if a token, log, commit message, file path or URL ever appears. Metrics, names, logins and dates are sent; code, logs, YAML and message bodies never are
Cost controls: 15-minute snapshot-fingerprinted caching, 20 requests/minute per token, a per-attempt and total wall-clock timeout, and a daily token budget (AI_DAILY_TOKEN_BUDGET). All in-process, so they bound cost per instance rather than globally — documented as a damage-limiter, not a hard spend cap

v4.0.112026-08-13

Improved

Team Analytics page redesign, implemented from a second Claude Design import ("Section Headers"): PR Leaderboard, Reviewer Load Matrix, Review Bottleneck, Workload Risk Radar, Bus Factor Map, and Runner Utilization now use a polished card-style collapsible header with a live data count badge, instead of a plain chevron-and-text toggle. Content components unchanged — same API, same data

v4.0.102026-08-13

Improved

Team Health Scorecard loading state: a spinner + "Collecting data across up to N repositories…" status line now sits above the skeleton tiles, so a slow fetch (or one hitting GitHub's rate limits) visibly reads as working rather than frozen

v4.0.92026-08-13

Fixed

Every page logged "Error with Permissions-Policy header: Unrecognized feature: 'interest-cohort'" — a stale FLoC-related directive no browser recognizes anymore. Removed; camera/microphone/geolocation stay disabled

v4.0.82026-08-13

Improved

Team Health Scorecard redesign, implemented from a Claude Design import: estate-distribution card + stat tiles in the header, a proper grouped table (column headers, risk-band sections with count chips), interactive filter/sort pills, and bus-factor pip indicators. Presentational only — same API, same data

v4.0.72026-08-13

Fixed

Team Health Scorecard had no visible entry point from the main dashboard — the only labeled link lived in the /org/[orgName] header, reachable solely via an unlabeled icon-only "Org overview" button. The main dashboard header now shows a labeled "Team Health Scorecard" button directly whenever an org is selected

v4.0.62026-08-13

Fixed

The four v4.0.0–v4.0.3 features (Team Health Scorecard, Workload Risk Radar, 1:1 Prep Sheet, Leadership Digest) were only mentioned in changelog prose — no dedicated Features doc page, no sidebar entry, so they were effectively undiscoverable outside this release-notes tab. Each now has a full feature page with a "New in vX.Y" badge
Feature Overview index still labeled Team Insights "(In Development)" and Contributor Profile "(Coming Soon)" even though both have been fully built for several releases

Improved

New VersionBadge convention: any doc content added for a new feature (or a meaningful addition to an existing one) now carries a "New in vX.Y" pill, so what's new stays visible outside the changelog

v4.0.52026-08-13

Fixed

Contributor profile failed with "Failed to fetch contributor profile" whenever the repo owner is a personal GitHub account rather than an organization — the route always called GET /orgs/{org}/repos, which 404s for personal accounts. Now falls back to GET /users/{username}/repos on 404

v4.0.42026-08-13

Added

Org switcher: "Go to org by name" input, always available — reaches any org directly regardless of what GitHub's org-discovery API returns for your token

Fixed

Org switcher showed only "Personal repos" with no explanation or workaround when GitHub's orgs.listForAuthenticatedUser API returned an empty list (stale OAuth scope, fine-grained PAT, or org visibility settings) — now explains why and offers the manual org input as a fallback
Navigating to an org not in the auto-discovered list incorrectly showed "Personal Repos" as the page title even though the repo list was correctly filtered to that org

v4.0.32026-08-13

Added

Weekly Leadership Digest — narrative org-wide summary emailed every Monday. New leadership_digest alert metric in the Alerts page (pick an org + email, no threshold or window needed). Reuses the Team Health Scorecard's computation; no new database table, no "last sent" state — the daily cron just checks if today is Monday

v4.0.22026-08-13

Added

1:1 Prep Sheet — one-click manager brief at /contributor/[login]/brief: period-over-period PR/review/cycle-time comparison plus auto-generated talking points (conversation prompts, not verdicts). Print-friendly. Zero extra GitHub API calls — reuses the contributor profile's existing data and SWR cache

v4.0.12026-08-13

Added

Workload Risk Radar — team-wide people-risk signals: sustained after-hours/weekend work, activity cliffs ("went quiet"), and concurrent-PR overload. New Team Analytics section, framed as a conversation-starter rather than a verdict

v4.0.02026-08-13

Added

Team Health Scorecard — org-wide ranked view (Healthy / Watch / At Risk) combining DORA tier and bus-factor risk per repo, worst-first, with a throughput trend. New page at /org/[orgName]/health

Improved

Bus-factor calculation extracted into a reusable lib function (src/lib/bus-factor.ts) so the scorecard can compute it per-repo without an internal HTTP round trip; the existing /api/github/bus-factor route is unchanged in behavior

v3.2.02026-08-13

Added

Custom domain: production now serves from gitdash.info
Scheduled background sync — daily Vercel Cron re-syncs every tracked repo so Reports and Alerts stay current without a manual click
Rate-limit budget widget in the sidebar — live GitHub API quota, free to check (GET /rate_limit is excluded from rate-limit accounting)
Runner Utilization — per-runner job counts, durations, and failure rates in Team Analytics
Partial-data indicators — fan-out routes (bus-factor, open-pr-health, repo-contributors, contributor-profile, repo-dora) now report when some sub-requests were rate-limited and show an amber banner instead of an undercounted number
Anomaly-driven alerts — new anomaly_count metric fires on statistical outliers using the same detector the workflow detail page already uses
Digest alert channel — bundle matching alerts into one daily summary email instead of a notification per event
Review Bottleneck — flags overloaded reviewers and single-point-of-failure review dependencies in Team Analytics

Fixed

RecentFailuresWidget on the repo dashboard never showed newly-loaded failures — its cache-read memo depended on an identity-stable SWR cache object; now driven by a tick counter

Improved

Consolidated all Recharts imports behind one shared module — eliminates duplicated ~388 KB chunks across routes
DORA drill-down and PR lifecycle sections (collapsed by default on the repo page) now lazy-load via next/dynamic
RepoRow on the home page is memoized — search keystrokes no longer re-render the full visible table
Fan-out routes' batch-of-N loops converted to the bounded worker pool (pLimitSettled) — no more waiting on the slowest member of each batch

v3.1.32026-03-13

Added

3.x release line — responsive mobile navigation, redesigned settings, global footer, demo mode, and test infrastructure (Vitest)
DB-backed reporting and alert rules with browser and email delivery (Resend / SendGrid / SMTP)
Cost analytics via the GitHub Enhanced Billing API
Reverse-proxy / Zscaler support — redirects use the public-facing origin from x-forwarded headers; dedicated /api/health endpoint for Kubernetes probes

Fixed

Middleware activation and reverse-proxy redirect fixes (issues #3, #4); ALWAYS_PUBLIC check now runs before the HTTPS redirect so probes and static assets are never redirected

Improved

API Reference rebuilt to cover all REST endpoints; docs corrected (Neon Postgres only — no SQLite; middleware path is src/middleware.ts)
See CHANGELOG.md for the full 3.0.0 → 3.1.3 history

v2.10.12026-03-03

Added

Feature Overview index page — 13 clickable feature cards replacing the long scroll, each navigates to a dedicated feature sub-page
13 individual feature detail pages in docs (one per page in the app): Repositories, Repository Overview, Workflow Detail, Audit Trail, Security Scan, Repo Team Stats, Team Insights, Contributor Profile, Cost Analytics, Reports, Alerts, Settings, Org Overview
Workflow Detail tabs section — each of the 5 tabs (Overview, Performance, Reliability, Triggers, Runs) now has its own card with description and screenshot slots; Performance shows a 2-column screenshot grid
Screenshot slots in every feature page — drop a PNG into public/screenshots/<name>.png and it renders automatically; absent files show a placeholder with the expected filename
Screenshots for Repositories, Workflow Detail (Overview, Performance, Reliability, Triggers, Runs), and Cost Analytics now pre-loaded from docs/screenshots/
Sidebar sub-items for all 13 feature pages — visually indented, smaller text, lighter colour when inactive
⌘K search index expanded with all 13 feature sub-pages (individually searchable with accurate excerpts)
FeaturePageHeader shared component — consistent icon + name + path badge + chip row across all feature pages

Improved

Full feature audit — API Reference and feature pages refreshed to match live code
DocSearch section labels now distinguish Features / Reference / Support groups

v2.9.02026-03-03

Added

DORA 4 Keys at repository level — Deploy Frequency, Lead Time, CFR, MTTR computed from real merged PRs and GitHub Releases
DORA drill-down charts: PR Cycle Time Breakdown (segmented bar), PR Size vs Velocity (scatter + regression line), PR Throughput (12-week bar), Workflow Stability (30d line with Elite/High reference lines)
Team Insights page (/team) — global view with repo picker, sortable contributor leaderboard (8 metrics), and reviewer load heatmap
Contributor Profile page (/contributor/[login]) — KPI cards, 52-week GitHub-style activity heatmap, weekly commit chart, PR lifecycle funnel, commit hour distribution, languages touched, recent PRs table
Bus factor analysis — per-module contributor Herfindahl index, flags modules with <2 active contributors
People-based alert metrics: PR throughput drop, review response P90, after-hours commit %, PR abandon rate, unreviewed PR age
Reviewer Load Matrix component (author × reviewer heatmap) in both /team and repo team pages
/api/github/repo-dora — new endpoint returning full RepoDoraSummary (5-min cache)
/api/github/repo-contributors — new endpoint: per-contributor stats + reviewer matrix + bus factor
/api/github/contributor-profile — new endpoint: full contributor data from PR, review, and commit APIs
/api/github/bus-factor — new endpoint: per-module concentration analysis
Docs redesigned to single-section navigation (opencode.ai style) — clicking nav replaces content, no more full-page scroll
Prev / Next navigation at the bottom of each docs section

Improved

Action Duration Trend (renamed from Duration Trend) on repo overview page
Docs search (⌘K) now switches to the selected section instead of scrolling
Version badge in sidebar reads from package.json via NEXT_PUBLIC_APP_VERSION at build time

v2.3.02026-03-01

Added

Modernized /docs page with component library, search, collapsible sidebar, IntersectionObserver ToC
GitHub webhook receiver at /api/webhooks/github — workflow_run events auto-sync to Neon DB
Alert rule evaluation wired into POST /api/db/sync — rules checked after every sync
Slack webhook delivery for alert rules (channel=slack)
/docs publicly accessible before authentication (no login required)

Fixed

Sidebar no longer renders on /docs for unauthenticated visitors
SWR global 401 handler no longer redirects away from /docs

Improved

upsertRuns() replaced N+1 per-row SQL loop with single Neon HTTP transaction — up to 500× fewer round-trips per sync

v2.2.02026-02-20

Added

Reports page — DB-backed historical reporting with daily area chart and quarterly breakdown
Alert rules CRUD UI at /alerts with per-repo and per-org scopes
Neon PostgreSQL integration with idempotent schema migration (ensureSchema)
POST /api/db/sync — incremental GitHub → DB sync with cursor tracking

Improved

Sync cursor prevents re-fetching already-stored runs on repeated syncs

v2.1.02026-02-10

Added

Cost Analytics page at /cost-analytics — GitHub Actions billing breakdown by SKU/runner type
Monthly burn rate progress bar with warning/critical thresholds
Org dashboard at /org/[orgName] — reliability heatmap and sortable repo table
Audit tab, Security tab, Team stats tab at /repos/[owner]/[repo]/*

Fixed

OAuth state now expires after 5 minutes to prevent stale CSRF tokens

Improved

Repo overview fetches up to 10 workflows in parallel batches of 5

v2.0.02026-01-15

Added

Organization mode — GitHub OAuth App login with isolated per-user sessions
Multi-arch Docker image (linux/amd64 + linux/arm64)
iron-session v8 with AES-256-GCM cookie encryption
HTTP security headers: CSP, HSTS, X-Frame-Options, X-Content-Type-Options
Rate limiting on /api/auth/setup (5 req/min) and /api/auth/login (10 req/min)

Improved

SESSION_SECRET minimum length enforced at startup in production
Lazy Neon DB singleton — no build-time crash when DATABASE_URL is absent

v1.0.02025-12-01

Added

Initial release — standalone mode with PAT-based authentication
Repositories list with fuzzy search and keyboard navigation
Workflow dashboard with 5 tabs: Overview, Performance, Reliability, Triggers, Runs
Auto-refresh every 30 seconds while runs are in-progress
Browser notifications for new workflow failures (opt-in)
CSV export from the Runs tab