v2.3.0 — 10 new plot types & end-to-end auto-merge

✨ New

• /map page — force-directed spec map clustered by tag similarity, with a teaser on the landing page (#5647, #5649, #5665).
• OG images redesigned in the any.plot() visual identity — branded cards with light/dark variants (#5659).
• Mobile debug: on-device console + richer ErrorBoundary so crashes on phones are inspectable (#5808).
• Legal page gained an avatar in the operator block (#5726).
• 10 new plot types × 9 libraries: area-stacked, bar-horizontal, bland-altman-basic, count-basic, ice-basic, radar-multi, scatter-embedding, scatter-regression-polynomial, shap-waterfall, spiral-timeseries.
• Coverage filled: horizon-basic got seaborn (#5851).

🛠️ Fixes

• iOS Safari: polyfilled requestIdleCallback so the site loads on iOS (#5816).
• POTD: GitHub source link now includes the language segment (#5663); letterbox strips around the preview image removed (#5943).
• Frontend lint: 32 → 0 errors. Five react-hooks/set-state-in-effect violations refactored to React 19's "adjust state on prop change" pattern; Math.random purity issue extracted to a module-level helper (#5818).
• Analytics safety: fire-and-forget tasks in api/analytics.py could be GC'd mid-flight; now held in a module-level set (#5727).

🏷️ Tag-vocabulary policy

The auto-polish step was stripping valid niche tags (bioinformatics, confidence-interval, ohlc, …) because it treated the recommended-values tables as canonical for every dimension. Vocabulary policy is now asymmetric (#5856, #5859):

• plot_type is canonical — table is authoritative; new types added via PR.
• data_type / domain / features are advisory — any well-formed, recognized data-viz term is valid; polish only moves misclassified tags or canonicalizes synonyms.

Tables expanded with legitimate tags surfaced by an audit of all 327 specs.

🔧 Pipeline

• End-to-end auto-merge: polish PRs and impl PRs squash-merge themselves once green. gh pr merge --auto is set on creation (#5955); auto-update-pr-branches.yml keeps behind branches in sync after every push to main (#5957, #5958).
• Autonomous spec polish + cross-library similarity audit in daily-regen pre-flight (#5714).
• Daily-regen defaults to Haiku, gained spec_id and model inputs (#5706).
• Resumable regen — split into single-step invocations for local-model workflows (#5650).
• Transient retry hardened across impl-generate / impl-review (#5819, #5725, #5724).
• evaluate-plot.py switched to the canonical 6-category rubric and bumped claude_model default to claude-sonnet-4-6 (#5818).

🗂️ Repo

• Plot implementations moved under plots/{spec}/implementations/python/{library}.py (metadata under metadata/python/) to make room for other languages (#5648).
• 64 Optional[X] → X | None modernizations via UP007/UP045 (#5727).

📚 Docs

• Library expansion roadmap (#5951).
• Plausible: FCP/TTFB Web-Vitals events documented (#5818).
• /seo-proxy/map handler so bots get a real <title> for the network map (#5818).

Full Changelog: v2.2.0...v2.3.0

v2.2.0 — Regen wave, Cloudflare Access on /debug & 4-attempt pipeline

A quality-regeneration wave (12 specs re-rolled end-to-end on Sonnet, ~3.5× throughput vs. v2.1.0), two coverage gaps filled, a 4th repair attempt added to the impl pipeline, and /debug moved behind Cloudflare Access. No breaking changes.

🔁 Regen wave

12 specs regenerated through impl-generate / review / repair / merge — same plot types, fresh code, higher quality scores. Daily-regen now runs on Claude Sonnet (#5452) and skips the 20:00–24:00 Berlin window (#5517).

✅ Coverage filled

• network-basic: 8/9 → 9/9
• area-mountain-panorama: bokeh added (highcharts still in-flight)

🔐 Cloudflare Access on /debug

/debug is gated behind Cloudflare Access with a service-token fallback for CI (#5522). Same-origin /api proxy fixes the cookie scope (#5551), and a hard-reload on TypeError fires the page-gate cleanly when the JWT lapses (#5552).

🔧 Pipeline

• Repair budget bumped 3 → 4 attempts with cascading thresholds (90/80/70/60/50).
• ADMIN_TOKEN PAT for impl-merge so --admin bypasses the main-branch ruleset (#5523, #5521).

🧰 Tooling

• /Regen command for unattended library regeneration.
• /audit framework split into per-agent files; 7 new auditors added (#5413/#5451). Critical/High findings applied across multiple waves.
• Mandatory PR follow-through codified in CLAUDE.md (#5553).

⬆️ Dependencies

8 Dependabot bumps landed: uvicorn, setuptools, sqlalchemy, cloud-sql-python-connector, matplotlib, react-router-dom, plus npm-minor / python-minor / actions groups.

Full Changelog: v2.1.0...v2.2.0

v2.1.0 — Tri-state theme, pipeline resilience & 2 new plot types

Three days after the v2.0.0 relaunch: two new plot types, two coverage gaps filled, five existing specs regenerated, a tri-state theme mode, and pipeline hardening. Drop-in over v2.0.0.

🎨 New plot types

Spec	Coverage	Notes
venn-labeled-items	9/9	Venn diagram with labeled members per region
area-mountain-panorama	7/9	Spec refined mid-flight to require jagged piecewise-linear ridgelines (#5411). bokeh + highcharts still in-flight.

✅ Coverage filled

• donut-basic (plotnine): 8/9 → 9/9
• contour-basic (pygal): 8/9 → 9/9

🔁 Regen

Five specs regenerated for quality / pipeline alignment.

🌗 Tri-state theme mode

Theme toggle gained a system / light / dark state with live OS sync — site follows the OS theme by default while still respecting an explicit user override (#5412).

🔧 Pipeline resilience

The impl-generate / review / repair pipeline is now self-healing against transient Claude Code Action failures (#5410): generate retry budget bumped 2 → 3, repair auto-retries once via marker comment, review auto-retries once via repository_dispatch. All marker counts now use --paginate --per_page=100.

📊 Analytics, polish, backend

• Tracking added for landing-page hero/libraries/palette/nav/footer interactions (#5400).
• Scroll restoration on route change (#5403/#5404), code-view paper aesthetic (#5401), mobile-search width fix (#5363).
• In-memory caching for spec lookups; review feedback from prior attempts now passed back into regeneration runs.

Full Changelog: v2.0.0...v2.1.0

v2.0.0 — anyplot.ai Relaunch

The project rebrands from pyplots.ai to anyplot.ai, ships a redesigned homepage, an Okabe-Ito color palette, and first-class dark mode for both the site and every plot preview. Major because URLs, schema, and the public domain all change.

⚠️ Breaking — bookmarks should switch to https://anyplot.ai. Spec URL structure changed (see below).

Highlights

• 🎨 New homepage & visual identity: hero, libraries / palette / specifications sections, MonoLisa typography, Okabe-Ito colorblind-safe palette, theme-adaptive chrome.
• 🌗 Dark mode (site + plots): backend renders dual-theme previews — every plot ships plot-light and plot-dark variants. Theme tokens drive all chrome; mandatory theme-readability check in the AI quality pipeline.
• 🌍 Rebrand: domain migration from pyplots.ai to anyplot.ai with nginx redirect-loop fixes.
• 🔗 URL restructure (breaking): canonical spec URL is now /{specId}/{language}/{library}. Hub view consolidated under /{specId} with optional ?language= filter. Catalog → Plots rename.
• 🗄️ Schema (breaking): multi-language path restructure with new languages table. Implementations expose preview_url_light / preview_url_dark (legacy preview_url aliases the light variant).

CI / infrastructure

• 6 workflows migrated from service-account keys to Workload Identity Federation.
• Bulk-generate moved to paced sequential dispatch with a global mutex.
• Daily cron regenerates the oldest not-recently-updated spec on a 4h cadence with a Berlin-evening blackout.

Quality

• Mandatory dual-theme rendering and readability checks in impl-generate / review.
• WCAG-AA contrast pass for dark-mode adaptive chrome.
• MUI 7 → 9, Vite 7 → 8, FastMCP v3, FastAPI 0.136, anthropic 0.97, plus security bumps for cryptography / mako / authlib / python-multipart.

Migration notes

• Bookmarks: use https://anyplot.ai.
• Deep links: update /specs/{id}/{library} → /{specId}/{language}/{library} (e.g. /scatter-basic/python/matplotlib).
• MCP clients: point config at the new anyplot.ai endpoint.
• API consumers: prefer preview_url_light / preview_url_dark; legacy preview_url keeps working as the light variant.

Full Changelog: v1.1.0...v2.0.0

v1.1.0 — Visual Redesign & Theme Tokenization

Visual redesign and theme tokenization pass.

Highlights

• Theme tokenization — 80+ hardcoded hex colors and font strings centralized into app/src/theme/index.ts. New tokens for fontSize.micro/xxs and colors.primaryDark/accentBg/codeBlock/highlight/tooltipLight. Shared style constants (headingStyle, subheadingStyle, codeBlockStyle, etc.) and a centralized LIB_ABBREV map.
• WCAG AA — fixed 15+ contrast violations (gray.400 was 2.9:1); all text now meets 4.5:1. Keyboard-accessible zoom container with focus-visible.
• Plot of the Day redesign — terminal-style frame with $ python plots/... prompt linking to GitHub source, split layout, fade-in animation, real library + Python versions, status bar.
• Responsive — mobile breadcrumb abbreviations (pyplots.ai → pp, matplotlib → mpl), smoother RelatedSpecs grid progression (2 → 3 → 4 → 6).
• Code quality — fixed pre-existing lint errors (setState in useEffect → render-time ref), AbortController fetch cleanup on unmount, full style-guide doc.

Full Changelog: v1.0.0...v1.1.0

v1.0.0

pyplots v1.0.0

First formal release of pyplots.ai — an AI-powered platform for Python data visualization examples across 9 libraries.

At launch: 309 specifications, 2,668 implementations, 439K+ lines of generated Python plotting code, 1,798 interactive HTML visualizations.

What's in it

• Browse, filter, and search plots across 11 tag categories (AND/OR logic).
• Implementations with syntax-highlighted code, AI quality reviews, and interactive previews; copy code, download PNG, open fullscreen.
• Stats Dashboard at /stats — library quality/LOC histograms, coverage dot matrix, top-rated implementations, tag cloud, timeline.
• Plot of the Day — daily featured high-quality implementation on homepage.
• Similar Specifications / Implementations via tag-based Jaccard similarity.
• MCP server for AI assistant integration.

Stack

FastAPI + async SQLAlchemy + PostgreSQL · React 19 + Vite + MUI 7 + TypeScript · 13 GitHub Actions workflows (spec creation, impl generation, AI review, auto-merge) · Google Cloud Run + Cloud SQL + GCS · 1,081 unit tests.

Links

• Website: https://pyplots.ai
• MCP: claude mcp add pyplots --transport http https://api.pyplots.ai/mcp/