# Changelog

## 0.14.1 - 2026-04-26

### Features

- Models: add OpenAI fast-service support via `--fast` / `--service-tier` and `--thinking`, keep `gpt-fast` / `fast` as compatibility aliases, and add explicit `codex-fast` for Codex CLI users.

## 0.14.0 - 2026-04-26

### Features

- Chrome extension: add a background extractor router with Reddit thread `.json` extraction and diagnostics while preserving media URL hard-switch behavior (#207, fixes #174, thanks @solomonneas).
- CLI extraction: support `--extract` for local PDF files through the existing markitdown preprocessing path, without requiring an LLM (#203, thanks @mvance).

### Fixes

- CLI auto models: include config-provided environment values when selecting `auto` candidates, so API keys in `~/.summarize/config.json` are honored for URL summaries (#206, fixes #205, thanks @kaihendry and @solomonneas).
- Chrome extension: skip always-on content scripts on Facebook, Instagram, and Meta CDN pages to avoid site compatibility issues (#208, fixes #106, thanks @solomonneas).

### Maintenance

- Dependencies: refresh direct and transitive packages to the latest compatible versions, including `es-toolkit` 1.46.0.

## 0.13.1 - 2026-04-22

### Fixes

- YouTube and cache: make `--no-cache` bypass cached URL extraction, forward `OPENAI_API_KEY` into media transcription, and surface yt-dlp transcription failures in diagnostics (#197, thanks @mvance).
- Chrome extension chat: isolate side-panel chat history by both tab and URL so navigating within a tab no longer shows another page's conversation (#189, thanks @Youpen-y).
- Chrome extension chat: honor `openai.useChatCompletions` for side-panel chat requests, including fixed and auto-selected OpenAI models (#155, thanks @Zevan770).
- Spotify podcasts: skip encrypted Spotify embed audio, fall back to publisher RSS enclosures, and surface podcast transcription failures instead of summarizing a bare URL.
- X extraction: surface unauthorized `xurl` responses with actionable auth/fallback guidance when Nitter is unavailable (#200, thanks @coygeek).
- OpenClaw CLI: call current OpenClaw with `-m/--message` and reject oversized prompts before hitting argv limits (#199, thanks @Silver-Aurora).
- Windows daemon: register the logon Scheduled Task via XML with battery-safe hidden launch settings, fix restart/uninstall process cleanup, and document the Administrator install flow (#192, thanks @ajmeese7).
- CLI providers: use stable default aliases for Gemini (`flash`) and Cursor Agent (`auto`) so installed CLI versions resolve supported models reliably (#193, thanks @mvance).
- CLI build: mark the generated `dist/cli.js` wrapper executable so `npm link` and global installs can run the binary directly on Unix-like systems (#191, thanks @maciej).
- CLI progress: show only the active transcription provider/model in status text instead of the full remote fallback chain.

## 0.13.0 - 2026-04-08

### Features

- Slides: support `--slides` for local video files in the main CLI and `summarize slides`, route local videos through the shared slide-aware flow, and document the local-file workflow (#149, thanks @steipete).
- Models: add explicit `github-copilot/...` model support backed by GitHub Models, including shorthand ids like `github-copilot/gpt-5.4` and `GITHUB_TOKEN` / `GH_TOKEN` auth.
- Models: add OpenCode as a first-class CLI provider across CLI flags, config, auto fallback, daemon picker/chat flows, and Chrome extension settings, while preserving existing OpenClaw behavior (#169, thanks @maciej).
- CLI providers: add OpenClaw as a configurable CLI backend (`--cli openclaw`, `cli/openclaw/...`, `openclaw/...`) across config, daemon discovery, and docs (#165, thanks @yqf-ai).
- Config: allow setting a default summary length via `output.length`, and keep prompt-override runs aligned with the configured length/language defaults in both CLI and daemon flows (#178, thanks @maciej).
- Media detection/cache: recognize `.m3u8` HLS playlists as direct media inputs and preserve the playlist extension in the media cache (#159, thanks @mdsakalu).

### Fixes

- OpenAI models: route GPT-5.4 / GPT-5.4 mini / GPT-5.4 nano / GPT-5 mini / GPT-5 nano text requests through direct provider APIs instead of the stale generic parser, preserve the real `gpt-5.4-mini` / `gpt-5.4-nano` ids end-to-end, and fall GitHub Models OpenAI GPT-5-family requests back to `gpt-5-chat` when GitHub rejects the raw id.
- GitHub Models: make `github-copilot/...` shorthand inference family-based instead of pinning old exact prefixes, so newer ids like `gpt-5.4`, `o5`, and `claude-opus-4.6` normalize correctly when the backend exposes them.
- Slides/local video: transcribe direct videos for slide summaries, avoid fake local-file “Downloading audio” phases, and keep progress text visible while slide extraction runs.
- Chrome extension slides: restore slide text/session state more reliably so reruns and reloads do not leave stale or blank slide summaries.
- Transcription: retry Groq Whisper uploads via `curl` when Node multipart uploads get a 403, fixing local `.ogg` regressions on some environments.
- YouTube: detect obviously truncated caption-track transcripts on long videos and fall through to yt-dlp transcription instead of caching a broken partial result (#184, thanks @sportiz91).
- YouTube: treat yt-dlp “no audio stream” videos as a non-fatal unavailable transcript case so summarize can continue cleanly with an explanatory note (#161, thanks @mdsakalu).
- Cache: include the prompt `<context>` block in summary cache hashing and bump the cache format version so stale cross-page summary collisions cannot be reused (#171, thanks @mvance).
- CLI providers: stream OpenClaw prompts over stdin instead of `--message`, make daemon side-panel chat honor `openai.useChatCompletions`/custom OpenAI-compatible base URLs, and stop leaking raw Codex JSONL events like `thread.started` when no assistant text was produced.
- Chrome extension: add a copy button for rendered summaries so results can be copied without manual selection.
- Chrome extension chat: handle plain-string assistant replies in the side-panel agent loop instead of crashing on `.filter()` tool-call extraction (#186, thanks @Youpen-y).
- Windows containers: let `summarize daemon install` start the daemon for the current container session without Scheduled Task registration, keep `0.0.0.0` binding Windows-only, and probe slide tools by spawning commands when PATH lookup is unreliable (#152, thanks @mathicg).
- Windows daemon: keep Scheduled Task startup hidden without breaking `summarize daemon restart` or uninstall by tracking the hidden daemon PID and killing that process tree before reruns/removal (#146, thanks @mathicg).
- Whisper.cpp: honor config-resolved transcription env overrides for readiness checks, model display, and local transcription so custom binary/model paths work outside `process.env` (#160, thanks @mdsakalu).
- Daemon models: gracefully fall back for unrecognized custom models when using proxy base URLs instead of crashing on undefined API metadata (#175, thanks @douo).
- Docs/setup: switch Homebrew instructions from the old tap to the official `brew install summarize` formula, including the side-panel setup UI and release checklist (#172, thanks @zeldrisho).
- Chrome extension: detect blank `userAgentData.platform` browsers like Vivaldi by falling back to `navigator.platform` before choosing OS-specific setup instructions (#158, thanks @bytrangle).
- Firecrawl: reject `--firecrawl always` for YouTube URLs with an explicit guidance error instead of silently skipping Firecrawl on the transcript-first path (#145, thanks @steipete).
- YouTube: keep Gemini-only no-caption runs on the transcription path by forwarding the Google API key from the top-level URL flow into link-preview transcription config (#148, thanks @bytrangle).
- Homebrew: make the tap formula fail clearly on Linux instead of installing a macOS binary, and add generator/test coverage for the macOS-only guard (#147, thanks @steipete).
- Maintenance: update the GitHub Pages workflow to `actions/configure-pages@v6` and `actions/deploy-pages@v5` (#182, thanks @dependabot).

## 0.12.0 - 2026-03-11

### Features

- Models: add `nvidia/...` provider alias (uses `NVIDIA_API_KEY` + optional `NVIDIA_BASE_URL`) for NVIDIA OpenAI-compatible endpoints.

### Fixes

- Transcription: add AssemblyAI as a first-class remote provider across direct media, podcast/RSS, and yt-dlp YouTube fallback; refactor remote fallback ordering, expand config/env support (`ASSEMBLYAI_API_KEY`, legacy `apiKeys.assemblyai`), and add AssemblyAI unit + live coverage (#126).
- X/Twitter: prefer `xurl` for tweet extraction when installed, fall back to `bird`, preserve long-form/article text plus media URLs, add live `xurl` extraction/media coverage, and replace the stale dead-`bird` install tip with a current X CLI recommendation (#70).
- Models: make daemon agent `artifacts` schemas Gemini-safe, improve Google empty-response handling with preview-to-stable fallback, and switch CLI/auto Gemini defaults away from brittle preview behavior (#82, #96).
- Agents: expand model auto-resolution errors with checked models, missing env/CLI setup, and daemon restart guidance (#107).
- Daemon: support multiple saved extension tokens, migrate legacy single-token configs, and accept any configured token for auth (#116).
- Chrome extension: harden side-panel slides so SSE keepalives no longer false-time out, seeded placeholders no longer block pending/cached slide runs, retries can start a fresh summarize+slides run, and reruns replace stale slide state.
- Chrome extension: refactor side-panel navigation/run attachment policy so late summary/slide runs no longer attach to the wrong page after tab or URL switches, and expand headless regression coverage for pending-run resume and slide-mode transitions.
- Chrome extension: default fresh installs to slide mode, keep passive tab navigation out of chat, and align slide cards with CLI `--slides` by preferring per-slide summary text over raw transcript/OCR fallback.
- Chrome extension tests: add stronger YouTube slide E2E coverage for loaded images, summary-backed slide text, and switching between videos mid-analysis without stale slide-summary bleed.
- Chrome extension: isolate slide-summary stream callbacks per run and harden Playwright settings hydration so late events no longer blank slide text when switching videos mid-analysis.
- Transcription: add Gemini audio/video transcription support across direct media, podcast/RSS, and yt-dlp YouTube fallback, including Files API uploads for larger media plus new Gemini live coverage (#89).
- npm packaging: publish CLI with `pnpm publish` so `@steipete/summarize-core` is version-pinned in published metadata (no `workspace:*` in registry package).
- Slides: detect WezTerm as an iTerm-compatible terminal for inline slide images in `--slides` mode. (#133) — thanks @doodaaatimmy-creator.
- CLI help: surface `summarize refresh-free` in `summarize help` output.
- CLI: report CLI provider timeouts explicitly, including the duration, command, and a `--timeout` hint instead of collapsing them into generic exec failures (#100, thanks @christophsturm).
- Daemon: restrict CORS responses to trusted extension and localhost origins, with regression coverage for allowed and denied `Origin` headers (#108, thanks @sebastiondev).
- Transcription: chunk oversized Groq Whisper uploads with ffmpeg in file mode instead of failing out on files above the 30MB limit (#134, thanks @WinnCook).
- Docs: tighten landing-page mobile layout so hero, cards, code blocks, and nav stay readable on narrow screens (#118, thanks @Acidias).
- Release: build macOS x64 Bun artifacts and add regression coverage for Homebrew formula rewrites during dual-arch releases (#122, thanks @androidshu).
- YouTube: tighten hostname validation across core, slides, and extension helpers so attacker-controlled lookalike hosts are no longer treated as YouTube URLs (#91, thanks @RinZ27).
- Config: honor `zai.baseUrl` config fallback for blank env values and keep Z.AI base URL overrides working outside the summary flow (#102, thanks @liuy).
- Chrome extension: tighten options and sidepanel UI spacing, copy actions, and advanced-controls layout for a cleaner panel experience (#86, thanks @morozRed).
- Slides: warn in summary mode when `--slides` dependencies are missing, and document required local installs for `ffmpeg`, `yt-dlp`, and optional `tesseract`.
- Docs: fix broken docs index links by setting an empty Jekyll `baseurl` (#113, thanks @Youpen-y).
- Models: preserve model id casing after the provider prefix so OpenAI-compatible proxies can route exact names correctly (#128, thanks @WinnCook).
- Cache: give extract entries with unavailable transcripts the same short retry TTL as negative transcript cache entries, so transient Apify failures can recover (#115, thanks @gluneau).
- Daemon: apply the saved env snapshot to `process.env` before `daemon run` starts so child tools inherit the right PATH and API/tool config under launchd/systemd (#99, thanks @heyalchang).
- Chrome automation: require sidepanel arming before debugger-backed native input can run in a tab, and auto-disarm after browser JS execution ends (#129, thanks @omnicoder9).
- Media setup: fix the local whisper.cpp install hint to use the current Homebrew formula name `whisper-cpp` (#92, thanks @zerone0x).
- CLI output: cap markdown render width on very wide terminals by default, with a `--width` override for manual control (#119, thanks @howardpen9).
- Slides: size inline slide images from terminal width instead of keeping them pinned to 32 columns, capped at 2x the previous width while preserving `COLUMNS` fallback behavior (#125, #135, thanks @WinnCook).
- Shell completions: add Fish shell completions for the current CLI flags and option values (#95, thanks @fbehrens).
- Bun fetch: only opt into compressed HTML/YouTube responses when running under Bun, and retry link-preview fetches with `Accept-Encoding: identity` after Bun decompression failures (#105, thanks @maciej).
- Daemon: support `cli/...` models in chat and agent endpoints, including CLI auto-fallback when no API-key transport is available (#109, thanks @jetm).

## 0.11.0 - 2026-02-14

### Highlights

- Auto CLI fallback: new controls and persisted last-success provider state (`~/.summarize/cli-state.json`) for no-key/local-CLI workflows.
- Transcription reliability: Groq Whisper is now the preferred cloud transcriber, with custom OpenAI-compatible Whisper endpoint overrides.
- Input reliability: binary-safe stdin handling, local media support in `--extract`, and fixes for local-file hangs/PDF preprocessing on custom OpenAI base URLs.

### Features

- CLI: add Cursor Agent provider (`--cli agent`) for CLI-model execution.
- CLI auto mode: add implicit auto CLI fallback controls (`cli.autoFallback`, `--auto-cli-fallback`) and provider priority controls (`cli.providers`, `--cli-priority`), with persisted provider success ordering.
- Transcription: add Groq Whisper as preferred cloud provider (#71, thanks @n0an).
- Transcription: support custom OpenAI-compatible Whisper endpoints via `OPENAI_WHISPER_BASE_URL` (with safe `OPENAI_BASE_URL` fallback) (#65, thanks @toanbot).
- Config: support generic `env` defaults in `~/.summarize/config.json` (fallback for any env var), while keeping legacy `apiKeys` mapping for compatibility (#63, thanks @entropyy0).

### Fixes

- CLI local files: avoid hangs when stream usage never resolves and preprocess PDFs automatically for custom OpenAI-compatible `OPENAI_BASE_URL` endpoints (e.g. non-`api.openai.com`).
- CLI stdin: support binary-safe piping/input temp files to prevent corruption on non-text stdin (#76).
- Extract mode: allow `--extract` for local media files (#72).
- Auto model/daemon fallback: skip model attempts when required API keys are missing and normalize env-key checks in daemon fallback (#67, #78).
- Cache: for auto presets (`auto`/`free`/named auto), prefer preset-level winner cache entries so stale per-candidate cache hits don’t override newer better-model results.
- Media: treat X broadcasts (`/i/broadcasts/...`) as transcript-first media and prefer URL mode.
- YouTube: keep explicit `--youtube apify` working when HTML fetch fails, while preserving duration metadata parity (#64, thanks @entropyy0).
- Transcription: stabilize Groq-first fallback flow (no duplicate Groq retries in file mode), improve terminal error reporting, and surface Groq setup in media guidance (#71, thanks @n0an).
- Media detection: detect more direct media URL extensions including `.ogg`/`.opus` (#65, thanks @toanbot).
- Slides: allow yt-dlp cookies-from-browser via `SUMMARIZE_YT_DLP_COOKIES_FROM_BROWSER` to avoid YouTube 403s.
- Daemon install: resolve symlinked/global bin paths and Windows shims when locating the CLI for install (#57, #62, thanks @entropyy0).
- Extraction: strip hidden HTML + invisible Unicode before summarization or extract output (#61).
- CLI: honor `--lang` for YouTube transcript→Markdown conversion in `--markdown-mode llm` (#56, thanks @entropyy0).
- LLM: map Anthropic bare model ids to versioned aliases (`claude-sonnet-4` → `claude-sonnet-4-0`) (#55, thanks @entropyy0).

### Improvements

- Tooling: remove Biome and standardize on `oxfmt` + type-aware `oxlint`; `pnpm check` now enforces `format:check` before lint/tests.
- Dependencies: update workspace dependencies to latest (including `@mariozechner/pi-ai` and `oxlint-tsgolint`).

## 0.10.0 - 2026-01-22

### Highlights

- Chrome Side Panel: **Chat mode** with metrics bar, message queue, and improved context (full transcript + summary metadata, jump-to-latest).
- Media-aware summarization in the Side Panel: Page vs Video/Audio dropdown, automatic media preference on video sites, plus visible word count/duration.
- Slides: **YouTube slide screenshots + OCR + transcript-aligned cards**, timestamped seek, and an OCR/Transcript toggle.
- CLI: robust URL + media extraction with transcript-first workflows and cache-aware streaming.

### Features

- Chrome Side Panel chat: stream agent replies over SSE and restore chat history from daemon cache (#33, thanks @dougvk).
- Media-aware summarization in the Side Panel: Page vs Video/Audio dropdown, automatic media preference on video sites, plus visible word count/duration.
- Transcripts: `--timestamps` adds segment-level timings (`transcriptSegments` + `transcriptTimedText`) for YouTube, podcasts, and embedded captions.
- Summaries: when transcript timestamps are available, prompts require timestamped bullet summaries; side panel auto-links `[mm:ss]` in summaries for media.
- Chrome Side Panel chat: timestamped transcript context plus clickable `[mm:ss]` links that seek the current media.
- Slides: extract slide screenshots + OCR for YouTube/direct video URLs in the CLI + extension (#41, thanks @philippb).
- Slides: top-of-summary slide strip with expand/collapse full-width cards, timestamps, and click-to-seek.
- Slides: slide descriptions without model calls (transcript windowing, OCR fallback) + OCR/Transcript toggle.
- Slides: stream slide extraction status/progress and show a single header progress bar (no duplicate spinners).
- CLI: transcribe local audio/video files with mtime-aware transcript cache invalidation (thanks @mvance!).
- Browser extension: add Firefox sidebar build + multi-browser config (#31, thanks @vlnd0).
- CLI: add Cursor Agent CLI provider (`cli/agent`, `--cli agent`).
- Chrome automation: add artifacts tool + REPL helpers for persistent session files (notes/JSON/CSV) and downloads.
- Chrome automation: expand navigate tool with list/switch tab support and return matching skills after navigation.

### Fixes

- Extract-only: remove implicit 8k cap; new `--max-extract-characters`/daemon `maxExtractCharacters` allow opt-in limits; resolves transcript truncation.
- Media: route direct media URLs to the transcription pipeline and raise the local media limit to 2GB (#47, thanks @n0an).
- Daemon (macOS): `daemon install` now falls back from `launchctl bootstrap gui/<uid>` to `user/<uid>` and resolves sudo/root uid targeting to avoid bootstrap `Input/output error` / `Domain does not support specified action` failures (#75).
- Slides: allow yt-dlp cookies-from-browser via `SUMMARIZE_YT_DLP_COOKIES_FROM_BROWSER` to avoid YouTube 403s.
- Chrome Side Panel chat: support JSON agent replies with explicit SSE/JSON negotiation to avoid “stream ended” errors.
- Chrome Side Panel: scope streams/state per window so other windows don’t wipe active summaries.
- Chrome Side Panel chat: keep auto-scroll pinned while streaming when you’re already at the bottom.
- Chrome Side Panel chat: clear streaming placeholders on errors/aborts.
- Chrome Side Panel: add inline error toast above chat composer; errors stay visible when scrolled.
- Chrome Side Panel: clear/hide the inline error toast when no message is present to avoid empty red boxes.
- Cache: include transcript timestamp requests in extract cache keys so timed summaries don’t reuse plain transcript content.
- CLI: `--no-cache` now bypasses summary caching only; transcript/media caches still apply.
- Media: treat X broadcasts (`/i/broadcasts/...`) as transcript-first media and prefer URL mode.
- Daemon: avoid URL flow crashes when url-preference helpers are missing (ReferenceError guard).
- YouTube: prefer English caption variants (`en-*`) when selecting caption tracks.
- Prompts: ignore sponsor/ads segments in video and podcast summaries.
- Prompts: enforce no-ads/no-skipped language and italicized standout excerpts (no quotation marks).
- Slides: render Slide X/Y labels and parse slide markers more robustly in streaming output.
- Slides: parse `Slide N/Total` labels and stabilize title/body extraction.
- Slides: ensure slide summary segments start with a title line when missing.
- Slides: detect headline-style first lines and render them as slide titles (no required `Title:` markers).
- Slides: progress updates during yt-dlp downloads and OSC progress mirrors slide extraction.
- Slides: reuse the media cache for downloaded videos (even with `--no-cache`).
- Slides: clear slide progress line before the finish summary to avoid stray `Slides x/y` output.
- CLI status line: ignore empty/ANSI-only spinner updates and duplicate texts to prevent flicker/blank status frames.
- Automation: require userScripts (no isolated-world fallback), with improved guidance and in-panel permission notice.
- CLI: clear OSC progress on SIGINT/SIGTERM to avoid stuck indicators.
- CLI local files: avoid hangs when stream usage never resolves and preprocess PDFs automatically for custom OpenAI-compatible `OPENAI_BASE_URL` endpoints (e.g. non-`api.openai.com`).

### Improvements

- Tooling: remove Biome and standardize on `oxfmt` + type-aware `oxlint`; `pnpm check` now enforces `format:check` before lint/tests.
- Transcription: add auto transcriber selection (default) with ONNX-first when configured + `summarize transcriber setup`.
- Cache: add media download caching with TTL/size caps + optional verification, plus `--no-media-cache`.
- CLI: add themed output (24-bit ANSI), `--theme`, and config/env defaults for a consistent color scheme.
- CLI: show determinate transcription progress percent when duration is known.
- CLI: show determinate OSC progress for transcription/download when totals are known.
- CLI: keep OSC progress determinate when recent percent updates are available.
- CLI: theme transcription progress lines and mirror part-based progress to OSC when duration is unknown.
- CLI: theme tweet/extraction progress lines for consistent loading indicators.
- CLI: theme file/slide spinner labels so all progress lines share the same styling.
- CLI: simplify media download labels (avoid “media, video” duplication).
- Slides: cap auto slide targets at 6 by default for long videos.
- Slides: render headline-style first lines as slide titles above the slide marker.
- Media: refactor routing helpers and size policy (#48, thanks @steipete).
- Daemon: emit slides start/progress/done metadata in extended logging for easier debugging.
- Prompts: allow straight quotes and encourage 1-2 short exact quotes when relevant.

### Docs

- README: 0.10.0 preview layout with clearer install flow, daemon rationale, and prominent Chrome Web Store link.
- README/docs: add UI theme config + ONNX install hints.
- README: document ONNX transcriber setup + auto selection.

## 0.9.0 - 2025-12-31

### Highlights

- Chrome Side Panel: **Chat mode** with metrics bar, message queue, and improved context (full transcript + summary metadata, jump-to-latest, smoother auto-scroll).
- Media-aware summarization in the Side Panel: Page vs Video/Audio dropdown, automatic media preference on video sites, plus visible word count/duration.
- Chrome extension: optional hover tooltip summaries for links (advanced setting, default off; experimental) with prompt customization.

### Improvements

- PDF + asset handling: send PDFs directly to Anthropic/OpenAI/Gemini when supported; generic PDF attachments and better media URL detection.
- Daemon: `/v1/chat` + `extractOnly`, version in health/status pill, optional JSON log with rotation, and more resilient restart/install health checks.
- Side Panel: advanced model row with “Scan free” (shows top free model after scan), a refresh summary control (cache bypass), plus richer length tooltips.
- Side Panel UX: consolidated advanced layout and typography controls (font size A/AA, line-height), streamlined setup panel with inline copy, clearer status text, and tighter model/length controls.
- Side Panel UX: keep the Auto summarize toggle on one line in Advanced.
- Streaming/metrics polish: faster stream flushes, shorter OpenRouter labels on wrap, and improved extraction metadata in chat.

### Fixes

- Auto model selection: OpenRouter fallback now resolves provider-specific ids (dash/dot slug normalization) and skips fallback when no unique match.
- Language auto: default to English when detection is uncertain.
- OpenAI GPT-5: skip `temperature` in streaming requests to avoid 400s for unsupported params.
- Side Panel stability: retryable stream errors, no abort crash, auto-summarize on open/source switch, synced chat toggle state, and caret alignment.
- YouTube duration handling: player API/HTML/yt-dlp fallbacks, transcript metadata propagation, and extension duration fallbacks.
- URL extraction: preserve final redirected URLs so shorteners (t.co) summarize the real destination.
- Hover summaries: proxy localhost daemon calls to avoid Chrome “Local network access” prompts.
- Install: use npm releases for osc-progress/tokentally instead of git deps.

## 0.8.2 - 2025-12-28

### Breaking

- ESM-only: `@steipete/summarize` + `@steipete/summarize-core` no longer support CommonJS `require()`; the CLI binary is now ESM.

### Highlights

- Chrome: add a real **Side Panel** extension (MV3) that summarizes the **current tab** and renders streamed Markdown.
- Daemon: add `summarize daemon …` (localhost server on `127.0.0.1:8787`) for extension ↔ CLI integration.
  - Autostart: macOS LaunchAgent, Linux systemd user service, Windows Scheduled Task
  - Token pairing (shared secret)
  - Streaming over SSE
  - Emit finish-line metrics over SSE (panel footer + hover details)
  - Commands: `install`, `status`, `restart`, `uninstall`, `run`
- Cache: add SQLite cache for transcripts/extractions/summaries with `--no-cache`, `--cache-stats`, `--clear-cache` + config (`cache.enabled/maxMb/ttlDays/path`).
  - Finish line shows “Cached” for summary cache hits (CLI + daemon/extension)
  - Daemon/Chrome stream cache status metadata (`summaryFromCache`)

### Features

- YouTube: add `--youtube no-auto` to skip auto-generated captions and prefer creator-uploaded captions; fall back to `yt-dlp` transcription (thanks @dougvk!).
- CLI: add transcript → Markdown formatting via `--extract --format md --markdown-mode llm` (thanks @dougvk!).
- X/Twitter: auto-transcribe tweet videos via `yt-dlp`, using browser cookies (Chrome → Safari → Firefox) when available; set `TWITTER_COOKIE_SOURCE` / `TWITTER_*_PROFILE` to control cookie extraction order.
- Prompt overrides: add `--prompt`, `--prompt-file`, and config `prompt` to replace the default summary instructions.
- Chrome Side Panel: add length + language controls (presets + custom), forwarded to the daemon.
- Daemon API: `mode: "auto"` accepts both `url` + extracted page `text`; daemon picks the best pipeline (YouTube/podcasts/media → URL, otherwise prefer visible page text) with a fallback attempt.
- Daemon/Chrome: stream extra run metadata (`inputSummary`, `modelLabel`) over SSE for richer panel status.
- Core: expose lightweight URL helpers at `@steipete/summarize-core/content/url` (YouTube/Twitter/podcast/direct-media detection).
- Chrome Side Panel: new icon + extension `homepage_url` set to `summarize.sh`.
- Providers: add configurable API base URLs (config + env) for OpenAI/Anthropic/Google/xAI (thanks @bunchjesse for the nudge).

### Fixes

- Packaging: ensure runtime deps and core tarball are included in published CLI bundles.

### Improvements

- Chrome Side Panel: stream SSE from the panel (no MV3 background stalls), use runtime messaging to avoid “disconnected port” errors, and improve auto-summarize de-dupe.
- Chrome Side Panel UI: working status in header + 1px progress line (no layout jump), full-width subtitle, page title in header, idle subtitle shows `words/chars` (or media duration + words) + model, subtle metrics footer, continuous background, and native highlight/link accents.
- Daemon: prefer the installed env snapshot over launchd’s minimal environment (improves `yt-dlp` / `whisper.cpp` PATH reliability, especially for X/Twitter video transcription).
- X/Twitter: cookie handling now delegates to `yt-dlp --cookies-from-browser` (no sweet-cookie dependency).
- X/Twitter: skip yt-dlp transcript attempts for long-form tweet text (articles).
- Transcripts: show yt-dlp download progress bytes and stabilize totals to prevent bouncing progress bars.
- Finish line: show transcript source labels (`YouTube` / `podcast`) without repeating the label.
- Streaming: stop/clear progress UI before first streamed output and avoid leading blank lines on non-TTY stdout.
- URL flow: propagate `extracted.truncated` into the prompt context so summaries can reflect partial inputs.
- Daemon: unify URL/page summarization with the CLI flows (single code path; keeps extract/cache/model logic in sync).
- Prompts: auto-require Markdown section headings for longer summaries (xl/xxl or large custom lengths).

## 0.7.1 - 2025-12-26

### Fixed

- Packaging: `@steipete/summarize-core` now ships a CJS build for `require()` consumers (fixes `pnpm dlx @steipete/summarize --help` and the published CLI runtime).

## 0.7.0 - 2025-12-26

### Highlights

- Packages: split into `@steipete/summarize-core` (library) + `@steipete/summarize` (CLI; depends on core). Versions are lockstep.
- Streaming: scrollback-safe Markdown streaming (hybrid: line-by-line + block buffering for fenced code + tables). No cursor control, no full-frame redraws.
- Output: Markdown rendering is automatic on TTY; use `--plain` for raw Markdown/text output.
- Finish line: compact separators (`·`) and no duplicated `… words` when transcript stats are shown.
- YouTube: `--youtube auto` prefers `yt-dlp` transcription when available; Apify is last-last resort.

### Fixed

- Streaming: flush newline-bounded output in `--plain` mode to avoid duplication with cumulative stream chunks.
- Website extraction: strip inline CSS before Readability to avoid extremely slow jsdom stylesheet parsing on some pages.
- Twitter/X: rotate Nitter hosts and skip Anubis PoW pages during tweet fallback.

### Changed

- CLI: remove `--render`; add `--plain` to keep raw output (no ANSI/OSC rendering).

## 0.6.1 - 2025-12-25

### Changes

- YouTube: `--youtube auto` now falls back to `yt-dlp` if it’s on `PATH` (or `YT_DLP_PATH` is set) and a Whisper provider is available.
- `--version` now includes a short git SHA when available (build provenance).
- `--extract` now defaults to Markdown output (when `--format` is omitted), preferring Readability input.
- `--extract` no longer spends LLM tokens for Markdown conversion by default (unless `--markdown-mode llm` is used).
- `--format md` no longer forces Firecrawl; use `--firecrawl always` to force it.
- Finish line in `--extract` shows the extraction path (e.g. `markdown via readability`) and omits noisy `via html` output.
- Finish line always includes the model id when an LLM is used (including `--extract --markdown-mode llm`).
- `--extract` renders Markdown in TTY output (same renderer as summaries) when `--render auto|md` (use `--render plain` for raw Markdown).
- Suppress transcript progress/failure messages for non-YouTube / non-podcast URLs.
- Streaming now works with auto-selected models (including `--model free`) when `--stream on|auto`.
- Warn when `--length` is explicitly provided with `--extract` (ignored; no summary is generated).

## 0.6.0 - 2025-12-25

### Features

- **Podcasts (full episodes)**
  - Support Apple Podcasts episode URLs via iTunes Lookup + enclosure transcription (avoids slow/blocked HTML).
  - Support Spotify episode URLs via the embed page (`/embed/episode/...`) to avoid recaptcha; fall back to iTunes RSS when embed audio is DRM/missing.
  - Prefer local `whisper.cpp` when installed + model available (no API keys required for transcription).
  - Whisper transcription works for any media URL (audio/video containers), not just YouTube.
- **Language**
  - Add `--language/--lang` (default: `auto`, match source language).
  - Add config support via `output.language` (legacy `language` still supported).
- **Progress UI**
  - Add two-phase progress for podcasts: media download + Whisper transcription progress.
  - Show transcript phases (YouTube caption/Apify/yt-dlp), provider + model, and media size/duration.

### Changes

- **Transcription**
  - Add lenient ffmpeg transcode fallback for local Whisper when strict decode fails (e.g. Spotify AAC).

- **Models**
  - Add `zai/...` model alias with Z.AI base URL + chat completions by default.
  - Add `OPENAI_USE_CHAT_COMPLETIONS` + `openai.useChatCompletions` config toggle.
- **Metrics / output**
  - `--metrics on|detailed`: finish line includes compact transcript stats (… words, …) + media duration (when available); `--metrics detailed`: also prints input/transcript sizes + transcript source/provider/cache; hides `calls=1`.
  - Smarter duration formatting (`1h 13m 4s`, `44s`) and rounded transfer rates.
  - Make Markdown links terminal-clickable by materializing URLs.
  - `--metrics on|detailed` renders a single finish line with a compact transcript block (… words, …) before the model.
- **Cost**
  - Include OpenAI Whisper transcription estimate (duration-based) in the finish line total (`txcost=…`); configurable via `openai.whisperUsdPerMinute`.

### Docs

- Add `docs/language.md` and document language config + flag usage.

### Tests

- Add JSON-LD graph extraction coverage.
- Extend live podcast-host coverage (Podchaser, Spreaker, Buzzsprout).
- Raise global branch coverage threshold to 75% and add regression coverage for podcast/language/progress paths.

## 0.5.0 - 2025-12-24

### Features

- **Model selection & presets**
  - Automatic model selection (`--model auto`, now the default):
    - Chooses models based on input kind (website/YouTube/file/image/video/text) and prompt size.
    - Skips candidates without API keys; retries next model on request errors.
    - Adds OpenRouter fallback attempts when `OPENROUTER_API_KEY` is present.
    - Shows the chosen model in the progress UI.
  - Named model presets via config (`~/.summarize/config.json` → `models`), selectable as `--model <preset>`.
  - Built-in preset: `--model free` (OpenRouter `:free` candidates; override via `models.free`).
- **OpenRouter free preset maintenance**
  - `summarize refresh-free` regenerates `models.free` by scanning OpenRouter `:free` models and testing availability + latency.
  - `summarize refresh-free --set-default` also sets `"model": "free"` in `~/.summarize/config.json` (so free becomes your default).
- **CLI models**
  - Add `--cli <provider>` flag (equivalent to `--model cli/<provider>`).
  - `--cli` accepts case-insensitive providers and can be used without a provider to enable CLI auto selection.
- **Content extraction**
  - Website extraction detects video-only pages:
    - YouTube embeds switch to transcript extraction automatically.
    - Direct video URLs can be downloaded + summarized when `--video-mode auto|understand` and a Gemini key is available.
- **Env**
  - `.env` in the current directory is loaded automatically (so API keys work without exporting env vars).

### Changes

- **CLI config**
  - Auto mode uses CLI models only when `cli.enabled` is set; order follows the list.
  - `cli.enabled` is an allowlist for CLI usage.
- **OpenRouter**
  - Stop sending extra routing headers.
  - `--model free`: when OpenRouter rejects routing with “No allowed providers”, print the exact provider names to allow and suggest running `summarize refresh-free`.
  - `--max-output-tokens`: when explicitly set, it is also forwarded to OpenRouter calls.
- **Refresh Free**
  - Default extra runs reduced to 2 (total runs = 1 + runs) to reduce rate-limit pressure.
  - Filter `:free` candidates by recency (default: last 180 days; configurable via `--max-age-days`).
  - Print `ctx`/`out` in `k` units for readability.
- **Defaults**
  - Default summary length is now `xl`.

### Fixes

- **LLM / OpenRouter**
  - LLM request retries (`--retries`) and clearer timeout errors.
  - `summarize refresh-free`: detect OpenRouter free-model rate limits and back off + retry.
- **Streaming**
  - Normalize + de-dupe overlapping chunks to prevent repeated sections in live Markdown output.
- **YouTube**
  - Prefer manual captions over auto-generated when both exist. Thanks @dougvk.
  - Always summarize YouTube transcripts in auto mode (instead of printing the transcript).
- **Prompting & metrics**
  - Don’t “pad” beyond input length when asking for longer summaries.
  - `--metrics detailed`: fold metrics into finish line and make labels less cryptic.

### Docs

- Add documentation for presets and Refresh Free.
- Add a “make free the default” quick start for `summarize refresh-free --set-default`.
- Add a manual end-to-end checklist (`docs/manual-tests.md`).
- Add a quick CLI smoke checklist (`docs/smoketest.md`).
- Document CLI ordering and model selection behavior.

### Tests

- Add coverage for presets and Refresh Free regeneration.
- Add live coverage for the `free` preset.
- Add regression coverage for YouTube transcript handling and metrics formatting.

## 0.4.0 - 2025-12-21

### Changes

- Add URL extraction mode via `--extract` with `--format md|text`.
- Rename HTML→Markdown conversion flag to `--markdown-mode`.
- Add `--preprocess off|auto|always` and a `uvx markitdown` fallback for Markdown extraction and unsupported file attachments (when `--format md` is used).

## 0.3.0 - 2025-12-20

### Changes

- Add yt-dlp audio transcription fallback for YouTube; prefer OpenAI Whisper with FAL fallback. Thanks @dougvk.
- Add `--no-playlist` to yt-dlp downloads to avoid transcript mismatches.
- Run yt-dlp after web + Apify in `--youtube auto`, and error early for missing keys in `--youtube yt-dlp`.
- Require Node 22+.
- Respect `OPENAI_BASE_URL` when set, even with OpenRouter keys.
- Add OpenRouter configuration tests. Thanks @dougvk for the initial OpenRouter support.
- Build and ship a Bun bytecode arm64 binary for Homebrew.

### Tests

- Add coverage for yt-dlp ordering, missing-key errors, and helper paths.
- Add live coverage for yt-dlp transcript mode and missing-caption YouTube pages.

### Dev

- Add `Dockerfile.test` for containerized yt-dlp testing.

## 0.2.0 - 2025-12-20

### Changes

- Add native OpenRouter support via `OPENROUTER_API_KEY`.
- Remove map-reduce summarization; reject inputs that exceed the model's context window.
- Preflight text prompts with the GPT tokenizer and the model’s max input tokens.
- Reject text files over 10 MB before tokenization.
- Reject too-small numeric `--length` and `--max-output-tokens` values.
- Cap summaries to the extracted content length when a requested size is larger.
- Skip summarization for tweets when extracted content is already below the requested length.
- Use bird CLI for tweet extraction when available and surface it in the status line.
- Fall back to Nitter for tweet extraction when bird fails; report a clear error when tweet data is unavailable.
- Compute cost totals via tokentally’s tally helpers.
- Improve fetch spinner with elapsed time and throughput updates.
- Show Firecrawl fallback status and reason when scraping kicks in.
- Enforce a hard deadline for stalled streaming LLM responses.
- Merge cumulative streaming chunks correctly and keep stream-merge for streaming output.
- Fall back to non-streaming when streaming requests time out.
- Preserve parentheses in URL paths when resolving inputs.
- Stop forcing Firecrawl for --extract-only; only use it as a fallback.
- Avoid Firecrawl fallback when block keywords only appear in scripts/styles.

### Tests

- Add CLI + live coverage for prompt length capping.
- Add coverage for cumulative stream merge handling.
- Add coverage for streaming timeout fallback.
- Add live coverage for Wikipedia URLs with parentheses.
- Add coverage for tweet summaries that bypass the LLM when short.
- Add coverage for content budget paths and TOKENTALLY cache dir overrides.

### Docs

- Update release checklist to all-in-one flow.
- Fix release script quoting.
- Document input limits and minimum length/token values.

### Dev

- Add a tokenization benchmark script.

### Fixes

- Preserve balanced parentheses/brackets in URL paths (e.g. Wikipedia titles).
- Avoid Firecrawl fallback when block keywords only appear in scripts/styles.
- Add a Bird install tip when Twitter/X fetch fails without bird installed.
- Graceful error when tweet extraction fails after bird + Nitter fallback.

## 0.1.2 - 2025-12-20

### Fixes

- Release tooling: repair script quoting (no user-visible changes).

## 0.1.1 - 2025-12-19

### Fixes

- Accept common “pasted URL” patterns like `url (canonical)` and clean up accidental `\\?` / `\\=` / `%5C` before query separators.

## 0.1.0 - 2025-12-19

First public release.

### CLI

- `summarize` CLI shipped via `@steipete/summarize` (plus optional library exports).
- Inputs: URL, local file path, or remote file URL (PDFs/images/audio/video/text).
- Automatic map-reduce for large inputs.
- Streaming output by default on TTY, with Markdown → ANSI rendering (via `markdansi`).
- Final “Finished in …” line: timing, token usage, cost estimate (when pricing is available), and service counts.
- Flags:
  - `--model <provider/model>` (default `google/gemini-3-flash-preview`)
  - `--length short|medium|long|xl|xxl|<chars>` (guideline; no hard truncation)
  - `--max-output-tokens <count>` (optional hard cap)
  - `--timeout <duration>` (default `2m`)
  - `--stream auto|on|off`, `--render auto|md|plain`
  - `--extract` (URLs only; no summary)
  - `--json` (structured output incl. input config, prompt, extracted content, LLM metadata, and metrics)
  - `--metrics off|on|detailed` (default `on`)
  - `--verbose`

### Sources

- Websites: fetch + extract “article-ish” content + normalization for prompts.
- Firecrawl fallback for blocked/thin sites (`--firecrawl off|auto|always`, via `FIRECRAWL_API_KEY`).
- Markdown extraction for websites in `--extract` mode (`--format md|text`, `--markdown-mode off|auto|llm`).
- YouTube (`--youtube auto|web|apify`):
  - best-effort transcript endpoints
  - optional Apify fallback (requires `APIFY_API_TOKEN`; single actor `faVsWy9VTSNVIhWpR`)
- Files (remote or local): MIME sniffing + best-effort forwarding to the model.
  - text-like inputs are inlined for provider compatibility

### LLM providers

- Direct-provider API keys (no gateway).
- OpenAI-compatible base URL support (`OPENAI_BASE_URL`, `OPENROUTER_API_KEY`).
- Model ids: `openai/...`, `anthropic/...`, `xai/...`, `google/...`.
- Auto-handling of provider/model limitations (e.g. no streaming support → non-streaming call; unsupported media types → friendly error).

### Pricing + limits

- Token/cost estimates and model limits derived from LiteLLM’s model catalog, downloaded + cached under `~/.summarize/cache/`.

### Quality

- CI: lint, tests (coverage), and pack.
- Tooling: Biome (lint/format) + Vitest (tests + coverage gate).