minutes-sdk
Conversation memory SDK — query meeting transcripts, decisions, and action items from any AI agent or application
Description
Conversation memory SDK — query meeting transcripts, decisions, and action items from any AI agent or application
Release History
| Version | Changes | Urgency | Date |
|---|---|---|---|
| 0.13.3 | Imported from npm (0.13.3) | Low | 4/21/2026 |
| v0.13.3 | # v0.13.3: Meeting-reliability fixes from the field 0.13.3 is a patch release focused on two bugs that showed up in real calls, plus a handful of platform and tooling fixes that make the rest of the stack easier to trust. If you're on 0.13.0 or later you can upgrade in place. If you somehow held at 0.12.x, read [v0.13.0](https://github.com/silverstein/minutes/releases/tag/v0.13.0) and [v0.13.2](https://github.com/silverstein/minutes/releases/tag/v0.13.2) first; those shipped the settings, iden | High | 4/21/2026 |
| v0.13.2 | 0.13.2 adds the mic-mute toggle for passive attendance and rolls up everything that shipped in the 0.13 series this week. If you installed Minutes before 0.13.0 landed two days ago, this is the single version to move to. If you updated to 0.13.0 or 0.13.1 already, you only get the mic-mute change on top of what you have. ## What's new in 0.13.2: mic-mute You're on a webinar, an all-hands, or a panel where you're not going to speak. You want the meeting transcript, but you don't want your breat | High | 4/18/2026 |
| v0.13.1 | # v0.13.1: Fix Claude Desktop extension install on 1.3109.0 ## What changed This is a patch release that fixes a broken install surface in the Claude Desktop extension, reported in [#149](https://github.com/silverstein/minutes/issues/149). The v0.13.0 `minutes.mcpb` bundle contained two Next.js chunks whose filenames literally include `..`: ``` .vercel/output/static/_next/static/chunks/0c1n75d4d01s..js .vercel/output/static/_next/static/chunks/0yhyx4ni~s7z..js ``` Claude Desktop 1.3109.0 sh | High | 4/18/2026 |
| v0.13.0 | # v0.13.0: Settings you can trust, smarter call capture, and better operator tooling ## What changed Minutes 0.13.0 is a “come back and try it again” release. The biggest theme is trust: - Settings behavior is much tighter now. Important toggle rows no longer drift out of sync with what you clicked, and the command palette shortcut finally has a real UI row instead of a half-wired path. - Identity setup is stronger. You can add your name, emails, and name variants so Minutes recognizes you m | High | 4/17/2026 |
| v0.12.2 | # v0.12.2: Live transcription that actually works on quiet audio ## Why this release exists A meeting today produced 11 fragments over 40 minutes of live transcript — mostly whisper placeholder tokens like `[typing]` and `[BLANK_AUDIO]`. Meanwhile the post-recording batch transcript cleanly recovered 2,259 words from the same WAV. Same audio, two different code paths, 200x quality gap. Three problems were working against live mode: 1. The recording sidecar used a simple energy-threshold VAD. | High | 4/15/2026 |
| v0.12.1 | # v0.12.1: Frontmatter polish, Parakeet hardening, and LLM titles ## What changed This release is a focused polish pass over the work that landed in v0.12.0. The transcription pipeline was doing a lot — transcribe, diarize, summarize, extract entities, match calendar events, write markdown — and the seams between those steps were leaking. Speaker labels from diarization were ending up in attendee lists, email addresses from calendars were spawning duplicate people, and "Untitled Recording" was | High | 4/15/2026 |
| v0.12.0 | # v0.12.0: Parakeet is multilingual by default ## What changed This release makes the Parakeet path feel like a first-class part of Minutes instead of an advanced toggle you had to babysit. The biggest change is the default model. If you choose Parakeet and do not pin a model explicitly, Minutes now defaults to `tdt-600m`, the multilingual v3 model, instead of the older English-only `tdt-ctc-110m`. That means better support for bilingual meetings, better WER, and a much more honest default fo | Medium | 4/14/2026 |
| v0.11.3 | # v0.11.3: Long meetings work the way they should ## What changed This release fixes a nasty class of failures in the Parakeet transcription path. Long recordings could fail in a few different ways. Some chunks came back with zero tokens. Some long runs could blow up in the GPU path. The app often surfaced all of that as a vague parsing error. That is fixed now. Long Parakeet runs are chunked before decode, empty chunks are skipped instead of taking down the whole meeting, and the exact real- | Medium | 4/13/2026 |
| v0.11.2 | Your AI used to forget what happened after `minutes stop`. Now it doesn't. ## The lifecycle is connected Before v0.11.2, the meeting lifecycle had a gap right in the middle. You'd record a meeting, run `minutes stop`, and... nothing. The tool that knows everything about your conversation history went quiet at the exact moment you needed it most -- right after the meeting, when decisions are fresh and follow-ups haven't slipped yet. Now `minutes stop` nudges you toward the next valuable step. | Medium | 4/10/2026 |
| v0.11.1 | Artifacts, a design system, and GPU backends. ## What's new **Artifacts** -- Create, edit, preview, and revert documents directly from Recall conversations. The artifact picker lets you choose what to create (relationship brief, deal memo, meeting summary), then opens it in a split view beside the AI assistant. Source editing with inline save, snapshot-based revert, and a review panel showing before/after diffs. **Unified design system** -- Light cream (#F8F4ED) default with warm-dark mode. I | Medium | 4/9/2026 |
| plugin-v0.8.0 | # Minutes Plugin v0.8.0: Four new skills, a proactive brief hook, and bug fixes two review passes deep ## The lifecycle is done. Brief runs before your call, mirror tells you what you did, tag marks the outcome, graph lets you query everyone and everything across your history. All four ship with Python helper scripts that do the counting deterministically so an LLM doesn't have to guess. Before v0.8.0 the plugin covered prep, record, debrief, weekly. After v0.8.0 it covers the full arc: `brief | Medium | 4/9/2026 |
| v0.11.0 | # Minutes v0.11.0: Command palette, faster whisper, better call capture ## The keyboard-first launcher ships. Whisper uses your GPU now (it wasn't before). Two contributors fixed bugs you'll actually notice. v0.11.0 ships the command palette as the headline feature, plus a round of fixes that change what Minutes feels like in daily use. The capture cues sound like real objects instead of compiler beeps. Native Meet calls stop labeling you as SPEAKER_00. The screen share privacy toggle, which h | Medium | 4/8/2026 |
| v0.10.4 | ## Minutes was telling you the wrong thing v0.10.4 fixes four bugs where Minutes confidently reported something that wasn't true. Nothing crashed; you just couldn't trust the answers. ### Your knowledge base was enabled. Except when it wasn't. The MCP `knowledge_status` tool was doing a naive string match for `enabled = true` anywhere in `config.toml`. So if you had `[recording] enabled = true` and `[knowledge] enabled = false`, the tool would tell Claude (or Cursor, or Codex) that your knowl | Medium | 4/7/2026 |
| v0.10.3 | ## What changed Three platform fixes that unblock real users on Linux, macOS, and Windows. None of these were nice-to-haves. All three were silently breaking the core capture-and-summarize loop on a specific OS. ### Linux: PipeWire system audio capture (#85) Linux users running `minutes sources` were seeing their speakers and headphones bucketed under "Microphones" with no clue they could record system audio from them. The bug was in our device categorization, not in cpal. cpal's PipeWire bac | Medium | 4/7/2026 |
| v0.10.2 | Your AI assistant remembers every conversation you've had. Here's what's new. ## Try it without a microphone `minutes demo --full` seeds 5 interconnected sample meetings (Snow Crash themed), then `minutes demo --query` shows cross-meeting intelligence in action. Da5id promised to fix the rendering pipeline 14 days ago. He didn't. Minutes caught it across three separate conversations. No setup, no mic, no model download. Just `cargo install minutes-cli && minutes demo --full`. ## Knowledge ba | Medium | 4/7/2026 |
| v0.10.1 | Patch release: fixes agent summarization, calendar timeouts, and ships the first building block of conversation-to-knowledge-base integration. ## Bug fixes **Agent summarization was broken on v0.10.0.** The `--no-input` flag was passed to the Claude CLI, which doesn't accept it. Every `[summarization] engine = "agent"` call failed silently — meetings were transcribed but never summarized. Fixed. **Calendar overlap detection timed out with many calendars.** AppleScript couldn't enumerate 29 ca | Medium | 4/6/2026 |
| v0.10.0 | Minutes now auto-updates from GitHub Releases and ships a major diarization overhaul from three community contributors. ## Auto-update The desktop app checks for updates on launch and every 6 hours. When a new version is available, a non-modal banner appears: "Restart Now" or "Later." Updates are never installed during a recording, live transcript, or dictation session. Signed with Ed25519. The release workflow produces signed artifacts and a `latest.json` manifest. Zero infrastructure beyond | Medium | 4/6/2026 |
| v0.9.4 | ## What's new ### Fix: MCP recordings no longer capture silence Recordings started via Claude Desktop's `start_recording` tool were producing silent WAV files on recent macOS versions. The CLI worked fine from Terminal -- same binary, same mic permissions, same machine. **Root cause:** The MCP server spawned `minutes record` with `detached: true`, which calls `setsid()` and creates a new macOS audit session. This severed the child process from Claude Desktop's TCC microphone grant. CoreAudio | Medium | 4/2/2026 |
| v0.9.3 | ## What's new ### Multi-source capture Record system audio and microphone simultaneously as separate stems. Each source gets its own WAV file, and speaker attribution uses per-source energy levels instead of relying solely on diarization. The shared resampler (32-tap Hann windowed-sinc) is now factored out of the capture path so both live streaming and file processing use identical downsampling. This is the foundation for call recording where you want to distinguish "what they said" from "wha | Medium | 4/1/2026 |
| v0.9.2 | ## What's new ### Brand-derived design system The entire desktop app has a new color palette pulled from the brand mark's sky-blue and amber identity. Every surface -- main window, dictation overlay, note popup, meeting prompt, terminal, and the MCP dashboard inside Claude Desktop -- now uses the same cohesive palette instead of scattered defaults. - Sky blue accent (`#4da8d9`) derived from the icon's upper glow - Warm amber section headers and warning states from the icon's lower glow - Do | Medium | 4/1/2026 |
| v0.9.1 | Back-to-back meetings just got smoother in the desktop app. **What's new:** - The record button now stays visible while background processing runs — you can start your next meeting without switching to the terminal (#38) - Filter diagnostics for blank transcript debugging - Default quickstart model changed from `tiny` to `small` for better out-of-box accuracy **Fixes:** - v0.9.0 release builds were broken by a missing struct field in CI — this release restores DMG, Windows EXE, and Linux binar | Medium | 3/31/2026 |
| v0.9.0 | Minutes now handles Bluetooth headsets, USB mics, and audio device switching mid-recording — no more silent recordings when you grab your AirPods and walk away from the desk. ## What's new **Audio device selection** — pick which mic to record from ```bash minutes record --device "AirPods Pro" # Use a specific device minutes devices # List available devices with sample rates ``` Or set it permanently in `~/.config/minutes/config.toml`: ```toml [recording] device | Medium | 3/31/2026 |
| v0.8.5 | Meetings now emit structured insight events that agents and external systems can subscribe to. Every decision, commitment, and open question extracted from a meeting becomes a typed event in the event log with a confidence level — so downstream systems know when to act and when to wait. Only strong and explicit insights are considered actionable, preventing noise from tentative language like "we should consider" from triggering false workflows. ## What's new **Meeting Insights** — structured | Medium | 3/31/2026 |
| v0.8.4 | # Release Notes for 24dfbcf9be5f3807cd3ef4664aada81fee25ac12 - Channel: stable - Range: v0.8.3..24dfbcf9be5f3807cd3ef4664aada81fee25ac12 - Stable release intended for broad usage. ## What changed ### Shared engine - fix: add version to whisper-guard dep for crates.io publish (`6c21803`) - fix: graceful Ctrl+C shutdown for live, dictate, and record (#29) (#30) (`9453b19`) - Accumulate dictation output across pauses (`cbd2b76`) ### Desktop - Accumulate dictation output across pauses (`cbd2b76` | Medium | 3/29/2026 |
| v0.8.3 | # Release Notes for 611d6a9df8c97736722d494380ef075fdaea3ce5 - Channel: stable - Range: v0.8.2..611d6a9df8c97736722d494380ef075fdaea3ce5 - Stable release intended for broad usage. ## What changed ### Shared engine - security: harden pipeline against injection, traversal, and data exposure (`6dc42e0`) ### CLI - fix: pin ort to rc.10 for Windows diarization build + document LLVM prereqs (`1e37dee`) ### Distribution and release policy - fix: pin ort to rc.10 for Windows diarization build + doc | Medium | 3/29/2026 |
| v0.8.2 | # Release Notes for d06d9e721f2afdd4fd073f3df9565e2187565846 - Channel: stable - Range: v0.8.0..d06d9e721f2afdd4fd073f3df9565e2187565846 - Stable release intended for broad usage. ## What changed ### Shared engine - feat: add Parakeet ASR as opt-in transcription engine (`6322ee4`) - style: apply cargo fmt to all workspace files (`2f39a4a`) - Extract whisper-guard crate + add nnnoiseless noise reduction (`d64d53d`) - whisper-guard: clean API for publish — borrow instead of own (`12eae36`) - fi | Medium | 3/28/2026 |
| v0.8.1-live-coach | ## Live Meeting Coach Real-time meeting transcription with AI coaching. Your AI assistant reads the transcript mid-meeting and coaches you on what to say. ### How it works 1. Click **Live** in the Minutes menu bar (or press **Cmd+Shift+L**) 2. Green pulsing indicator shows it's listening 3. Open Claude Desktop / Claude Code / Codex / Gemini 4. Ask: "What should I say about the pricing objection they just raised?" 5. Your AI reads the last 5 minutes of transcript and coaches you 6. Click **Sto | Medium | 3/28/2026 |
| v0.8.0 | ## Relationship Intelligence Minutes now tracks **who you talk to**, **what you promised**, and **how your relationships evolve** across every meeting. Ask your AI *"What did I promise Sarah?"* — and get an answer. ### New features **Conversation Graph** — SQLite index derived from your meeting markdown. Tracks people, commitments, topics, and relationship health. Rebuilds in <50ms. ```bash minutes people # Relationship overview with scores minutes commitments --perso | Medium | 3/27/2026 |
| v0.7.3 | ## Dictation UX Overhaul The dictation overlay gets a complete UX upgrade — from a status-only indicator to a trust-building interface. ### What's new - **Instant feedback** — Overlay appears immediately on shortcut press with "Loading model..." state. No more 5-15s dead gap wondering if it worked. - **Model preloading** — Whisper model loads in a background thread at app startup, so the first dictation press goes straight to "Listening..." with zero delay. - **Text confirmation on success** | Medium | 3/25/2026 |
| v0.7.2 | ## Windows support + dictation polish Minutes now builds and runs on Windows. Plus: dictation hotkeys that survive app restarts. ### What shipped **Windows desktop app** — First-class Windows support via NSIS installer. Tray icon, recording, dictation, and the full pipeline work on Windows. Sentinel-based stop path handles the platform's file-locking quirks. **Dictation hotkey persistence** — Your custom dictation shortcut now survives app restarts. Previously it reverted to the default on e | Medium | 3/24/2026 |
| v0.7.1 | Patch release with streaming dictation, plugin marketplace, SDK improvements, and docs. ### Streaming Whisper (new) - **Text appears as you speak** — partial transcription every 2 seconds during dictation - Full re-transcription from t=0 each pass for consistent output - Whisper init noise suppressed after first partial - `minutes dictate` CLI shows progressive text with live line updates ### Distribution (new) - **Claude Code plugin marketplace**: `claude plugin marketplace add silverstein/mi | Medium | 3/24/2026 |
| v0.7.0 | ## Cross-Device Ghost Context Layer Record a thought on your phone → it becomes searchable AI memory on your desktop. Claude surfaces recent memos proactively. ### What's new - **Duration-based routing**: Short voice memos (<2 min) route through the fast memo pipeline. Long recordings get full meeting treatment. Threshold configurable. - **Phone → desktop pipeline**: Works with any folder sync — iCloud Drive, Dropbox, Google Drive, Syncthing, or manual file drop. - **Ghost context**: SessionS | Medium | 3/24/2026 |
| v0.6.0 | Your AI remembers every conversation you've had. Now it remembers every thought you speak. **v0.6.0 adds dictation** — hold a key, speak, and your words land in your clipboard AND your AI's memory. Every dictation tool forgets what you said. Minutes remembers it. "What did I dictate into Slack about the pricing page?" just works. ## What's new ### Dictation mode Speak naturally. Text goes to your clipboard after each pause. Every utterance is also logged to your daily note and saved as a sea | Medium | 3/24/2026 |
| v0.5.0 | ## Try minutes right now — no install needed ```json { "mcpServers": { "minutes": { "command": "npx", "args": ["minutes-mcp"] } } } ``` Drop that into your Claude Desktop config. Claude can now search your meeting history, find action items, build people profiles, and track decisions — all from your local `~/meetings/` folder. No Rust toolchain. No API key. No build step. --- ### What's new in v0.5.0 **`npx minutes-mcp` — zero-install MCP server** The MCP server no | Medium | 3/23/2026 |
| v0.4.0 | ## What's New **Cross-Platform Support** — minutes now builds and tests on macOS, Windows, and Linux: - CI runs on all 3 platforms (core + CLI tests pass everywhere) - Windows: WASAPI audio capture support, HANDLE-based PID management - Linux: ALSA audio capture support - `cargo install --path crates/cli` works on any platform **Vault Sync for Obsidian / Logseq** — connect your knowledge vault: - `minutes vault setup` — auto-detect vaults, recommend symlink/copy/direct strategy - `minutes vaul | Low | 3/22/2026 |
| v0.3.0 | ## What's New **Interactive MCP App Dashboard** — 5 tools now render an interactive UI inline in Claude Desktop via [MCP Apps](https://modelcontextprotocol.io/specification/2025-03-26/server/utilities/apps): - Meeting list with stats bar, type filters, and instant search - Detail view with fullscreen mode and "Send to Claude" context injection - Clickable attendee chips → person profile navigation - Live recording indicator with stop button - Person profiles and consistency reports **Smart Att | Low | 3/21/2026 |
| v0.2.0 | > **A newer version is available: [v0.3.0](https://github.com/silverstein/minutes/releases/tag/v0.3.0)** — Interactive dashboard, smart attendee detection, semantic search. ## What's New **Interactive Meeting Lifecycle** — 3 new skills that chain together: - `/minutes prep` — relationship brief + talking points before a call - `/minutes debrief` — post-meeting analysis, compares outcomes to prep intentions - `/minutes weekly` — weekly synthesis with decision arcs and forward brief **Settings | Low | 3/20/2026 |
Dependencies & License Audit
Loading dependencies...
Similar Packages
@senso-ai/shipablesCLI for installing, managing, and publishing AI agent skills from the Shipables registry0.1.2
automagik-genieSelf-evolving AI agent orchestration framework with Model Context Protocol support2.5.27
council-mcpThree-tier AI agent MCP orchestration system: Chancellor (Opus), Executor (Sonnet), Aide (Haiku)0.4.0