Getting started

Anoka.ai is a local-first desktop app that records your meetings (microphone + system audio), transcribes them with Whisper, and uses AI to turn the transcript into kanban tasks — grouped by project, with priorities, due dates, comments and one-click push to ClickUp, Jira and GitHub. Your data stays on your machine.

Quick start

1. Download the desktop app and open it (see Install).
2. In Settings → AI, set an AI provider (cloud with your own key, or 100% local).
3. Go to Record: record a call (or paste a transcript), Transcribe, then Extract tasks.
4. Review the proposed tasks, pick a project, and add them to your Board.

Install

macOS (Apple Silicon)

1. Download the .zip from the home page and double-click to unzip.
2. Move Anoka.app to your Applications folder (optional).
3. The app isn't notarized yet, so macOS blocks it on first open. Remove the quarantine flag once:

xattr -dr com.apple.quarantine ~/Downloads/Anoka.app

Then open it normally. (Alternatively: System Settings → Privacy & Security → Open Anyway.)

Windows / Linux

Coming soon. In the meantime you can run from source:

git clone <repo> && cd anoka-ai
npm install
npm start

AI providers & models

Open Settings → AI. Choose One provider for everything, or Advanced to mix providers per area (e.g. Whisper to transcribe + Claude to extract). Bring your own API keys, or run everything locally.

100% local stack

Extraction with Ollama, transcription with Speaches (OpenAI-compatible):

# Extraction
ollama serve
ollama pull qwen2.5:7b

# Transcription
docker run -d --name anoka-speaches --restart unless-stopped \
  -p 8000:8000 ghcr.io/speaches-ai/speaches:latest-cpu
curl -X POST http://localhost:8000/v1/models/Systran/faster-whisper-small

Note: Claude (Anthropic) has no audio transcription. For meetings, use Whisper (OpenAI/Groq/Speaches) to transcribe and Claude to extract — that's what Advanced mode is for.

Connect the MCP server

Anoka ships an MCP server that reads your local tasks, so an AI assistant (Claude Desktop, Claude Code…) can view and create tasks, leave comments and read transcripts.

Claude Code

claude mcp add anoka -- node /path/to/anoka-ai/mcp/server.mjs

Claude Desktop

Add this to claude_desktop_config.json and restart:

{
  "mcpServers": {
    "anoka": {
      "command": "node",
      "args": ["/path/to/anoka-ai/mcp/server.mjs"]
    }
  }
}

Available tools

• list_projects — projects with task counts and context
• list_tasks — filter by project / status (incl. priority & due date)
• tasks_by_project — overview grouped by status
• create_task — title, description, project, status, priority, dueDate
• add_comment — leave a comment on a task
• list_transcripts — saved transcript history

If you change the MCP (e.g. new fields), reconnect it so the client picks up the schema: /mcp → reconnect, or restart the client. New tasks appear in the app after a reload (Cmd+R).

Connectors

From a task's drawer you can push it to ClickUp, Jira or GitHub Issues. Enable and configure them in Settings → Connectors; on create, the task opens in your browser.

Troubleshooting

"Anoka can't be opened" / "Apple could not verify…"

The app isn't notarized. Remove the quarantine flag and reopen:

xattr -dr com.apple.quarantine /Applications/Anoka.app

"No system audio captured"

macOS needs the Screen Recording permission to capture system audio. Go to System Settings → Privacy & Security → Screen Recording, enable the app (or Electron in dev), and restart it. The microphone works without this.

Transcription error: 404 or fetch failed

• fetch failed → the transcription server is down. If you use Speaches, make sure the container is running: docker ps / docker start anoka-speaches.
• 404 ... model not installed → download the model into Speaches: curl -X POST http://localhost:8000/v1/models/Systran/faster-whisper-small.

Extraction error: model not found, try pulling it first

The model name must match exactly what ollama list shows (e.g. qwen2.5:7b, not qwen-2.5-32b). Pull it with ollama pull qwen2.5:7b and pick it in Settings.

Tasks created via MCP don't show in the app

The app reads its data on launch. Reload with Cmd+R (or reopen) to see tasks created externally.

Large recordings transcribe slowly

Local Whisper on CPU is slow for long audio (20+ min can take several minutes — it's not frozen). Use a smaller model (faster-whisper-tiny) or a cloud provider (OpenAI/Groq) for speed.