Baez

Sync Granola meeting transcripts into an Obsidian vault as enriched, read-only markdown — with optional AI summaries powered by Claude.

What it does

• Fetches all your Granola transcripts via the API
• Writes Obsidian-flavored markdown with [[wiki-links]], #tags, and Dataview-compatible frontmatter
• Organizes files by date: Vault/Granola/2025/01/2025-01-15_standup.md
• Preserves raw JSON API responses for archival
• Incremental sync — only fetches documents that changed since last run
• Generates AI meeting summaries using Claude (Anthropic API), with Obsidian-native formatting
• Extracts a knowledge graph — People, Concepts, and Projects notes with [[wiki-links]] across meetings
• Resumable summarization — picks up where it left off if interrupted

Install

cargo install --path .

API Keys

Baez needs two separate API keys: one for Granola (to fetch transcripts) and one for Anthropic (to generate AI summaries). They are configured independently.

Granola API Key

The Granola token is used to authenticate with the Granola API and fetch your meeting transcripts. Baez resolves it in this order:

1. --token flag (highest priority)

   baez sync --token <granola-session-token>

2. BAEZ_GRANOLA_TOKEN environment variable

   export BAEZ_GRANOLA_TOKEN=<granola-session-token>
   baez sync

3. BEARER_TOKEN environment variable (deprecated, use BAEZ_GRANOLA_TOKEN)

4. Config file (~/.config/baez/config.json → granola_token field)

5. Granola's local session file (automatic, no setup needed) If you're logged into the Granola desktop app on macOS, baez reads the token directly from ~/Library/Application Support/Granola/supabase.json. This is the default — if you use the Granola app, you don't need to configure anything.

Anthropic API Key (for AI Summaries)

The Anthropic API key is used to generate meeting summaries via Claude. This is optional — sync works without it, you'll just get transcripts without summaries.

1. BAEZ_ANTHROPIC_API_KEY environment variable (highest priority)

   export BAEZ_ANTHROPIC_API_KEY=sk-ant-...
   baez sync

2. ANTHROPIC_API_KEY environment variable (also accepted, standard Anthropic convention)

   export ANTHROPIC_API_KEY=sk-ant-...
   baez sync

3. Config file (~/.config/baez/config.json → anthropic_api_key field)

4. macOS Keychain (persistent, recommended on macOS)

   baez set-api-key sk-ant-...

   This stores the key in the macOS system keychain under the service baez with the account name anthropic_api_key. The key persists across terminal sessions and reboots.

If no Anthropic API key is found, baez sync prints a warning and continues without summarization. You can also explicitly skip summaries with --no-summarize.

Usage

Set up your vault

# Option 1: Pass vault path directly
baez sync --vault /path/to/your/obsidian-vault

# Option 2: Set environment variable
export BAEZ_VAULT=/path/to/your/obsidian-vault
baez sync

# Option 3: Save in config file (supports all settings)
mkdir -p ~/.config/baez
echo '{"vault": "/path/to/your/obsidian-vault"}' > ~/.config/baez/config.json
baez sync

Config file

All settings can be stored in ~/.config/baez/config.json:

{
  "vault": "/path/to/your/obsidian-vault",
  "granola_token": "eyJ...",
  "anthropic_api_key": "sk-ant-..."
}

All fields are optional. CLI flags and environment variables take precedence over config file values.

Security note: If you store API keys in the config file, restrict its permissions so only your user can read it:

chmod 600 ~/.config/baez/config.json

Baez will print a warning at startup if the config file contains keys and is readable by other users. On macOS, the keychain (baez set-api-key) is the most secure option for the Anthropic API key.

Commands

baez sync                  Sync all documents (with AI summaries if key is set)
baez sync --force          Force re-sync, ignoring cache
baez sync --no-summarize   Sync without AI summaries
baez sync --dry-run        Preview what would be synced
baez summarize-all         Batch-summarize all unsummarized docs (no API fetch needed)
baez summarize-all --force Re-summarize everything
baez summarize-all --dry-run Preview what would be summarized
baez list                  List all documents
baez fetch <id>            Fetch a specific document
baez open                  Open vault directory
baez fix-dates             Fix file modification dates

AI Summaries

Summaries are generated during baez sync when an Anthropic API key is available. If sync is interrupted, re-running baez sync or baez summarize-all picks up where it left off — already-summarized docs are tracked in a separate summary cache and skipped automatically.

baez summarize <doc-id>           Print summary to stdout
baez summarize <doc-id> --save    Update the ## Summary section in the markdown file
baez summarize-all                Batch-summarize all unsummarized docs from local files
baez summarize-all --force        Re-summarize everything (e.g. after changing model)

summarize-all reads from the raw JSON files already on disk — it never hits the Granola API. This makes it safe to run repeatedly and useful for catching up after enabling summarization on an existing vault.

Configure the model and input limits:

baez set-config --show                            Show current configuration
baez set-config --model claude-sonnet-4-20250514  Change model
baez set-config --context-window 300000           Set max input size (chars)
baez set-config --prompt-file my-prompt.txt       Use a custom prompt

The default model is claude-opus-4-6 with a 600,000 character input limit (~150K tokens) and 8,192 max output tokens.

Knowledge Graph

When summarization is enabled, baez extracts entities from each meeting and creates interconnected notes in your vault:

Vault/
├── People/              # One note per person mentioned across meetings
│   └── Alice Smith.md
├── Concepts/            # Reusable ideas and patterns
│   └── API Design.md
└── Projects/            # Projects discussed in meetings
    └── Project Atlas.md

Each entity note tracks which meetings it appeared in via ## Sources links. Meeting notes link back via related frontmatter fields. The result is a navigable knowledge graph in Obsidian's graph view — people, concepts, and projects connected through meetings.

Vault layout

Vault/
├── People/                       # Entity notes (auto-created)
├── Concepts/
├── Projects/
└── Granola/
    ├── 2025/
    │   ├── 01/
    │   │   ├── 2025-01-15_standup.md
    │   │   └── 2025-01-16_planning.md
    │   └── 02/
    │       └── ...
    └── .baez/
        ├── raw/                  # Raw JSON API responses
        ├── tmp/                  # Atomic write temp dir
        ├── summary_config.json   # Summarization settings
        ├── .sync_cache.json      # Incremental sync state
        └── .summary_cache.json   # Summarization progress tracking

Markdown format

Each document gets Dataview-compatible frontmatter:

doc_id: abc123
source: granola
date: "2025-01-15"
created: 2025-01-15T10:00:00Z
title: Sprint Planning
attendees: [Alice, Bob]
duration_minutes: 30
tags: [planning]
related:
  - "[[Alice Smith]]"
  - "[[API Design]]"
status: substantive
generator: baez

The body includes [[wiki-links]] for attendees and speakers, #granola and #meeting/* tags, an AI-generated ## Summary section (with Key Decisions, Action Items, Discussion Highlights, and Open Questions), user notes, and the full transcript.

License

MIT