transcribe-it

A lightweight CLI for ingesting meeting transcripts (Gmail or Slack), enriching them with an LLM, and storing the results as local files.

Source -> Extract -> LLM Enrich -> Local Files

Prerequisites

• Python 3.12+
• A Google Cloud project with Gmail API + Google Drive API enabled (for the Gmail source), or a Slack bot token (for the Slack source)
• An API key for one of the supported LLM providers (Anthropic, OpenAI, or Groq) — only needed if you want LLM enrichment

Install

uv tool install transcribe-it

Or with pipx:

pipx install transcribe-it

Setup

Setup is split into two steps: a one-off global step for credentials, and a per-project step for what to ingest.

Step 1: Configure credentials (once per machine)

transcribe-it setup

Pick which credentials to set up — Google OAuth (for Gmail), Slack bot token, and/or LLM provider — and the values are written to ~/.config/transcript/env. Re-run any time to add or rotate values; existing values are preserved unless you confirm overwrite (or pass --force).

Step 2: Initialise a project (per directory)

From the directory where you want transcripts to land:

transcribe-it init

This asks which sources to enable, source-specific config (sender filter, channel ID, etc.), output path, and lookback window. Writes .transcripts/config.yaml. No secrets prompts — it'll warn if the credentials a chosen source needs aren't set yet.

Gmail credentials

setup asks for GOOGLE_OAUTH_CLIENT_ID and GOOGLE_OAUTH_CLIENT_SECRET. Two options:

1. Reuse someone else's OAuth client — ask a teammate for the values and have them add your Google account as a Test user on their OAuth consent screen.
2. Create your own — in Google Cloud Console, create an OAuth 2.0 Client ID of type Desktop app, then copy the client ID and secret from the resulting credentials.

After setup and init, authenticate:

transcribe-it auth gmail

Slack credentials

setup asks for SLACK_BOT_TOKEN (xoxb-...). The bot needs to be a member of the channels you want to ingest from. The channel ID itself is configured per-project in init.

Usage

By default, ingestion only extracts the raw transcript — no LLM call, no API key required. Pass --enrich to also generate a summary, topics, and participants via LLM.

# Last N days, raw extraction only (default)
transcribe-it ingest gmail --days 7

# With LLM enrichment
transcribe-it ingest gmail --days 7 --enrich

# Enrichment + cleaned transcript variant (--clean implies --enrich)
transcribe-it ingest gmail --days 7 --clean

# Specific date range
transcribe-it ingest gmail --from 2026-04-01 --to 2026-04-05

# Preview matching emails without fetching or writing
transcribe-it ingest gmail --days 1 --dry-run

# Ingest a single transcript file directly
transcribe-it ingest file path/to/transcript.txt

Output

Raw mode (default) writes a single .txt file per transcript:

transcripts/
  2026-04-09-ai-labs-daily.txt

With --enrich, each transcript becomes a folder:

transcripts/
  2026-04-09-ai-labs-daily/
    raw.txt          # Original transcript (immutable)
    metadata.json    # Source, date, participants, topics, summary

With --clean, an additional clean.md is written (structured: title, summary, topics, cleaned transcript).

Prompts

LLM prompts are bundled with the package under transcribe_it/prompts/. To customise, fork the repo and edit prompts/enrich.md.

Commands

Ingest options (Gmail)

Configuration files