Personalized email

Runs the personalized email cookbook in examples/cookbooks/personalized_email/. It remembers your Granola meeting notes and your Gmail inbox and sent mail in the cognee dataset personalized_email, then drafts a reply to the newest inbox email. Setup, the scripts and what each one does are in the cookbook's README.md.

Run every command from the repo root.

Steps

With neither Gmail nor Granola set up, the commands below run on a sample mailbox (the [setup] line says so) and draft a reply to a fictional email; only LLM_API_KEY is needed. Tell the user the reply is to sample data. Add --sample to use the sample even when their accounts are set up.

1. Run uv run python examples/cookbooks/personalized_email/personalized_email.py --check.
   • Exit 0: go to step 2.
   • Exit 2: each [setup] MISSING: line names one fix. Don't create credentials yourself. Tell the user exactly which line to fix, using the README's "What it needs" table, and stop. If only the Granola line is missing, ask whether to run with --no-granola instead.
2. Run uv run python examples/cookbooks/personalized_email/personalized_email.py. Add --days N if the user named a time range for meetings, and --no-granola if they don't want Granola used.
3. The draft is everything after the line [draft] Reply:. Give it to the user, and say which email it answers (the [draft] Answering: line) and which sources went in (the [ingest_granola] and [ingest_email] lines). The draft is never sent: hand it to the user to send.
4. If a script fails, personalized_email.py exits 1 with a line naming what went wrong. Run that script from examples/cookbooks/personalized_email/scripts/ on its own to look closer.

Rules

• Gmail access is read-only and nothing is sent.
• Don't print the contents of credentials.json, token.json or .env.
• A sample run forgets the cookbook's dataset personalized_email before it ingests (the [clear] line). On the user's own data, pass --clear only when they asked to start over: it forgets everything the dataset holds.
• Ingesting reads the user's real meetings and mailbox, and every run uses LLM credits, so run it only when the user asked for it.

Clean up

uv run cognee-cli forget --dataset personalized_email