Quickstart¶
Requirements¶
- Python 3.11 or newer.
- SQLite with FTS5.
- One model backend: Codex (the default), Claude Code, Antigravity, or OpenRouter.
- macOS for the local Messages, WhatsApp, Calendar, and Reminders integrations.
Install¶
git clone https://github.com/cuuush/memcal.git
cd memcal
./install.sh
The installer puts a memcal launcher on your PATH that points at this checkout,
initializes ~/.memcal on the first run, imports available contacts, and writes the
first brief.
Chat sources need extra libraries (see each source page), installable together:
pip install "memcal[chat]"
signal-cli is not a pip package — it is a JVM program, installed separately
(see Signal).
Choose a model backend¶
memcal setup
The setup flow asks for the provider and model, then saves only the owned settings in
~/.memcal/.env. Existing source credentials and hand-written settings are preserved.
Check the result:
memcal doctor
memcal brief
First ingest¶
See which connectors are available, then pull from one source with a small limit before collecting everything:
memcal sources
memcal ingest slack --limit 20
memcal ingest all
Preview the size and estimated cost of an extraction pass, then run it:
memcal dream --dry-run
memcal dream
To collect and update automatically, install the nightly job:
memcal schedule install
Connect an agent¶
memcal openclaw setup
openclaw gateway restart
ln -s /path/to/memcal/integrations/hermes/memcal ~/.hermes/plugins/memcal
hermes memory setup
Any other harness: read brief.md and connect python3 -m memcal.mcp_server.
See Integrations.
Next steps¶
- Architecture — observe, gate, dream, brief.
- Remember, Recall, Correct — the tool surface.
- Sources — every input memcal reads.
- Evaluation — how memory quality is measured.