A proactive AI assistant that operates on a "Push" paradigm - anticipating needs and taking action rather than waiting for prompts.
Full Documentation | Architecture | MCP Setup | Ideas & Future | Discussions | Contributing
Modern AI assistants are passive - they wait for prompts. Reeve is different:
- Proactivity First: Reeve anticipates needs based on time, context, and history. It initiates conversations.
- Cognitive Offloading: If you have to remember to ask the assistant, the assistant has failed.
- The "Living" System: Reeve runs continuously, organizing your digital life while you sleep.
Instead of waiting for commands, Reeve operates on a Pulse - scheduled wake-ups that let it check on your world and take action. Learn more
Reeve acts as a high-fidelity proxy between you and the world:
- Input Filter: Reads the noise (group chats, emails) so you don't have to
- Output Delegate: Drafts replies, coordinates logistics, manages vendors
- Gatekeeper: Knows what deserves your attention vs. what can wait
Unlike "black box" agents with hidden state, Reeve's entire mind is visible in plain Markdown files. You can literally open its brain, read it, and edit it. Learn more
Your personal context lives in a separate Git repository (example Desk). This separation is a core design philosophy that enables data ownership and rollback. Learn more
Scenario: Weather agent detects 18" of powder at Mammoth.
Reeve's Action: Sends you a Telegram alert: "Powder Alert: 18 inches forecast for Mammoth this weekend. Shall I check if the Shred Crew is free?"
Outcome: Upon approval, messages the WhatsApp group, parses replies, summarizes headcount, offers to draft Airbnb booking. Zero mental load.
Also see:
- Deep Work Defender - Proactively blocks focus time, silences group chat noise, but breaks through for real emergencies
- Adaptive Coach - Detects burnout patterns and shifts from taskmaster to supporter
| Component | Purpose |
|---|---|
| Pulse Queue | SQLite-backed scheduler for proactive wake-ups |
| MCP Servers | Tools for Reeve to schedule pulses and send notifications |
| Telegram Integration | User messages trigger immediate high-priority pulses |
| The Desk | Separate Git repo of Goals/, Responsibilities/, Preferences/ |
Reeve wraps specialized CLIs (Claude Code, Hapi) rather than implementing its own agent loop - letting billion-dollar companies compete on that while we focus on orchestration and context hygiene.
# Clone and install
git clone https://github.com/reubenjohn/reeve-bot.git
cd reeve-bot
uv sync
# Run migrations
uv run alembic upgrade head
# Run tests
uv run pytest tests/ -v
# Start the daemon (requires configuration)
export PULSE_API_TOKEN=your-secret-token
uv run python -m reeve.pulseOpenClaw (157k+ stars) is an excellent all-in-one runtime. Reeve takes a different architectural bet:
| OpenClaw | Reeve | |
|---|---|---|
| Paradigm | Custom runtime | Orchestrator wrapping CLIs |
| Session | Continuous context | Isolated per wake-up |
| Observability | Debug via logs | Glass Box: inspect & edit in real-time |
| Rollback | Filesystem writes | Git-versioned Desk = Undo button |
Contributions welcome! See CONTRIBUTING.md for guidelines.
Before diving in, consider the strategic questions facing the project - feedback on architectural trade-offs is especially valuable. Join the Discussions to share ideas or ask questions.
For exploratory ideas and concepts still in development, see IDEAS.md.
AGPL-3.0 License - see LICENSE for details.