First-day walkthrough. Takes about 10 minutes end-to-end and leaves you with a working alpi — model selected, workspace pinned, first session running, and alpi doctor green. Profiles, gateways, apps, and ALP come next.
For concepts behind any step, see ARCHITECTURE.md. For the security posture, see SECURITY.md. Why alpi is built the way it is: see README.md.
1. Install
uv tool install alpi-agent # install
alpi setup # use
PyPI package is alpi-agent; the binary is alpi. Alternatives and update path → INSTALL.md.
Needs Python ≥ 3.10. The browser tool downloads Chromium (~200 MB) the first time it runs — no separate install step.
Check it works:
alpi --version
2. Pick a model
The fresh profile ships without a default model — you choose your provider and your model. Open the setup wizard:
alpi setup
Pick Model / Provider, choose a provider, paste the API key when prompted, pick a model. For recommendations see docs/MODELS.md — if you want a single sensible choice, Claude Sonnet 4.6 (anthropic/claude-sonnet-4-6) is the pragmatic daily driver. If you want fully local (no cloud), install Ollama first and pick Ollama in the wizard.
3. Pin a workspace
Same wizard, pick Workspace. Point it at the directory alpi is allowed to read and write — typically the project you're working on. Without a workspace, alpi falls back to the current working directory at launch with a warning; that's fine for trying things, not for real use.
4. Send your first message
alpi
Type hola and press Enter. The TUI streams the model's reply. Useful first commands to try inside the TUI:
/help— list every slash command./memory— show what alpi remembers about you (USER.md,MEMORY.md,AGENT.md)./skills— list installed skills./status— session snapshot (turns, tokens, cost)./new— start a fresh session.
And try asking alpi about itself — "how do I configure the TUI theme?", "how does the ALP protocol work?". The alpi_knowledge tool answers from the packaged docs without leaving the terminal.
Leave it with Ctrl-C or /exit.
5. Resume where you left off
alpi -c
-c / --continue rehydrates the last session. To make this the default behaviour without typing -c every time, enable tui.auto_resume in ~/.alpi/config.yaml (or set it via alpi setup).
6. Optional: connect a gateway
If you want alpi to answer you on Telegram / IMAP / Gmail / Matrix, run the wizard again:
alpi setup → Gateways
Pick the channel, follow the prompts, then make sure the unified daemon is running so alpi answers 24/7:
alpi daemon status
alpi daemon start
Gateways are compatibility bridges for 24/7 reach. The primary surfaces are the terminal, desktop app, and mobile app talking to the daemon directly. The single alpi daemon process supervises every profile on the machine and hosts gateways, scheduler ticks, ALP listeners, workgroup pollers, and the host plane.
7. Add a second profile (optional)
If you want a work profile that's completely isolated from your personal one (different API keys, different memory, different Telegram bot):
alpi profile create work
alpi -p work setup
Everything is per-profile: home directory, sessions, memory, skills, keys, peers. See docs/PROFILES.md for the full model.
8. Link to other alpis (optional)
If you plan to link profiles or machines with other alpis, the daemon already exposes the Alpi Link Protocol listener when the profile's ALP subsystem is enabled. Then exchange pubkeys with peers:
alpi daemon status
alpi setup → ALP → Peers
See docs/DEPLOYMENTS.md for topologies (laptop + home server, multi-device, team, enterprise).
9. Check everything is healthy
alpi doctor
Runs live checks: model reachable, gateways logged in, MCP servers handshaking, services alive, ALP socket listening, peers reachable. Green = you're done.
Next steps
- docs/SKILLS.md — teach alpi reusable recipes (skills are live-by-default, scanner-gated, per-profile).
- docs/OPERATIONS.md — logs, services, upgrades, backups.
- docs/SECURITY.md — approval gate, sandbox, and how alpi protects you from the LLM.
- docs/DEPLOYMENTS.md — multi-alpi topologies from personal to enterprise.