-
Notifications
You must be signed in to change notification settings - Fork 0
Getting Started
- Linux host you trust (a VM/container is great for first runs)
- Python 3.10+
- Optional but recommended: Docker (for Postgres + Redis), and a Claude/ChatGPT subscription or an LLM API key
git clone https://github.com/yohn1985/jarvisbot.git
cd jarvisbot
./install.sh upinstall.sh up sets up a virtualenv + dependencies, brings up the optional services, and starts the dashboard. It prints your login link at the end:
http://<this-host>:8787/?token=…
The link embeds a private access token (the dashboard is gated). Open it once — the dashboard stores a cookie and you stay logged in.
Lost the link?
./install.sh url # re-print the full login link
./install.sh token # print just the raw tokenThe token is your password. Keep the link private; anyone with it can reach the dashboard on your network.
Open Settings → Brain & Models:
- Subscription login (cheapest): a Claude (Max/Pro) or ChatGPT/Codex plan via the CLI.
- API key (pay per use): paste an OpenAI/Anthropic/Ollama-Cloud key.
Then set per-role routing (the main brain should be a frontier model; spawned agents can use a cheaper one). Effort/thinking controls appear for models that support them.
Go to Chat and talk to it. Try:
- “what machine are you on and what’s running?”
-
/web is ubuntu 26 out— run a skill from chat (type/to see them) - attach an image (📎 or paste) and ask about it
Jarvis ships in shadow mode (it proposes, never executes). To let it act on its own, raise its mode and grant action classes in Settings → Behavior — deliberately, and watch the Activity tab.
./install.sh breathe # one shadow tick (perceive → decide → think → act)
./install.sh run # the persistent wake loop
./install.sh doctor # health check
./install.sh selfcheck # the tamper-proof fitness score
./install.sh skill list|install <name>|run <name> [args]