Get started

Quickstart

Chat with a local agent in Studio in five minutes.

Studio is a local web UI over the exact same runtime that ships in apps. You'll install the three signed example packs, route requests to them, and approve an agent action.

  1. 1

    Sign the example packs (creates packs/dist)

    The example packs must be signed before the runtime will install them. This helper mints a local dev key on first run, signs each pack into packs/dist, and prints the public key to trust.

    PUB=$(./scripts/sign-example-packs.sh)
  2. 2

    Launch Studio with the example packs

    Or simply run ./scripts/try.sh, which does the signing and this launch in one step (add --with-llm for a real model).

    cargo run -p lybo-cli -- studio --port 4242 \
      --packs packs/dist --semantic --builtin \
      --trust "lybo-examples:$PUB"
  3. 3

    Open the printed URL and talk to the agent

    Studio prints http://127.0.0.1:4242/?token=… — open that exact URL (the token is required). Try: “what does the manual say about the pump seal” — a RAG answer grounded in local knowledge, with sources.

  4. 4

    Trigger a human-in-the-loop approval

    Try: “capture this transcript asset_id: PUMP-7 condition: leaking seal severity: medium”. The workflow extracts a structured record, then pauses with Approve/Reject buttons before saving anything.

  5. 5

    Inspect what happened

    The Traces panel shows every step (routing, model calls, schema validation, policy decisions). Click “verify audit chain” — traces are tamper-evident.

No GPU, no API key, no network needed: a deterministic model tier guarantees the runtime works before any neural model is installed. For a real model, run ./scripts/try.sh --with-llm (ships a tiny Qwen 0.5B), attach any GGUF from Studio's Local model (BYO) panel, or wire your own engine with --model-cmd (see *Models & routing*).