Try it in your browser

This is the recommended first step — no local installs, nothing to configure. The hosted [Edge Studio Canvas](/studio) runs entirely on LyboAI Cloud: design a workflow visually, give it knowledge, gate it with evals, sign and download a .lybopack, all in the browser. Prefer code? The Google Colab notebook (Option B) drives the same runtime from Python. *(Option A below — a full source workspace in GitHub Codespaces — requires licensed source access, since the repository is private.)*

1. 1

   Open a Codespace

   On the GitHub repo page: Code → Codespaces → Create codespace on main. First boot builds the workspace and runs the test suite (~2–3 minutes).

2. 2

   Launch Studio

   In the Codespace terminal. This signs the example packs into packs/dist (first run mints a local dev key) and starts Studio with them trusted. Add --with-llm to also download a tiny real model (Qwen2.5-0.5B) and attach it so answers are generative:

   ./scripts/try.sh              # built-in deterministic tier
   ./scripts/try.sh --with-llm   # + a real tiny LLM (Qwen 0.5B)

3. 3

   Open the forwarded port

   Codespaces forwards port 4242 and pops an “Open in Browser” toast. Studio prints a URL containing a ?token=… in the terminal — open the forwarded port and append that same ?token=… (the token gates the API). Studio automatically allows the Codespaces forwarded domain.

4. 4

   Attach your own model (optional)

   In Studio's “Local model (BYO)” panel, click Download to fetch a GGUF (defaults to Qwen 0.5B) or paste a local .gguf path, then Attach — new turns immediately use that model. No restart needed.

5. 5

   Things to try

   what does the manual say about the pump seal
   capture this transcript asset_id: PUMP-7 condition: leaking seal
   create a task for calibrating the flow sensor
   help me plan the launch: write docs then record a demo

Option B — Google Colab (Python notebook)

Prefer notebooks? notebooks/lyboai_quickstart.ipynb is a ready-to-run Google Colab notebook: one setup cell bootstraps the runtime inside the Colab VM, then you build an agent cell by cell — local knowledge with sourced answers, a human-approval workflow (you click approve), routing trained from labelled examples with train_triggers, a pack built and signed with PackBuilder, an eval gate, and the tamper-evident audit check. No API keys, no local installs.

1. 1

   Open the notebook in Colab

   Open the URL directly:

   https://colab.research.google.com/github/rajandua20/lyboai-releases/blob/main/notebooks/lyboai_quickstart.ipynb

2. 2

   Run the setup cell (~30 seconds)

   Downloads the prebuilt lybo binary from the public GitHub release and installs the lyboai SDK from PyPI — no source build, no Rust toolchain.

3. 3

   Run the rest top-to-bottom

   Each cell is one concept: knowledge → RAG → approval workflow → training → pack signing → evaluation → observability.