Skip to content

Latest commit

 

History

History
37 lines (25 loc) · 2.74 KB

File metadata and controls

37 lines (25 loc) · 2.74 KB

Quickstart

Get a governed, sandboxed agent running on your laptop in three commands — no Azure account, no Rust, no clone.

kars first run on a local kind cluster
First run on a local kind cluster (replayable asciinema cast).

📋 You need: kind · kubectl · any container runtime (Docker, Podman, or nerdctl — kind drives all three) · Node.js 22+ · a GitHub Copilot seat (any tier). Nothing else — no Azure account, no Rust.

# 1. Install the CLI (public, signed, SLSA-attested)
npm i -g @kars-runtime/cli

# 2. Bring up a governed agent on a local kind cluster (the real production pod shape)
kars dev --release --target local-k8s

# 3. Chat with it
kars connect dev-agent

On first run, kars dev asks you to pick an inference provider — choose GitHub Copilot (one device-code login, no Azure account). That's it: you now have an agent on a real Kubernetes cluster whose every model call, tool call, and network request is brokered by the in-pod Rust router — exactly as it runs on AKS.

Even faster, less faithful. Drop --target local-k8s to run a single container instead of a kind cluster (kars dev --release). It's the quickest path to a chat, but it co-locates the agent and router — not the production pod shape. This path uses the docker CLI specifically.

What just happened?

kars dev --release --target local-k8s pulled the published, cosign-signed images and ran them on a local kind cluster in the real sandbox shape — agent (UID 1000) and inference router (UID 1001) in separate containers, with the egress-guard init container and a NetworkPolicy. The agent runs with no credentials of its own — the router holds them and enforces identity, content safety, token budgets, tool policy, and a tamper-evident audit chain on every call. See Architecture → Two modes for exactly what is and isn't isolated in each mode.

Next steps