Skillget

litellm-agent-platform

agent
Security Audit
Fail
Health Pass
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Community trust — 48 GitHub stars
Code Fail
  • fs.rmSync — Destructive file system operation in cli/bin/lap.mjs
  • os.homedir — User home directory access in cli/bin/lap.mjs
  • process.env — Environment variable access in cli/bin/lap.mjs
  • network request — Outbound network request in cli/bin/lap.mjs
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

Self-hosted platform for running coding agents (Claude Code, Codex) in isolated sandboxes with vault proxy.

README.md

LiteLLM Agent Platform

Discord

Self-hosted platform for running coding agents in isolated sandboxes.

Run Claude Code, Codex, or any coding-agent harness in its own Kubernetes sandbox. A vault proxy sits in front of each sandbox and swaps stub credentials for real ones on outbound calls — the agent process never sees your raw credentials.

Xnapper-2026-05-08-19 10 50

Why this exists

In most enterprises you can't just run claude --dangerously-skip-permissions on a corporate laptop. IT won't approve it. So developers babysit a permission popup every two minutes instead of shipping.

Run the agent inside a sandbox where the env contains only stub credentials and the vault swaps them at egress, and bypass-permissions becomes safe to enable.

How it works — developer flow

A developer never deals with Kubernetes or the vault. They install one CLI and run one command.

1. Install the lap CLI

git clone https://github.com/BerriAI/litellm-agent-platform.git
cd litellm-agent-platform/cli
npm install
chmod +x bin/lap.mjs
ln -sf "$PWD/bin/lap.mjs" ~/.local/bin/lap

2. Log in to your platform

lap login
#   Agent platform URL: https://lap.acme.dev
#   Master key:         ••••••••••••••••
#   ✓ saved to ~/.lap/config.json

3. List the agents your team has configured

lap agents

4. Open a sandbox

lap claude-code-cli1

That command spins up a fresh Kubernetes pod running Claude Code, attaches your local terminal to its TTY over a WebSocket, and drops you straight into the agent. Same feel as ssh — your iTerm / tmux / wezterm stays exactly where it is. Press Ctrl-D to detach; the session stays alive for 24h and you can reconnect by running lap <agent> again.

What's running in the sandbox

  • The actual claude CLI under node-pty
  • Working tree at /work/repo, optionally cloned at boot
  • Credentials in the pod's env are stub placeholders:
    GITHUB_TOKEN=stub_github_a8f1
LITELLM_API_KEY=stub_litellm_bb20
    Vault swaps them for the real values inline on every outbound TLS connection. The agent can echo $GITHUB_TOKEN all it wants and only get the stub.

Full CLI docs: docs/lap-cli.md.

Self-hosting the platform

Sandboxes run on Kubernetes via the kubernetes-sigs/agent-sandbox CRD. Local dev uses kind.

Prereqs: Docker Desktop, kind, kubectl, helm, a LiteLLM gateway URL.

bin/kind-up.sh
docker compose up

bin/kind-up.sh is idempotent — provisions a kind cluster agent-sbx, installs the agent-sandbox controller, and loads the harness image. docker compose up boots Postgres, runs the schema migration, and starts web (:3000) + worker.

Open localhost:3000 to create an agent. Then point lap at it and run through the steps above.

Architecture and tuning: docs/k8s-backend.md.

Deploying to production

Recommended path: AWS EKS for the sandbox cluster, Render for web + worker. See deploy/ — bin/eks-up.sh provisions the cluster, the Render Blueprint at the top of deploy/render/README.md is one click.

Developer API

Create an agent, open a session, send a message, read the reply — directly with curl. See docs/spawn-task-agent.md and src/server/DEVELOPER.md.

License

MIT — see LICENSE.

Reviews (0)

No results found