Agents & Skills
Skills, not glue code.
A skill is a packaged capability an agent can pick up: a role, a model, and the workload it drives. These ship with every workspace — attach one from the console, wire an AI provider, and it runs under your workspace identity.
All skillsBrowserCodeOps
🖥
VM daemonops
runs on: your VMs
An agent living beside your VM that operates it over SSH — install software, debug services, manage files. Chat with it in the console, dispatch tasks over the API, and watch every step in its trajectory.
bring your own providerAttach
🔍
Browser agentbrowser
runs on: browser
Drives a headless browser through a trajectory of sub-goals — research, scrape, fill forms. Every step comes back with a review verdict you can accept or reject before it proceeds.
bring your own providerAttach
⌘
Master agentops
runs on: workspace-wide
An orchestrator for your workspace: it plans across your workloads and provisions, restarts or rewires resources through the same public API documented here.
bring your own providerAttach
✎
App scaffoldercode
runs on: Create with AI
Describe an app in the console; it writes the repo in your workspace Git org, builds from source (no Dockerfile) and deploys a live container app with its own hostname.
bring your own providerAttach
⟲
Code reviewercode
runs on: workspace repos
Watches the repos in your workspace org and reviews changes as they land — findings arrive before the auto-build rolls your app.
bring your own providerAttach
Engines & AI providers
Every agent runs on an AI provider you connect once on the Integrations page — the key is stored write-only and shared by all agents in the workspace. The engine an agent runs on is picked automatically from the provider; there is nothing to configure.
🔑
API-key providers
OpenAI, Anthropic, Gemini, GLM coding plans and a dozen more — paste the provider's API key and every agent kind can use it.
✳
Claude subscription (Pro/Max)
Bring your own Claude plan instead of an API key: run claude setup-token on your machine and paste the token. Agents on it run the Claude Agent engine, and the model picker shows exactly what your plan can use — usage draws on your subscription, not on per-token billing. If the token expires or the plan hits its usage window, the agent says so plainly and you reconnect from Integrations.
Provider keys and tokens are write-only: the platform can use them on your agents' behalf but can never show them back — to anyone.
Teach your own agent.
Everything an agent needs to operate LiveLLM lives in one SKILL.md — hand it to Claude, a workspace daemon or your own harness and it can drive the whole API.