Agents that work for you — and stay under your control.

Build agents that run on your own models

Plugsky gives you everything needed to build AI agents: a models API with function calling, agent SDKs for Python and TypeScript, and a playground where you can test autonomous agents safely before shipping them. Agents run on your choice of 30+ models — including on sovereign infrastructure.

Agent features

• Function calling — agents can use tools with structured outputs on all modern models.
• Long context — up to 1M context for agents that remember.
• Audit logs — every agent action is logged for review.
• Human-in-the-loop gates — pause agents before money-moving or irreversible actions.
• Sovereign runtime — run your agents in your region or on-prem.

Agent guides

• Agent API quickstart (Python)
• Agent API quickstart (TypeScript)
• Agent security checklist
• Agent orchestration guide
• Failure and retry design
• Agent evaluation framework

Try the playground

Test autonomous agents in the playground beta — give an agent a goal, watch it plan, and review its work before it acts. No API key needed to try it.

Agents FAQ

Which models support function calling?

All modern models in the catalog. See the model reference for per-model tool support.

Can agents run on-prem?

Yes — the agent API runs against any Plugsky endpoint, including on-prem deployments.

Is the agent API OpenAI-compatible?

The model calls use the OpenAI-compatible API, so existing agent frameworks work by changing base_url.