Documentation

Serverless agents

AgentContainer lets you run AI agents the way serverless lets you run functions — describe the work, hit the API, get the result. No containers to manage, no orchestration to build, no infrastructure to maintain. Each task spins up an isolated environment with Pi, the open source agent runtime. Use one-shot autonomous tasks when you want the agent to run to completion, or conversational tasks when you want a persistent back-and-forth with the same task.

You pay per task: container time at a flat per-minute rate. Bring your own AI provider key — token costs go straight to your provider, not through us. No idle costs, no subscriptions, no upfront commitment. EU-hosted infrastructure with containers running in Amsterdam.

How it works

The core model is simple: define an agent once, then dispatch tasks.

1. Agent — a reusable worker definition. Set a prompt, a model, reference files, and runtime defaults, including the agent runtime.
2. Task — a single execution of that agent. It can run autonomously or as a conversation that resumes across turns.

Don't need custom configuration? Use a standard agent and dispatch tasks immediately — no setup required.

curl -X POST https://agentcontainer.co/api/v1/agents/agt_standard_coding/tasks \
  -H "Authorization: Bearer ac_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "instructions": "Read the attached PDF and extract all financial tables into a CSV."
  }'

What you get

Where to go from here

• Quickstart — go from zero to a completed task in under 5 minutes
• Core Concepts — understand how agents, runtimes, and tasks fit together
• Tasks & Output — dispatching work, output policies, and retrieving results
• Skills — install reusable agent instructions into task workspaces
• API Reference — every endpoint, request body, and response schema