Skip to main content
Funky is a durable runtime for agent swarms. Define how an agent should behave, choose where it should run, and start sessions for individual tasks. Funky handles the execution infrastructure and keeps each session’s history and state available as the agent works.

​
Core concepts

Create agents and environments once, then reuse them across sessions. Start a new session for each independent task or durable conversation.

​
How it works

  1. Create an agent. Choose a model and define its instructions and tool policy.
  2. Create an environment. Configure the sandbox and the network access the agent needs.
  3. Start a session. Reference the agent and environment to create an isolated run for a task.
  4. Send a message. Give the session its first task. The agent can use tools and return its response while Funky records the session’s events.

​
When to use Funky

Funky is a good fit when your agents need:
  • Long-horizon execution for tasks that take many steps or run for minutes or hours
  • Background operation that continues beyond the lifetime of a web request, worker, or local process
  • Durability so a session can pause, recover from infrastructure failures, and resume from its recorded state
  • High concurrency for running many independent sessions or coordinating large agent swarms in parallel
Ready to run an agent? Continue to the Quick Start.