paf

paf is a small, opinionated agent framework built on pydantic-ai. You describe an agent in a YAML spec, choose which capabilities it may use, and paf assembles the agent and runs it with durable conversations, long-term memory, context management, and runtime-enforced security.

It ships with a terminal chat (paf chat), an SSE server for web front ends, and a Python API (Session) that both are built on. The repository README covers installing and running it.

Documentation

• Spec reference: every key an agent spec can set, with its default.
• Capabilities: what an agent can do, and how each part works.
  • Conversations and memory
  • Context management
  • Files and code
  • Web and vision
  • Orchestration and extensions
• Security: how the defenses fit together and how to choose a setup.
  • Approvals: a person says yes or no to one call.
  • Shell command policy: allow, ask or refuse each shell command.
  • Tool gating: allow, ask or refuse each tool call, by path.
  • Prompt injection: screening tool results, and tightening the run after a hit.
  • Secrets: handles in, values never.
  • Sandbox: Docker, run_python, credential files, and the file tools.