Skip to content

ACP Kit 1.1.0

1.1.0 adds the ACP-backed Pydantic AI provider bridge and tightens the v1 release surface around current framework versions and maintained examples.

Highlights

  • pydantic-acp now exports AcpProvider, AcpModel, and AcpHostBridge so an existing ACP agent can be consumed as a Pydantic AI v2 provider/model pair.
  • AcpProvider(acp_agent=...) is the explicit construction API for wrapping an ACP agent instance.
  • provider.model() leaves model selection to the wrapped ACP agent's session default; pass a concrete model name only when the ACP agent accepts that session/set_model id.
  • Maintained examples now write generated workspaces and local session state under agent_demos/, with .gitignore ignoring that single runtime output directory.
  • Workspace package version files are synchronized at 1.1.0.

Compatibility

The release gate covers:

  • Python 3.11 through 3.13
  • Pydantic AI 2.0 through 2.4
  • LangChain 1.3.11
  • LangGraph 1.2.7
  • DeepAgents 0.6.12
  • ACP Python SDK 0.9.0

Migration Notes

Use AcpProvider only when the runtime you have is already an ACP agent. If you are exposing a pydantic_ai.Agent through ACP, keep using create_acp_agent(...) or run_acp(...).

If an ACP agent rejects "agent" as an explicit model id, call provider.model() without arguments. That path no longer sends session/set_model("agent"); it lets the wrapped agent keep its own default.