python_1.1.4

2FastLabs/agent-squadpython_1.1.4Sep 23, 2026by brnaba-aws

Upgrading?

AI Summary

Python 1.1.4 mirrors the TypeScript release with the new JevClassifier and MCP enhancements, while fixing critical issues with the BedrockClassifier regarding the citations field and compatibility with newer Claude models.

Key Highlights

  • Added JevClassifier using TypeSafe's decision model
  • MCP protocol 2026-07-28 support via mcp 2.x
  • Streamable HTTP transport in MCPToolProvider
  • Tool-UI (widget) support in GroundedAgent
  • BedrockClassifier compatibility fixes

New Features

  • JevClassifier routing mechanism
  • MCP protocol support
  • Streamable HTTP transport
  • Tool-UI support in GroundedAgent
  • BedrockClassifier fixes for citations and temperature parameters

Full Release Notes

## What's new

- Added `JevClassifier`: routes with TypeSafe's Jev decision model instead of an LLM. The registered agents become a typed `choice` question, and Jev returns the chosen agent plus a calibrated confidence, billed on input tokens only. Standard library only, no new dependencies. See the [docs](https://2fastlabs.github.io/agent-squad/classifiers/built-in/jev-classifier) (#699)
- MCP protocol 2026-07-28 support via `mcp` 2.x, keeping 1.x compatible (#643)
- Streamable HTTP transport in `MCPToolProvider` (#639)
- MCP tool-UI passthrough in `MCPToolProvider` (#604)
- Tool-UI (widget) support in `GroundedAgent` (#591)

## Fixes

- `BedrockClassifier` failed on every call since `ConversationMessage` gained a `citations` field; it now sends only `role` and `content` (#698)
- Bedrock classifier and agents can now drop `temperature`/`topP` for models that reject them (Claude Opus 5, Sonnet 5, Opus 4.7/4.8), by setting them to `None` in `inference_config`. Defaults are unchanged (#698)
- Don't send an empty tool list to Bedrock/Anthropic (#603)

Related Repositories