Skip to content

v0.8.0

Choose a tag to compare

@evalstate evalstate released this 28 Jun 23:28
2f43843

fast-agent 0.8.0 release summary

Highlights

Python compatibility

  • Lowers minimum Python requirement from >=3.13.5 to >=3.12.

Harness API (unstable)

  • Introduces a protocol-neutral harness app boundary for UI/protocol adapters.
  • Refactors runtime startup, session handling, and agent invocation paths
    around this abstraction.

Auth and Hugging Face deployment improvements

  • Adds MCP auth forwarding support for Hugging Face-style inbound bearer token forwarding.
  • Adds Hugging Face auth provider compatibility helpers and token verifier coverage.
  • Adds Hugging Face Spaces MCP examples and docs.

New A2A support (unstable)

  • Adds fast_agent.a2a package with A2A client and server runtime support.
  • Supports connecting to remote A2A agents from CLI/TUI and AgentCards.
  • Adds A2A serving for fast-agent apps, including streaming, task state mapping, file/data parts, input-required flows, and auth-required handling.

MCP serving and adapter improvements

  • Replaces/expands the older MCP agent server path with harness-based MCP serving.
  • Adds HarnessMCPAdapter documentation and implementation for custom MCP servers backed by fast-agent sessions.
  • Improves MCP connection handling, auth propagation, prompt serialization, resource/content handling, and provider-managed MCP
    docs.
  • Deprecates "shared" request scope.

CLI/runtime refactor

  • Splits large CLI startup logic into focused modules for harness startup, model bootstrap, one-shot execution, session resume, and shell
    cwd preflight.
  • Adds A2A-related CLI options and parsing.
  • Improves result export and command-surface tests.

Session and history improvements

  • Adds session hydration, formatting, trajectory support, and reusable
    session runtime helpers.
  • Improves CLI/TUI session resume behavior and session command handling.
  • Adds E2E coverage for CLI session smoke flows.

Provider/model updates

  • Default model references/docs updated toward gpt-5.4-mini.
  • Updates provider SDK pins:
  • Adds Google streaming/usage telemetry improvements.
  • Adds Groq/Qwen and model namespace/catalog updates.
  • Deprecates/removes some older DeepSeek/distill references.

GEPA and batch/example updates

  • Updates GEPA and Trackio integration and docs.
  • Adds Hugging Face batch/demo dataset examples.
  • Refreshes HF API agent workflow card generation.

Notable compatibility notes

  • Legacy APIs or compatibility shims from earlier versions removed