Skip to content

AgentBridge Roadmap

v0: SDK Foundation

  • Requirements, research, architecture, and roadmap docs.
  • Python package scaffold.
  • Core framework-neutral types.
  • Adapter registry.
  • Mock backend.
  • Optional adapter skeletons for Pydantic AI, CrewAI, and LangGraph.
  • Examples for refunds, research, and AG-UI-shaped events.
  • Unit tests without paid API calls.

v0.1: Real Backend Execution

  • Pydantic AI adapter execution with tool calling.
  • LangGraph minimal graph execution, conditional routing, checkpoint interrupt reporting, and compiled-runtime resume.
  • Contract tests gated behind optional dependency markers.
  • CLI commands for running, validating, comparing, inspecting, and conformance-checking backends.
  • Backend capability metadata for supported features and known gaps.
  • docs/version_policy.md kept current as adapter dependency ranges change.
  • Static JSON/YAML manifests for CLI runs.
  • Scaffold command for lightweight external adapter plugins.
  • External plugins for OpenAI Agents SDK, Strands Agents, direct LangChain, and Google ADK with adopted version ranges, offline contract tests, and capability-specific diagnostics.
  • CrewAI dependency strategy remains unresolved in this Python 3.14 environment and should be verified separately before claiming native execution.

v1: Migration and Comparison

  • Agent manifest file format.
  • agentbridge compare command for backend fit analysis.
  • Tool registry for manifest-declared tools.
  • Existing-project import helpers for common CrewAI and LangGraph structures.
  • Existing-project import helpers for common LangChain agent apps.
  • Conservative MigrationReport helpers for LangChain and LangGraph object shapes.
  • Tracing and diagnostics surfaces for plugin backends, with provider-specific smoke tests where credentials/config are available.
  • Provider-neutral offline evaluation reports with per-example scores, failures, and aggregate summaries.
  • Deep scenario reports that demonstrate framework-specific features, migration deltas, normalized outputs, and model routing variants.
  • Full AG-UI server example.

v2: Full Framework Coverage

  • Capability matrix for each supported framework.
  • Framework-specific extension modules for features that do not fit the common core.
  • Conformance test suite that every adapter must pass.
  • Coverage reports showing which framework features are fully supported, partially supported, unsupported, or intentionally native-only.
  • Scenario report suite covering deep framework nuances, framework-to-framework migrations, and model routing across hosted APIs, local models, OpenRouter, NVIDIA NIM, and custom gateways.
  • Adapter authoring guide for new frameworks and tools.
  • First-class external adapter packages for OpenAI Agents SDK, Strands Agents, LangChain, Google ADK, CrewAI, and additional ecosystem frameworks.

Continuous Compatibility Program

  • Weekly upstream version-drift checks with isolated core and external-plugin conformance jobs.
  • Machine-readable framework feature inventories linked to capability, extension, native-only, and unsupported outcomes.
  • Credentialed smoke-test lanes for provider, observability, governance, and deployment integrations.
  • Automated compatibility reports and visible scheduled failures when upstream releases leave adopted ranges.
  • AgentCore Runtime, Memory, Gateway, Identity, protocol, and observability bridge with generic native API access.
  • LangSmith thread/run helpers and Langfuse ingestion/score helpers layered over dependency-free transports.