AgentBridge Documentation

Use this page as the navigation map for the repository.

Start Here

• README: Public project overview, quickstarts, architecture summary, and development workflow.
• Vision And Target State: Final target, goals, plugin families, and long-term ecosystem map.
• Requirements: Product requirements, target users, non-goals, success criteria, and acceptance criteria.
• Roadmap: v0, v0.1, and v1 milestones.

Product And Research

• Articles: Shareable GitHub posts about why AgentBridge exists and how the compatibility layer works.
• Research: Comparison of AG-UI, LiteLLM, CrewAI, LangGraph, and Pydantic AI.
• Adapter Target Research: Next adapter targets including OpenAI Agents SDK, Strands, direct LangChain, and Google ADK.
• Capability Coverage: Long-term strategy for covering framework nuance without flattening everything into a weak abstraction.
• Scenario Reports: Target format for deep framework-nuance examples, migration comparisons, outputs, and model-provider routing.
• Model Routing: How AgentBridge records hosted, local, OpenAI-compatible, and offline model routes without becoming a model gateway.
• Version Policy: Adopted package ranges, verified local versions, and adapter version rules.
• Release Workflow: Package build checks, tag workflow, and future PyPI publishing path.

Engineering

• Architecture: System architecture, data flow, plugin loading, manifest flow, and normalized type model.
• Visual System Maps: Shareable Mermaid diagrams for market positioning, migration, capability layers, plugin boundaries, and support status graduation.
• Design: Design decisions, tradeoffs, extension principles, and boundaries.
• Adapters: Backend support matrix, current limitations, and adapter-specific notes.
• CLI: Command reference for validating, running, comparing, and inspecting agents.
• Conformance: Adapter contract checks and offline plugin model rules.
• Migration Helpers: Conservative import reports for LangChain and LangGraph project shapes.
• Plugin Authoring: How to build adapter plugins outside the core package.

Examples

• examples/refund_agent.py: Same agent spec against selectable framework adapters.
• examples/research_agent.py: Tool use and streaming event demo.
• examples/agui_events.py: Convert normalized AgentBridge events into AG-UI-shaped events.
• examples/framework_extensions.py: Framework-specific extension config examples for LangChain, OpenAI Agents, Strands, and Google ADK.
• examples/langchain_native_memory_retriever.py: Offline LangChain example with native retriever, checkpointer, and store objects.
• examples/deep_scenario_report.py: No-key report-shaped example for deep framework features, migrations, outputs, and model routing.
• examples/openai_agents_approval_report.py: No-key OpenAI Agents-to-LangGraph approval queue/resume report with output comparison.
• examples/crewai_prototype_report.py: No-key CrewAI-to-LangGraph prototype migration report with role/task/crew nuance.
• examples/rag_migration_report.py: No-key LangChain-to-LangGraph RAG migration report with output comparison.
• examples/strands_agentcore_report.py: No-key Strands-to-LangGraph AgentCore production-path report with output comparison.
• examples/google_adk_enterprise_report.py: No-key Google ADK-to-LangGraph enterprise services/eval/deployment report with output comparison.
• examples/pydantic_validation_report.py: No-key Pydantic AI-to-LangGraph typed validation report with output comparison.
• examples/scenario_report_suite.py: Aggregate no-key scenario report suite and framework/backend coverage index.
• examples/credentialed_smoke_matrix.py: Readiness matrix for double-gated hosted/local/provider smoke paths.
• examples/model_routes.py: Machine-readable model route catalog for offline, hosted, local, OpenRouter, NVIDIA NIM, and custom gateways.
• examples/refund_agent.yaml: Static manifest for CLI runs.

Maintainer Checklist

Before claiming support for a framework feature:

• Add or update adapter capability metadata.
• Add contract tests for every capability marked full.
• Run conformance for the changed backend.
• Update Version Policy with adopted and verified versions.
• Update Adapters with limitations and native escape hatches.
• Add example coverage when the feature changes user-facing behavior.