Eduard van Valkenburg
fc9c81b0b1
* Fix #3600: Pass JSON schemas through without Pydantic conversion This change optimizes FunctionTool and MCP flows by passing JSON schemas directly to providers without converting them to Pydantic models first. Key changes: - Store JSON schema as-is when supplied to FunctionTool - Skip Pydantic model_validate for schema-supplied tools in invoke() - Return MCP tool schemas directly without conversion - Add comprehensive tests for schema passthrough behavior Performance benefits: - Eliminates expensive Pydantic model creation for supplied schemas - Preserves exact schema structure (additionalProperties, custom fields, etc.) - Reduces memory overhead and initialization time Maintains backward compatibility: - Function signature inference still uses Pydantic models - Explicit Pydantic models passed as input_model work as before - All existing tests pass * Fix schema passthrough validation and remove helper * Simplify FunctionTool without generic model dependency * Fix FunctionTool typing fallout in 3600 * Remove FunctionTool[Any] compatibility shim * Use serializable kwargs in OTEL tool args
Agent Framework Lab
This is the experimental package for Microsoft Agent Framework, agent-framework-lab, which contains
various lab modules built on top of the core framework.
Lab modules are not part of the core framework and may experience breaking changes or be deprecated in the future.
What are Lab Modules?
Lab modules are extensions to the core Agent Framework that fall into one of the following categories:
- Incubation of new features that may get incorporated by the core framework.
- Research prototypes built on the core framework.
- Benchmarks and experimentation tools.
Lab Modules
- gaia: Evaluate your agents using the GAIA benchmark for general assistant tasks
- tau2: Evaluate your agents using the TAU2 benchmark for customer support tasks
- lightning: RL training for agents using Agent Lightning
Repository Structure
agent-framework-lab/
├── pyproject.toml # Single package configuration for agent-framework-lab
├── README.md # This file
├── LICENSE # License file
├── namespace/ # Centralized namespace package files
│ └── agent_framework/
│ └── lab/
│ ├── gaia/ # Re-exports from agent_framework_lab_gaia
│ ├── lightning/ # Re-exports from agent_framework_lab_lightning
│ └── tau2/ # Re-exports from agent_framework_lab_tau2
├── gaia/ # GAIA module implementation
│ └── agent_framework_lab_gaia/
├── lightning/ # Lightning module implementation
│ └── agent_framework_lab_lightning/
└── tau2/ # TAU2 module implementation
└── agent_framework_lab_tau2/
This structure maintains a single PyPI package agent-framework-lab while supporting modular imports through the namespace package mechanism.
Installation
To install each lab module, use the extras syntax with pip:
pip install "agent-framework-lab[gaia]"
pip install "agent-framework-lab[tau2]"
pip install "agent-framework-lab[lightning]"
Usage
Import and use lab modules from the agent_framework.lab namespace.
For example, to use the GAIA module:
# Using GAIA module
from agent_framework.lab.gaia import GAIA
Should I consume Lab Modules?
If you are looking for stable and production-ready features, you should not use lab modules. Stick to the core framework.
If you are looking for experimentation, research, or want to benchmark different approaches -- most importantly, if you don't mind breaking changes and potential deprecations -- then lab modules are for you.
Contributing to Lab Modules
Microsoft-maintained modules
For Microsoft-maintained modules in this repository, please follow standard contribution guidelines and submit pull requests directly to this repository.
Community modules
If you want to contribute a community-maintained lab module:
- Create a new repository on GitHub for your module
- Tag your repository with
agent-framework-labfor discoverability - Submit a PR to add a link to your repository in the Lab Modules section above
- Use the PR title format:
[New Lab Module] Your Module Name
We will review your submission based on the guidelines below.
Guidelines
- Purpose: Community modules should fit into one of the three categories of lab modules (incubation, research, benchmarks)
- Namespace: Community modules should avoid the
agent_framework.labnamespace (reserved for modules maintained in this repository) - Dependencies: Minimize external dependencies, always include
agent-frameworkas a base dependency - Documentation: Include comprehensive README with installation instructions and usage examples
- Tests: Write comprehensive tests with good coverage
- Type hints: Always include type hints and a
py.typedfile - Versioning: Use semantic versioning, start with
0.1.0for initial releases