Metadata-Version: 2.4
Name: agent-learning
Version: 0.7.0
Summary: Evidence-driven decision learning for AI agents with inspectable autonomy gates.
Author: Chris Tava
License: MIT License
        
        Copyright (c) 2026 Microsoft
        
        Permission is hereby granted, free of charge, to any person obtaining a copy
        of this software and associated documentation files (the "Software"), to deal
        in the Software without restriction, including without limitation the rights
        to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
        copies of the Software, and to permit persons to whom the Software is
        furnished to do so, subject to the following conditions:
        
        The above copyright notice and this permission notice shall be included in all
        copies or substantial portions of the Software.
        
        THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
        IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
        FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
        AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
        LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
        OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
        SOFTWARE.
        
Project-URL: Homepage, https://github.com/microsoft/agent-learning
Project-URL: Repository, https://github.com/microsoft/agent-learning
Project-URL: Issues, https://github.com/microsoft/agent-learning/issues
Keywords: agentic-decision-making,reinforcement-learning,ai-agents,azure,evaluation
Classifier: Development Status :: 4 - Beta
Classifier: Intended Audience :: Developers
Classifier: License :: OSI Approved :: MIT License
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.10
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.10
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: numpy>=1.24
Requires-Dist: azure-identity>=1.15
Requires-Dist: azure-ai-evaluation>=1.0.0
Requires-Dist: pydantic>=2.0
Provides-Extra: dev
Requires-Dist: pytest>=7.4; extra == "dev"
Requires-Dist: pytest-asyncio>=0.21; extra == "dev"
Requires-Dist: pytest-cov>=4.1; extra == "dev"
Requires-Dist: ruff>=0.1; extra == "dev"
Requires-Dist: mypy>=1.6; extra == "dev"
Provides-Extra: cosmos
Requires-Dist: azure-cosmos>=4.5; extra == "cosmos"
Provides-Extra: nlp
Requires-Dist: scikit-learn>=1.3; extra == "nlp"
Requires-Dist: scipy>=1.10; extra == "nlp"
Requires-Dist: joblib>=1.3; extra == "nlp"
Provides-Extra: examples
Requires-Dist: pyyaml>=6.0; extra == "examples"
Provides-Extra: slm
Requires-Dist: onnxruntime-genai>=0.5; extra == "slm"
Provides-Extra: llm
Requires-Dist: azure-ai-evaluation>=1.0.0; extra == "llm"
Requires-Dist: azure-identity>=1.15; extra == "llm"
Dynamic: license-file

# agent-learning

Native reinforcement learning SDK for AI agents. An in-process Learner optimizes a small, interpretable TaskPolicy over discrete agent choices (understand intent and complete task by choosing the right outcome).

TaskPolicies model reusable decisions among executable alternatives such as
models, skills, tools, workflows, or workloads. Factual questions, ordinary
chat, reporting, and learning automation are not policy tasks.

## How it works

The SDK improves agents without LLM weight fine-tuning. There are no GPU fine-tune jobs and no opaque update cycles — just three pieces that run in your existing Python process:

1. **TaskPolicy** is a softmax distribution over `N` discrete actions (e.g., "take action A", "take action B", "take action C"). It lives in Python and updates in milliseconds.

2. **Score** evaluates each episode on-device with three stdlib scorers for intent resolution, task adherence, and task completion. Their scores are combined into a single scalar reward with no scoring endpoint or environment variables required. Azure AI evaluators remain available as an opt-in.

3. **Learner** applies REINFORCE-with-baseline to update TaskPolicy logits directly from logged episodes. Updates are tiny gradient steps that run on local compute and persist through a pluggable store — in-memory or local files by default, with Azure Cosmos DB optional.

`task-policy-decide` closes the loop at execution time by returning the selected
action plus historical correctness, reward, result summaries, and per-metric
quality feedback for the agent to use on its next delegated decision.

It also returns a complexity-proportional autonomy assessment. A persisted
profile covers intent ambiguity, context variability, outcome observability,
decision impact, reversibility, and mandatory approval; action-space size is
derived. The resulting low, standard, high, or critical tier scales required
outcomes, Wilson confidence, reward, probability, margin, stable snapshots, and
drift-audit rate. Autonomous executions continue learning from observable
outcomes, while tier-scaled samples request user feedback to detect drift.
An explicit accepted-feedback episode is a separate durable authorization path:
it pins that action for the task policy and suppresses future feedback prompts
until the user explicitly rejects it.

Every episode, reward, run, and deployment is captured by the configured store — in-memory or local files by default, or Azure Cosmos DB — giving you a complete lineage and audit trail of how the policy evolved over time.