ase-python 0.1.6

Agent Simulation Engine — pytest for AI agent tool calls

ASE

ASE is the open pre-production testing and certification layer for agent systems.

ASE helps teams validate what an agent did, not just what it said:

• ase watch shows live tool calls
• ase test runs scenarios with assertions on behavior
• ase compare diffs two runs after a prompt, model, or adapter change

Install

python3.11 -m venv .venv
source .venv/bin/activate
pip install ase-python

Quickstart

ase test examples/customer-support/scenarios/refund-happy-path.yaml
ase compare /tmp/baseline.trace.json /tmp/candidate.trace.json

What ASE Does

• captures tool calls, state transitions, handoffs, and protocol events
• evaluates traces with assertions like tool_called, max_tool_calls, and policy checks
• certifies adapter-backed frameworks through a neutral event protocol
• works across proxy, instrumented, and adapter-backed agent runtimes

Project Links

• Source repository: see the project homepage that publishes this package
• Documentation: see README.md and docs/ in the source repository

Status

ASE is release-hardening toward a broader framework certification story. The current public positioning is:

the open pre-production testing and certification layer for agent systems