AutoGen conversational agents with LLM calls routed through the AMC Gateway.
- Creates a MathTutor assistant and a Student user proxy
- Runs a multi-turn conversation to solve a calculus problem
- AMC captures all LLM interactions via the gateway proxy
- Python ≥ 3.10
- AMC CLI installed with the checksum-verified GitHub Release installer (npm is not public yet; see the root README)
- An OpenAI API key
cd examples/autogen
pip install -r requirements.txtamc up
amc wrap autogen-cli -- python main.py[AMC] Routing LLM calls through gateway: http://localhost:3700/v1
=== AutoGen Conversational Agents ===
Student (to MathTutor): What is the integral of x^2 from 0 to 3?
MathTutor (to Student): Let me solve this step by step...
[AMC] All LLM calls captured as evidence via gateway proxy.
AutoGen's config_list accepts a base_url parameter. AMC's autogen-cli adapter sets OPENAI_BASE_URL and OPENAI_API_KEY environment variables, which are read into the config. The adapter also sets OPENAI_API_BASE for older AutoGen versions.
The adapter definition is in src/adapters/builtins/autogenCli.ts.