Quickstart¶
This gets you from zero to an agent-driven array analysis without EdgeFEM (the C++ full-wave solver). Array patterns and system metrics work with pure-Python dependencies.
1. Install¶
pip install "apab[ollama]"
Install Ollama, then pull the default model:
ollama pull qwen2.5-coder:14b
Any Ollama model with tool-calling support works; set it in apab.yaml
under llm.model.
2. Scaffold a project¶
apab init --name my_array --quickstart
cd my_array
--quickstart writes a ready-to-run apab.yaml for an 8x8 array with
no solver dependency. Use --quickstart-fullwave instead if you have
EdgeFEM installed.
3. Check your environment¶
apab doctor
This verifies Python dependencies, EdgeFEM availability, and that the Ollama server is reachable with your configured model.
4. Run the agent¶
Interactive session:
apab design
you> Compute the pattern for this array at 28 GHz and evaluate a
500 m comms link with 200 MHz bandwidth.
Or non-interactive, driven by the config:
apab run --config apab.yaml
The agent picks tools (pattern_compute, system_evaluate, ...),
executes them, and reports metrics. Every session writes a run bundle:
workspace/runs/<run_id>/
├── audit.json every tool call, with arguments per your redaction mode
├── manifest.json config hash, dependency versions, status, token usage
└── artifacts/ patterns, plots, system results, report
5. Next steps¶
- Turn on OpenTelemetry tracing to see the full span tree for each run
- Read run bundles to understand what gets recorded and why
- Try the LangGraph pipeline when you want the same analysis with no LLM in the loop