Skip to content

fugacio.copilot

The chemical-engineering design copilot (depends on fugacio.sim): it exposes the differentiable engine to a language model through a JSON tool registry, a vendor-neutral provider layer, agent loops, and human-readable reports.

from fugacio.copilot import default_registry, run_llm_agent, MockProvider

Chemical-engineering design copilot for Fugacio (depends on fugacio.sim).

The copilot exposes the differentiable engine to a language-model design agent through:

  • a tool registry (default_registry, tool_schemas, call_tool) of deterministic, JSON-in/JSON-out engineering calculations spanning properties, unit operations, distillation, reactors, optimization, design specs, and economics;
  • a vendor-neutral LLM provider layer (OpenAIProvider, AnthropicProvider, and the test MockProvider) behind the optional llm extra;
  • a model-agnostic agent loop (run_agent) plus a real multi-turn function-calling loop (run_llm_agent), with planner adapters (llm_planner, heuristic_planner);
  • human-readable reports (summarize_bubble_point, and the richer markdown summaries in fugacio.copilot.report).

Where to look next

Area Page Key symbols
Tool registry Tool registry ToolSpec, default_registry, tool_schemas, call_tool
Agent loops Agent loops run_agent, run_llm_agent, llm_planner, AgentResult
LLM providers LLM providers LLMProvider, OpenAIProvider, AnthropicProvider, MockProvider
Reporting Reporting summarize_optimization, summarize_economics, stream_table

See the AI design copilot section of the optimization guide for an end-to-end walkthrough.