Skip to content

ARCHITECTURE.md

Latest commit

 

History

History
48 lines (36 loc) · 1.72 KB

ARCHITECTURE.md

File metadata and controls

48 lines (36 loc) · 1.72 KB

Architecture

Overview

Natural-language CAD requests enter through the FastAPI gateway, run through a multi-agent pipeline (PipelineExecutor), and execute FreeCAD Python in a sandboxed headless runner.

Client (HTTP/WS)
    → api/ (routes, auth, rate limits)
    → orchestration/ (StateGraph pipeline)
    → agents/ (Planner → Generator → Validator → Executor)
    → freecad/ (headless_runner, state extraction)
    → sandbox/ (AST validation, safe execution)
    → redis_utils/ (cache, audit, pubsub)

Agents

Agent Role
Planner Decomposes user intent into a task graph
Generator Produces FreeCAD Python per task
Validator Scores geometry and script quality; drives retries
Executor Runs scripts via HeadlessRunner
Orchestrator Coordinates iteration and routing decisions

LLM calls go through llm/provider.py (UnifiedLLMProvider, Opper-backed).

Shared infrastructure

  • Redis — state cache, audit trail, rate limiting, pub/sub
  • WebSocket — realtime/websocket_manager.py for progress events
  • Schemas — Pydantic contracts in schemas/ (API + pipeline)
  • core/ — exceptions, structured logging, Prometheus metrics only

Configuration

Source Purpose
.env Secrets (OPPER_API_KEY, Redis, auth)
config/config.yaml FreeCAD paths, agent model overrides
llm/model_config.py Default per-agent model routing

Deployment

  • Production image: docker/Dockerfile.production → uvicorn ai_designer.api.app:app
  • Dev image: docker/Dockerfile.dev with mounted src/
  • Optional Prometheus profile: docker compose --profile monitoring up