●

LangGraph Agent / Routed Through Fire

Question Decomposer → Research Agent → Synthesizer, every call via Fire’s /v1/chat

checking Fire...

Decomposer Bot

Breaks your question into focused sub-questions

idle
↓

Answer Bot

Researches and answers each sub-question

idle
↓

Final Answer

Synthesized from all research steps

idle

How It Works

Browser
→
Web App (FastAPI)
→
LangGraph
→
Fire LLM Wrapper
→
Fire /v1/chat
→
LLM
  • LangGraph manages the workflow: decompose → answer → synthesize.
  • Fire provides every LLM call — no direct OpenAI, Anthropic, or Gemini calls exist in this app.
  • A custom Fire wrapper (FireChatModel) routes all model requests through /v1/chat.
  • Secrets stay server-side: FIRE_API_KEY lives only in a local .env file, never sent to the browser.