Align the Python quickstart with the current backend and web client flow

[digitallysavvy]

This PR agent-quickstart-python to match the current expected runtime shape for a Python backend plus React frontend. The backend now uses an async FastAPI path with requester-scoped agent sessions and stateless-safe stop behavior. The web client now uses the shared conversation UI flow, a StrictMode-safe RTC lifecycle, correct transcript normalization, and token renewal. The setup path was also fixed so bun install from the repo root actually manages the frontend package.

What Changed

• Backend
  • Switched the FastAPI service to AsyncAgora
  • Converted request handlers to async def
  • Scoped agent sessions to the requesting user instead of remote_uids=["*"]
  • Made stop idempotent with a fallback to client.stop_agent(...)
  • Reduced token lifetime to 1 hour so the client can renew RTC and RTM tokens
  • Aligned the default agent config more closely with the current reference behavior
• Frontend
  • Replaced the custom shell with the shared conversation UI primitives
  • Moved RTC client creation to a useRef pattern to avoid StrictMode double creation
  • Normalized transcript UIDs instead of treating every nonzero speaker as the agent
  • Added token renewal handling for RTC and RTM
  • Updated frontend package versions and Tailwind scanning for the shared UI package
  • Removed obsolete custom transcript/log/store code
• Setup and docs
  • Added the missing server-python/.env.example
  • Fixed setup instructions to point at the real env template
  • Updated architecture docs to match the implemented backend and frontend behavior
  • Converted the repo root to a Bun workspace so root bun install manages web-client
  • Removed the conflicting nested web-client/bun.lock

Why

The Python quickstart had drifted from the current implementation expectations in a few important ways:

• agent sessions were not requester-scoped
• the frontend UI and lifecycle differed from the shared conversation flow
• token handling differed from the renewal-based session model
• setup instructions referenced missing files
• root bun install did not actually manage frontend dependencies

This change closes those gaps and makes the repo’s setup, runtime behavior, and docs consistent with each other.

Validation

• python3 -m py_compile server-python/src/server.py server-python/src/agent.py
• bun install from the repo root after converting to a workspace

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
agent-quickstart-python-mason	Error		Apr 25, 2026 3:04am