[Suggestion] Voice agent with cross-session conversation memory using Redis (Python)

[deepgram-robot]

What to build

A voice agent that persists conversation history and user context across multiple sessions using Redis, enabling the agent to remember previous interactions, user preferences, and unresolved issues when a user reconnects. Demonstrates session persistence patterns for production voice agent deployments.

Why this matters

Production voice agents need to remember conversations across sessions — a customer calling back about an unresolved issue shouldn't have to repeat themselves. Developers building persistent voice experiences need a reference for storing and retrieving conversation state efficiently. Redis is the most common choice for session state in production systems, making this pattern immediately applicable to real deployments.

Suggested scope

• Language: Python
• Deepgram APIs: Voice Agent API with Nova-3 STT + Aura TTS
• State store: Redis (with fallback to in-memory for local dev)
• Key patterns: Session ID management, conversation history serialization, user profile persistence, context injection on reconnect, TTL-based session expiry
• Architecture: FastAPI server + Deepgram Voice Agent + Redis session store
• Complexity: Medium — session management, Redis integration, context injection into agent prompts

Acceptance criteria

• Runnable with minimal setup (clone, add API key, run — Redis optional via Docker or in-memory fallback)
• README explains the session persistence architecture
• Uses current SDK version
• Demonstrates conversation continuity: disconnect, reconnect, and agent remembers context
• Shows user preference persistence (e.g., name, previous topics discussed)
• Includes session expiry and cleanup logic

---

Raised by the DX intelligence system.