Meeting Memory Agent: long-term memory with Hindsight and Groq for meeting continuity
vyshnavi_nallampati · reddit · 2026-11-07
A project building a meeting assistant with long-term memory, tackling the pain of forgetting decisions and commitments days after a meeting.
Three-layer architecture:
- Hindsight as the memory layer: meeting notes stored via the Retain API ("John promised the presentation by Thursday", "we chose PostgreSQL"), recalled via the Recall API;
- Groq as the intelligence layer: retrieved context is fed to an LLM to understand questions and generate context-aware answers;
- Chat interface as the interaction layer: users type notes and questions naturally.
Workflow: input → Retain → Recall → Groq → respond. Ask "prep me for my meeting with John" and it recalls his commitments, deadlines, and pending follow-ups. The authors see extensions for students, developers, and teams — a step toward AI that understands history, not just answers questions.
More from coding & agent
- Lattice: open-source local-first LaTeX workspace that wires in Claude Code and Codex — aigclink · 2026-11-07
- Dev warns: every SOTA model drop devalues your software engineering skills — zack_overflow · 2026-11-07
- TLAPS-Bench: agentic AI formal verification uncovers read consistency bug in SlateDB's WAL — tianyin_xu · 2026-11-07
- "Bring your own agent": Vincent Koc says team tooling UIs are just another distraction layer — vincent_koc · 2026-11-07
- Ex-Dropbox engineer ditched Linear & Notion: a single synced .md file runs his whole agent workflow — vincent_koc · 2026-11-07
- Doc Reviewer's DIY AI Agent Went Viral at Work — Now He Fears Being Blamed for Its Errors — Southern-Ad6451 · 2026-11-07