memory-happens: an MCP tool that recalls from existing traces instead of a second memory ledger
terra-viii · reddit · 2026-10-13
memory-happens, an experimental MCP tool maintained by a Codex agent (Astra) working with humans and Claude agents, argues you shouldn't need to know in advance what your AI will need to remember — or maintain a second account of your work just for memory.
- Idea: old conversations, handover notes, and Git commits are already memory traces. Connect supported notes/repos or export Claude/Codex session logs, then search when a question arises weeks later.
- Verifiability: originals remain the reference; findings carry reopenable addresses, summaries are checkable interpretations.
- Plugin-based recall: read source → select passages → expand context → optional model processing. Word search and the built-in trigram selector need no model or vector DB.
- Status: beta on Windows/Python 3.12; word search can miss paraphrases. Example sessions and a guide for connecting your own sources are provided.
More from coding & agent
- Constraint-context prompting: telling the LLM what NOT to do yields actionable output — JafarNajafov · 2026-10-13
- Reference-context prompting: paste examples and name the patterns for precise output — JafarNajafov · 2026-10-13
- Persona + expertise prompting turns generic reviews into senior-engineer code audits — JafarNajafov · 2026-10-13
- Context engineering, not prompt engineering: 8 techniques for 10x LLM output — JafarNajafov · 2026-10-13
- One-Person Company Runs 62 Grok Agents, 29 Scheduled Routines Write Its Newsletter — elonmusk · 2026-10-13
- Socratic prompting: stack questions before answers for strategy-level AI output — bigaiguy · 2026-10-13