Decide what an agent should remember and recall
Ask Jev whether each message is worth saving to agent memory, and which saved memories matter for the next task.
Author's figure: Author's figure: 0.3 s per memory decision (Jetwani Avinash). Reported by the builder, not measured by us.
The use case
Teams giving agents long-term memory find the memory file grows forever and gets loaded into every run. These builds ask Jev one question per message, whether it should be kept, and at recall time ask which stored memories matter for the task at hand, so only those are passed in. One lab tool lets you edit test cases to see what gets admitted. A developer reviews the memory store regularly and deletes anything personal or wrong, since a bad memory repeats its mistake on every run.
Example workflow
- Hook into the agent's message stream in Claude Code or Hermes
- Ask Jev a Noul question on whether each message is worth storing
- Save admitted items to a memory store
- At recall, ask Jev to choose relevant memories for the current task
- Review and prune the store monthly
Tools
- Jev (TypeSafe AI)
- Claude Code
- SQLite
Risks and guardrails
- Personal data ends up in long-lived memory
- Wrong memories get reused repeatedly
- Recall misses lead to repeated questions
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- A memory gate for Claude Code by Jetwani Avinash on Made with Jev: madewithjev.com/builds/claude-code-memory-gate
- A memory gate for Claude Code: original post or repo by Jetwani Avinash: x.com/jetwaniavinash/status/2103088357077328290
- perfectrecall by Arslan R. on Made with Jev: madewithjev.com/builds/perfectrecall
- perfectrecall: original post or repo by Arslan R.: github.com/arslanr-com/perfectrecall
- cairn-jev-lab by Cairn-ink on Made with Jev: madewithjev.com/builds/cairn-jev-lab
- cairn-jev-lab: original post or repo by Cairn-ink: github.com/Cairn-ink/cairn-jev-lab