Memory gate
Decide which messages in an agent's chat are worth remembering next session, and which change what it already knows.
Run it on each message in a coding or support agent's conversation. Jev decides whether the message holds something a later session needs (a decision, a standing rule, a bug and its fix, an approach that failed, a preference, agreed future work), whether it will still hold next week, and whether memory already has it or it replaces an old line. Your agent's memory file stays short and current, without asking an LLM to summarize every turn.
Try it live
Real calls to Jev. Pick an example or paste your own.
Pick an example or paste your own input, then run it.
Under the hood
Every question and rule, in the open. Fork it to change any of them.
What it asks
Reads message and memory, then asks 4 questions in one call.
- Q1Worth keepingPick one
What does the
messagecontain that a later session would need?A decisionA ruleA bugA dead endA preferenceLater workNothing - Q2Still true next weekYes / no
Will what the
messagesays still be true and useful next week, beyond the task at hand? - Q3Already savedYes / no
Does the
memoryalready say the same thing as themessage? With nomemory, answer no. - Q4Changes a saved lineYes / no
Does the
messagechange or reverse something thememorysays? With nomemory, answer no.
How it decides
Rules run in code, top to bottom. The first one that matches gives the verdict.
- 1If Worth keeping is NothingSkip
- 2If Changes a saved line ≥ 0.6Replace a line
- 3If Already saved ≥ 0.6Skip
- 4If Still true next week ≥ 0.5Save it
- ·DefaultSkip
Call it from your code
Every template runs as-is with an API key, no setup: POST /api/v1/judges/memory-gate/run.
curl -X POST https://tryevally.com/api/v1/judges/memory-gate/run \ -H "Authorization: Bearer $EVALLY_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "input": { "message": "Let'\''s move the primary store from SQLite to Postgres 16. The sync worker needs concurrent writes.", "memory": "- [decision] Use SQLite as the primary store for now.\n- [rule] Node 20 is the minimum supported version." }}'Seen in the wild
Community builds on Jev that use this pattern. Not affiliated with Evally.
More in Agents
- Agents
Tool-call firewall
Judge every agent tool call before it touches your machine: allow, ask a human, or deny.
AllowAsk a humanDeny10 questions - Agents
Is the task done?
Don't trust an agent's "done". Check the transcript for proof, open errors and skipped work.
DoneDone, but unprovenNot done5 questions - Agents
Agent loop & drift detector
Spot an agent that's stuck in a loop, drifting off-goal, or blocked, and stop wasting tokens.
ContinueNudge: change approachStop and escalate4 questions - Agents
Model router
Send each prompt to the cheapest model that can handle it.
Small modelMid-size modelFrontier model+14 questions - Agents
Tool router
Pick the right tool for a request without an LLM tool-calling round trip.
Ask to clarifyConfirm firstWeb search+62 questions - Agents
Context relevance filter
Decide what an agent keeps in context: keep, compress or drop each chunk.
KeepCompressDrop3 questions