Jev 101
Start Guides
Search
中文 EN
Agents outlineExperimental — memory text and queries go to Jev; the answer model is billed separately

Agent memory control

Who: Builders of long-running agents with persistent memory. Steps: On write, ask Noul / Choice questions for memory type, redundancy and links to stored memories; on read, ask which memory views to search, how much budget to spend, which candidates to keep and whether the evidence is enough; code enforces the limits and passes the selected evidence to the answer model. Expected effect: fewer language-model calls inside the memory loop.

Source: https://github.com/libingzheren/Jev-Mem ↗ Docs / source ↗
1

Treat this as an outline — adapt state and questions to your data.

2

Implement in code — call System One / Jev; compose answers yourself.

3

Gate on confidence — act, confirm, or escalate before side effects.

Outline sketch

Prompt 
# Agent memory control
State: the new observation or the query, plus short summaries of the related stored memories.
Ask: Choice for memory type and which memory view to search; Noul for redundant, related, keep candidate and evidence sufficient.
Then: code writes links, enforces traversal and budget limits, stops when the evidence is enough, and sends only the selected evidence to the answer model.

Needs access to: locomolaya

Who it's for

Builders of long-running agents with persistent memory

Steps / how it's set up

Steps (experimental)

  1. Keep the original observations with timestamps and provenance.
  2. Ask Jev typed questions on write (type, redundancy, relations) and on read (route, budget, candidate, stop).
  3. Enforce graph and retrieval limits in code, and let a separate model write the answer.

Sources

This pattern is experimental. Results are paper-reported on one benchmark — evaluate on your own agent history.

Expected effect

Moves frequent memory decisions out of the generative loop

Unofficial outline for learning. Paraphrased from public docs and tutorials — not a production recipe. Review sources before you automate anything.