Agent memory maintenance (admin)
nightly or on-demand tidy of each user’s agents’ personal memory (dedupe, expired daily notes). Not the same as memory forge (scan conversations → improve help/workspace knowledge
Source help/en/admin-ops/agent-memory-maintain.md
Audience: platform admins (admin console Agent memory maintenance).
In user language: nightly or on-demand tidy of each user’s agents’ personal memory (dedupe, expired daily notes). Not the same as memory forge (scan conversations → improve help/workspace knowledge overlay).
What you do in the admin console
- Sign in to the admin console → Overview → Open memory maintenance (route
/agent-memory-maintain). - View config summary: whether auto-tidy is on, nightly hour, time zone, next planned time.
- Tap Tidy now: run one round on agents that have memory under the runtime directory (do not wait for the night window).
- In Run history, open a run to see what this run changed (delete type, agent, path, reason).
Users can also view this assistant’s tidy log on each agent’s Memory page, and run “Tidy my memory” by hand. See Agent memory.
vs memory forge
| | Agent memory maintenance | Memory forge | |--|-------------------|--------------| | Serves | One agent’s personal memory | Help / workspace knowledge overlay and skills | | Typical action | Delete duplicates, expired daily notes | Scan archive → proposal → human accept new supplements | | Admin entry | /agent-memory-maintain | /memoryforge | | Changes original knowledge? | No | No (writes overlay only) |
When it auto-runs
All required:
schedule.enabledis trueagent_memory_maintain.enabledis true
Trigger time: local night in RuntimeTimezone (or MINDLINK_RUNTIME_TIMEZONE, default Asia/Shanghai), once a day in about a 2-hour window from daily_local_hour (default 2).
Config (mindlink.json)
"agent_memory_maintain": {
"enabled": true,
"daily_local_hour": 2,
"daily_note_retain_days": 14,
"stale_review_days": 90,
"min_entries_for_llm": 5,
"max_agents_per_run": 30,
"max_deletes_per_agent": 8,
"use_llm": true
}
Field notes: Server configuration § agent_memory_maintain.
Related APIs (admin)
| Method | Path | Notes |
|---|---|---|
GET | /api/v1/admin/agent-memory-maintain/status | Config summary + last result |
GET | /api/v1/admin/agent-memory-maintain/history | Run history (new → old) |
POST | /api/v1/admin/agent-memory-maintain/run | Tidy one round now |
On-disk locations (ops)
| Path | Content |
|---|---|
{runtime_dir}/.memory_maintain_status.json | Last global status |
{runtime_dir}/.memory_maintain_history.jsonl | Global run history |
{runtime_dir}/{agent-id}/memory/.maintain/log.jsonl | Change history related to that agent |
Related docs
- User help: Agent memory
- Mechanism: docs/core-mechanisms/智能体记忆.md
- Memory forge: admin
/memoryforge(knowledge feed-back, not personal-memory cleanup)