memory — Tool Reference

Manage memories. Actions: learn (store observation with auto-pattern detection), recall (search memories and skills), forget (soft-delete), correct (update with lineage tracking).

Aliases

Old tool names remember, recall, forget, and correct still work as aliases and route to the corresponding action on this tool.

Parameters

Name	Type	Required	Description
`action`	string	Yes	One of `learn`, `recall`, `forget`, `correct`.
`observation`	string	—	What happened. Used with `learn`.
`domain`	string	—	Domain or project name.
`task_type`	string	—	Type of task. Used with `learn`. Omit for pure memories.
`tools_used`	string[]	—	Tools used. Used with `learn`.
`outcome`	string	—	One of `success`, `failure`, `partial`, `unknown`. Used with `learn`.
`tags`	string[]	—	Tags for categorization.
`query`	string	—	Search query. Used with `recall`.
`memory_id`	string	—	Memory ID. Used with `recall` (by ID), `forget`, `correct`.
`include`	string	—	One of `memories`, `skills`, `all`. Defaults to `memories`. Used with `recall`.
`limit`	number	—	Maximum number of results.
`offset`	number	—	Pagination offset.
`format`	string	—	One of `full`, `compact`.
`skill_detail`	string	—	One of `full`, `summary`, `name_only`. Controls skill detail level in `recall`.
`summarize`	boolean	—	Condense results into a summary. Used with `recall`.
`correction`	string	—	Corrected text. Used with `correct`.
`superseded_by`	string	—	Replacement memory ID. Used with `forget`.

learn

Store a new observation. Kyew automatically detects patterns across related memories over time.

Parameters: observation (required), domain, task_type, tools_used, outcome, tags

{
  "name": "memory",
  "arguments": {
    "action": "learn",
    "observation": "The deploy script requires the --env production flag or it targets staging",
    "domain": "infrastructure",
    "outcome": "failure",
    "tags": ["deploy", "gotcha"]
  }
}

{
  "content": [
    {
      "type": "text",
      "text": "Memory saved (id: mem_a1b2c3). Domain: infrastructure. Pattern detection: 2 related memories found — 1 more before skill generation."
    }
  ]
}

recall

Search memories and optionally skills. Supports full-text search, domain filtering, and pagination.

Parameters: query or memory_id (at least one), domain, include, limit, offset, format, skill_detail, summarize

{
  "name": "memory",
  "arguments": {
    "action": "recall",
    "query": "deploy gotchas",
    "domain": "infrastructure",
    "include": "all",
    "limit": 5
  }
}

{
  "content": [
    {
      "type": "text",
      "text": "Found 3 memories and 1 skill:\n\n1. [mem_a1b2c3] The deploy script requires --env production flag or it targets staging (infrastructure, failure)\n2. [mem_d4e5f6] Always run npm run build:worker before deploying to catch type errors (infrastructure, success)\n3. [mem_g7h8i9] Database migrations must be run separately with db:migrate:remote (infrastructure, success)\n\nSkill: Deploy Checklist (skill_x1y2z3, active) — 3-step pre-deploy verification"
    }
  ]
}

forget

Soft-delete a memory. It stops appearing in searches but is kept in history.

Parameters: memory_id (required), superseded_by

{
  "name": "memory",
  "arguments": {
    "action": "forget",
    "memory_id": "mem_a1b2c3",
    "superseded_by": "mem_j0k1l2"
  }
}

{
  "content": [
    {
      "type": "text",
      "text": "Memory mem_a1b2c3 forgotten. Superseded by mem_j0k1l2."
    }
  ]
}

correct

Update a memory with corrected content. The original is preserved for lineage tracking.

Parameters: memory_id (required), correction (required)

{
  "name": "memory",
  "arguments": {
    "action": "correct",
    "memory_id": "mem_d4e5f6",
    "correction": "Run npm run build:worker AND npm run typecheck before deploying"
  }
}

{
  "content": [
    {
      "type": "text",
      "text": "Memory mem_d4e5f6 corrected. Previous version preserved in lineage."
    }
  ]
}