Memory API

List Memory Entries

GET /api/memory?type=semantic&page=1&page_size=20&sort_by=created_at&sort_order=desc&tag=python

Parameter	Type	Default	Description
`type`	string	all	Memory type filter: `semantic`, `episodic`, `procedural`, `conversation`, `task_digest`
`page`	int	1	Page number (1-indexed)
`page_size`	int	20	Items per page (max 100)
`sort_by`	string	`created_at`	Sort field: `created_at`, `updated_at`, `importance`
`sort_order`	string	`desc`	Sort direction: `asc` or `desc`
`tag`	string	—	Filter by tag (case-insensitive)
`search`	string	—	Semantic search query (uses vector similarity)

Response:

{
  "items": [...],
  "total": 91,
  "page": 1,
  "page_size": 20,
  "has_next": true,
  "has_prev": false
}

Get Memory Tags

GET /api/memory/tags?limit=20

Returns the most frequent tags across all memory types with their counts.

Search Memory

GET /api/v1/memory/search?query=user%20preferences%20for%20coding%20style&memory_types=semantic,episodic&limit=10

Parameter	Type	Default	Description
`query`	string	—	Search query (required, uses vector similarity)
`memory_types`	string	all	Comma-separated memory types, e.g. `semantic,episodic`
`limit`	int	10	Maximum results (1-50)
`session_id`	string	—	Optional session ID; ranking signals are tuned to that session’s preferences

Response:

{
  "results": [...],
  "scores": [0.91, 0.84],
  "query": "user preferences for coding style",
  "total": 2
}

results and scores are aligned by index.

Create Memory Entry

POST /api/memory

{
  "type": "semantic",
  "content": "User prefers TypeScript with strict mode",
  "importance": 0.8
}

Delete Memory Entry

DELETE /api/memory/:id

Memory Command Center

Get the full memory system health and diagnostics snapshot for the Settings UI. Memory Labs is experimental and off by default (MYRM_MEMORY_LABS on the server, NEXT_PUBLIC_MEMORY_LABS=1 for the WebUI). The /memory/command-center* routes are mounted only then, except GET /memory/command-center/graph and GET /memory/command-center/evidence/playback, which the default Journey graph tab and memory evidence drawer use and which are always mounted.

GET /api/v1/memory/command-center?project_id=<optional>

Parameter	Type	Default	Description
`project_id`	string	—	Optional project ID to scope the snapshot to a single project’s memory spaces

Response includes: overview statistics, memory spaces, governance items, health diagnostics, timeline events, migration status, runtime configuration, and more. When project_id is provided, the spaces, governance, and timeline sections are filtered to show only data related to that project’s SharedContext bindings. Empty strings are normalized to the global view.

Learning & Evolution Timeline API

Unified memory and skill lifecycle chronological stream with inline governance.

Get Unified Learning Timeline

GET /api/v1/statistics/learning-timeline?days=30&limit=50&kind_filter=fact_memory,skill_evolution&cursor=2026-10-01T08:00:00+00:00|mem_123

Parameter	Type	Default	Description
`days`	int	30	Lookback window in days (1-365)
`agent_id`	string	—	Keep only nodes of this agent (nodes without an agent always stay)
`kind_filter`	string	all	Comma-separated node kinds: `fact_memory`, `preference_memory`, `procedural_memory`, `episodic_memory`, `skill_evolution`, `skill_draft`
`limit`	int	50	Items per page (1-100)
`cursor`	string	—	Opaque `next_cursor` returned by the previous page

Response:

{
  "items": [
    {
      "id": "mem_123",
      "kind": "fact_memory",
      "title": "Factual Knowledge",
      "content": "User prefers strict TypeScript type checking",
      "created_at": "2026-10-01T08:00:00+00:00",
      "importance": 0.9,
      "confidence": 0.95,
      "is_user_protected": false,
      "metadata": { "memory_type": "semantic" }
    }
  ],
  "has_more": true,
  "next_cursor": "2026-10-01T08:00:00+00:00|mem_123",
  "total_count": 42
}

total_count is the number of nodes matching all filters (each source contributes at most its 500 newest records). Skill nodes carry metadata.manageable, metadata.archived and metadata.pinned. An invalid cursor or kind_filter returns 400. is_user_protected is true when the user protects the memory (pinned, or a locked procedural rule). source_chat_id is the conversation a memory was extracted from. Procedural rules carry no importance score, so their importance is a placeholder.

Update Timeline Memory Item

PUT /api/v1/statistics/learning-timeline/memory/:memory_type/:id

:memory_type is the node’s metadata.memory_type (semantic, episodic or procedural).

{
  "content": "Updated rule content",
  "importance": 0.9,
  "reasoning": "Explicit user correction via Timeline Editor"
}

importance is optional and only applied when supplied. An edit protects the memory: procedural rules are locked and semantic/episodic memories are pinned, so automated forgetting, merging and rewriting skip them. Memory items report this protection as is_user_locked: true. Returns 404 when the memory does not exist and 400 for rejected content.

Delete Timeline Memory Item

DELETE /api/v1/statistics/learning-timeline/memory/:id?memory_type=semantic

Semantic and episodic memories are archived (recoverable from the memory trash via POST /api/v1/memory/trash/:id/restore); procedural rules are removed permanently.

Archive / Restore Timeline Skill

POST /api/v1/statistics/learning-timeline/skill/:skill_name/archive?active=false

active=false archives the skill, active=true restores it. Pinned skills return 409; unknown skills return 404.

{
  "skill_name": "code-reviewer",
  "is_active": false,
  "message": "Skill archived"
}

Memory Command Center: Long-Horizon Economics Profiler

Available only when Memory Labs is enabled.

GET /api/v1/memory/command-center/economics?conversation_id=conv-xxx&window=50

Parameter	Type	Default	Description
`conversation_id`	string	`None`	Optional conversation filter (omitted to evaluate global stateful trajectories)
`window`	int	50	Recent turn horizon window size (max 200)

Response:

{
  "kpis": {
    "total_injected_turns": 42,
    "avg_load_ratio": 0.18,
    "avg_phase_latency_ms": {
      "retrieval": 12.4,
      "assembly": 4.1,
      "injection": 1.8
    },
    "long_horizon_roi": "optimal"
  },
  "turn_trajectories": [
    {
      "turn": 1,
      "retrieval_ms": 14.2,
      "assembly_ms": 3.8,
      "injection_ms": 1.5,
      "load_ratio": 0.12,
      "injected_tokens": 320
    }
  ],
  "parasitic_stale_items": [
    {
      "id": "mem-123",
      "content": "Legacy unreferenced trap rule",
      "created_at": "2026-06-01T00:00:00Z",
      "access_count": 0,
      "estimated_tokens": 150
    }
  ]
}

​Memory API

​List Memory Entries

​Get Memory Tags

​Search Memory

​Create Memory Entry

​Delete Memory Entry

​Memory Command Center

​Learning & Evolution Timeline API

​Get Unified Learning Timeline

​Update Timeline Memory Item

​Delete Timeline Memory Item

​Archive / Restore Timeline Skill

​Memory Command Center: Long-Horizon Economics Profiler

Memory API

List Memory Entries

Get Memory Tags

Search Memory

Create Memory Entry

Delete Memory Entry

Memory Command Center

Learning & Evolution Timeline API

Get Unified Learning Timeline

Update Timeline Memory Item

Delete Timeline Memory Item

Archive / Restore Timeline Skill

Memory Command Center: Long-Horizon Economics Profiler