OmniMem

Server health: Valkey connection, index counts, model status, uptime.

Return the current OmniMem version.

Store a memory with automatic dedup. Returns duplicate info if near-match exists; use force=True to override.

Index a long-form document by splitting it into chunks and storing each chunk as a memory.

Search memories by semantic similarity. Returns ranked results; abandoned-approach warnings appear first.

Lightweight recall: returns ranked summaries without full content. Use recall_detail() to fetch full content for selected keys.

Fetch full content for specific memory keys. Use after recall_index() to expand only the entries you need.

Reduce a memory's visibility without deleting. Accepts a key or natural language query.

Archive a memory — excluded from recall but still stored.

Reinstate a deprioritised/archived memory to active state.

Replace or adjust the tags on a memory without re-embedding it.

Permanently delete a memory. Requires confirm=True; returns preview otherwise.

Suppress a topic — matching memories filtered from recall.

Remove a topic from the suppression list.

List all suppressed topics.

Scan a namespace for clusters of near-identical memories.

Create or update a project's context (description, stack, goals, state).

Retrieve full context for a project by name.

List all stored project contexts, deduplicated by project name.

Update a project's current state and notes without re-embedding.

Gather all stored memories for a project and compile them into a structured context draft. Use this before set_project_context() to auto-produce or refresh a project's context from its episodic memories, experience data, and abandoned approaches.

Suggest work-type domains for a project from its stack and its own memories, with the evidence behind each one. Returns a draft by default; pass auto_save=True to store it.

Bulk delete every memory belonging to a project. Requires confirm=True; returns a preview otherwise.

Bulk deprioritise every active memory in a project (0.2x recall visibility, reversible). Requires confirm=True; returns a preview otherwise.

Bulk reinstate every deprioritised or archived memory in a project back to active. Requires confirm=True; returns a preview otherwise. The inverse of deprioritise_project().

Summary of all memories grouped by state. Useful for cleanup.

Explain why a topic surfaced by searching recall logs.

Return full metadata for a memory key.

Drop and recreate Valkey search indexes to clear orphaned vector entries.

Record effort, outcome, dead ends, and breakthroughs for a memory. High-effort successes surface more; high-effort failures (>=4, abandoned) auto-suppress the approach names.

Append a dead-end approach to a memory's abandoned list.

Return experience data for a memory key.

Aggregate experience stats: effort, outcomes, graveyard of abandoned approaches, breakthroughs.

Check if an approach was previously abandoned. Call before suggesting libraries or tools.

Scan for contradictions. Tier 1 (default): fast heuristic. Tier 2 (use_api=True): Claude API verification.

Compile domain procedure from experience and graveyard memories into a loadable SKILL.md. 'propose' (default) returns a reviewable diff; 'write' commits only a previously proposed and accepted diff — there is no silent-commit path.

Discover compiled skills: ranked skill IDs and descriptions for a query or domain. Load the winner intact with get_skill().

Load a skill whole: the complete SKILL.md body with frontmatter and structure intact, by ID, name, or domain.

Promote a single strong lesson to skill-eligible now, bypassing the reinforcement threshold at the next compile_skill().

Session-start briefing: project context, experience summary, stale memories, knowledge, contradictions, reinstate candidates.

Recent knowledge articles ingested by the RSS worker.

Mark a knowledge item as permanently useful, and optionally skill-eligible for a domain.

Record the redistribution rights of one or more memories.

Record where one or more memories came from.

Check the enrichment queue. Returns the number of pending jobs waiting for background fact extraction.

Export all memories to a JSON backup file.

Restore memories from backup. Default dry_run=True previews without writing. Merges with existing data.

List available backup files, newest first.