Omem
Shared Memory That Never Forgets — persistent memory for AI agents with Space-based sharing across agents and teams. Plugins for OpenCode, Claude Code, OpenClaw, MCP Server.
Last verified:
What is Omem?
ourmem (omem) is a shared persistent memory layer for AI agents that enables collective intelligence across sessions, devices, agents, and teams. It solves the problem of AI agents having amnesia—where every session starts from zero—and working in isolation without sharing knowledge. The tool provides one API key that reconnects agents to their memory from any machine, allowing preferences, decisions, and context to persist seamlessly.
Key features include three-tier Space architecture (Personal, Team, Organization) for knowledge flow with full provenance tracking, a Weibull decay model that manages memory lifecycle automatically so core memories persist while peripheral ones gracefully fade, and an 11-stage hybrid retrieval pipeline combining vector search, BM25, RRF fusion, cross-encoder reranking, and MMR diversity for precise recall. The system supports 7-decision reconciliation (CREATE, MERGE, SUPERSEDE, SUPPORT, CONTEXTUALIZE, CONTRADICT, SKIP) that makes memories smarter over time through smart evolution.
ourmem is designed for AI coding tool users (OpenCode, Claude Code, OpenClaw, MCP Server/Cursor) who want their agents to remember past context, and for developers building AI products who need a REST API with 48+ endpoints to embed persistent memory into their own agents and workflows. It supports multi-modal file processing including PDF, image OCR, video transcription, and code AST chunking, plus a GitHub connector for real-time webhook sync.
Omem pricing
Pricing model: Freemium
Open Source: $0 forever for self-hosted on your own infrastructure, includes all memory types, core integrations, self-hosted deployment, and community support. Hosted option available at ourmem.ai with nothing to deploy. Self-deploy costs approximately $5/month for server hosting. The GitHub repository indicates Apache-2.0 open source license. Enterprise deployment with on-premise/VPC, custom SLAs, 24/7 support, SSO, and advanced security requires custom pricing.
Omem pros
- Persistent memory across all AI agent sessions
- Shared memory across multiple agents and teams
- Works across devices with one API key
- Three-tier Space architecture (Personal/Team/Organization)
- Full provenance tracking for all shared memories
- 11-stage hybrid retrieval for precise semantic search
- Weibull decay model auto-manages memory lifecycle
- 7-decision reconciliation makes memories smarter over time
- Smart auto-capture on session end and auto-recall on session start
- Supports 4 platforms: OpenCode, Claude Code, OpenClaw, MCP Server
- Open source Apache-2.0 licensed
- Self-hostable with Docker one-liner (~$5/month)
- Multi-modal support: PDF, image OCR, video transcription, code AST
- GitHub connector for real-time webhook sync
- Quality-gated auto-sharing with customizable rules
- Privacy protection with <private> tag redaction
- 15 MCP tools + 1 resource exposed
- User profile auto-generation with static facts and dynamic context
- Cross-space search across all accessible spaces
- Idempotent sharing prevents duplicate memories
Omem cons
- Musl static build excludes AWS Bedrock support
- Self-hosting requires AWS credentials for Bedrock embedding
- Primarily focused on AI coding agents rather than general use
- No native mobile app available
- Documentation scattered across multiple markdown files
- Enterprise features require custom pricing negotiation
- S3 object storage configuration is optional but recommended for production
- Learning curve for understanding Space architecture concepts
- Rust-based may require compilation for custom modifications
- Hosted option at ourmem.ai has limited public pricing information
Frequently asked questions about Omem
What is ourmem and what problem does it solve?
ourmem gives AI agents shared persistent memory across sessions, devices, agents, and teams. It solves the problem of AI agents having amnesia (every session starts from zero), working in silos (Coder agent can't access what Writer agent learned), local lock-in (memory tied to one machine), no sharing (team agents can't share knowledge), and dumb recall (keyword match only without semantic understanding).
How do I install ourmem for my AI coding tool?
For OpenClaw, run 'openclaw skills install ourmem' then tell your agent to 'setup ourmem in self-hosted mode'. For Claude Code, OpenCode, or Cursor/VS Code, read https://ourmem.ai/SKILL.md and follow instructions to install and configure. One message to your AI agent handles everything—API key, plugin install, config, and verification.
What is the Space architecture?
Three-tier Spaces enable knowledge flow across agents and teams: Personal (one user, multiple agents, owner's agents only), Team (multiple users, team members access), and Organization (company-wide, all org members read-only). Examples include Coder + Writer sharing preferences in Personal, backend team sharing architecture decisions in Team, and tech standards/security policies in Organization.
How does memory lifecycle management work?
The Weibull decay model manages the memory lifecycle with tier-specific β values (Core=0.8, Working=1.0, Peripheral=1.3). Core memories persist while peripheral ones gracefully fade. Three-tier promotion moves memories between Peripheral, Working, and Core based on access. Auto-forgetting detects TTL for time-sensitive info like 'tomorrow' or 'next week'.
What is the 11-stage hybrid retrieval pipeline?
The retrieval pipeline combines: vector search + BM25 → RRF fusion → cross-encoder reranker → Weibull decay boost → importance scoring → MMR diversity → parallel cross-space aggregation. This provides precise semantic recall with user explainability showing per-stage input/output/score/duration.
How do I self-host ourmem?
Use Docker one-liner: 'docker run -d -p 8080:8080 ghcr.io/ourmem/omem-server:latest'. For Bedrock embedding (recommended), add '-e OMEM_EMBED_PROVIDER=bedrock -e AWS_REGION=us-east-1'. For OpenAI-compatible embedding, use '-e OMEM_EMBED_PROVIDER=openai-compatible -e OMEM_EMBED_API_KEY=sk-xxx'. Full guide at docs/DEPLOY.md.
What memory tools are available to my agent?
Memory tools include: memory_store (save facts/decisions/preferences), memory_search (11-stage hybrid search), memory_get (retrieve by ID), memory_update (modify content/tags/importance/tier), memory_delete (remove memory), memory_list (browse with filters/pagination), memory_ingest (smart-ingest conversations), memory_profile (auto-generated user profile), and memory_stats (analytics).
How does sharing work across users?
To share memories, pass another user's API Key as target_user. The system automatically creates a bridging Team Space between you with no manual space management needed. Convenience APIs include share-to-user (one-step cross-user share with auto-bridging) and share-all-to-user (bulk share). Shared copies include source vector embeddings for full searchability.
What file formats does ourmem support?
Multi-modal file processing supports PDF, image OCR, video transcription, and code AST chunking. There's also a GitHub connector for real-time webhook sync for code, issues, and PRs. Original text is preserved with dual-path search (vector + BM25 on source text) for content fidelity.
How is security and privacy handled?
Security includes Rust memory safety (no garbage collector, no data races), tenant isolation with X-API-Key auth and query-level tenant filtering, privacy protection with <private> tag redaction before storage, HTTPS for all API transit, server-side encryption at rest on S3, 5-dimension admission control scoring gate, and open source auditable Apache-2.0 licensing.