Built for Real Engineering Workflows
Offline-first, cross-device memory layer for AI coding agents. Hybrid FTS5 + sqlite-vec search, cursor-based sync, team namespaces.
Zero-config memory
Install in 30 seconds. Your agent starts remembering immediately.
Sign up & install
Create an account, paste the one-liner. Engrm writes your config, initialises the local database, and connects to your MCP agent.
Agent captures observations
As you work, Engrm hooks into your coding agent and automatically extracts discoveries, decisions, bugfixes, and patterns. No manual tagging needed.
Search & recall
Your agent queries memory at the start of each session. Hybrid search (FTS5 keyword + sqlite-vec semantic) finds relevant context instantly — online or offline.
Sync across devices & team
Observations push to Candengo Vector and pull to every connected device. Your whole team shares project knowledge automatically.
SQLite is source of truth · Sync outbox queues for push · Hybrid FTS5 + vec search
Built for real workflows
Offline-first, privacy-aware, and designed to stay out of your way.
Offline semantic search
sqlite-vec with all-MiniLM-L6-v2 embeddings. Full semantic search runs locally — no network required. Works on a plane.
Cross-device sync
Cursor-based change feed syncs observations to every device in your team. Conflict-free, idempotent, crash-safe.
Multi-agent support
Works with Claude Code, Cursor, Codex CLI, Windsurf, Cline, and any MCP-compatible agent. Same memory, any tool.
Privacy by default
Automatic secret scrubbing catches API keys, tokens, and credentials before they're stored. Sensitivity levels control what syncs.
Lifecycle management
Observations age, compact into digests, and eventually purge. No manual cleanup — your memory stays relevant automatically.
Team knowledge
Shared namespace means every team member's agent benefits from everyone's discoveries. Personal observations stay private.
Works with your agent
One MCP server. Every coding agent.
Built different
How Engrm compares to alternatives in the AI agent memory space.
| Engrm | claude-mem | mem0 | Cognee | |
|---|---|---|---|---|
| Local search | < 50ms | ~100ms | ~200ms+ | ~500ms+ |
| Offline support | ✓ | ✓ | ✕ | ✕ |
| Cross-device sync | ✓ | ✕ | Cloud only | ✕ |
| Team sharing | ✓ | ✕ | Limited | ✕ |
| Multi-agent (MCP) | ✓ | Claude only | API | Limited |
| Real-time audit | Sentinel | ✕ | ✕ | ✕ |
| Self-hosted | ✓ | Local only | ✕ | ✓ |
| Secret scrubbing | ✓ | ✕ | ✕ | ✕ |
| Semantic + keyword | Hybrid (RRF) | Vector only | Proprietary | Graph |
| Free tier | 10K obs, 2 devices | Local only | 1K calls/mo | Open source |
Why not...?
We get asked this a lot. Here's the honest answer.
Why not claude-mem?
Single-machine only, no cross-device sync, no team sharing, Claude-only. AGPL license restricts commercial use. Engrm is offline-first with full sync, any MCP agent, and Fair Source licensed (FSL-1.1-ALv2).
Why not mem0?
Cloud-only with no offline support. Proprietary search you can't self-host. Limited team features. Engrm works on a plane and lets you own your data.
Why not CLAUDE.md files?
Static, manual, one file per project. No search, no cross-device, no team sync. Engrm captures automatically and surfaces the right context at the right time.
Why not CodeRabbit / Qodo?
They review PRs after code is written. Sentinel blocks violations before they happen. They're stateless; Engrm remembers your team's past decisions and learns from every review.