Your AI Forgets Everything. Wyrm Remembers, and Learns From Failure.
The local-first memory and intelligence layer for AI agents. It gives Claude, Cursor and Copilot persistent memory across every session, plus a benchmarked firewall that blocks a known mistake before it repeats. Nothing leaves your machine. Free to use.
Or install the CLI: npm install -g wyrm-mcp
New here? 5-minute Getting Started · Questions or ideas? GitHub Discussions
Works with
Ghost Protocol is a member of the NVIDIA Inception program
Free Forever. Scale When Ready.
Local use is free with no limits. Paid plans add cloud sync, encryption, and team collaboration.
Free
For individual developers
- Local SQLite storage
- All 137 tools · 32-verb surface
- Fleet memory + negative learning
- Write receipts + trust lanes
- Live Memory + hybrid recall
- Cross-project memory
- Full-text + semantic search
- Unlimited projects
- Community support
Pro
For power users
- Everything in Free
- Cloud sync across devices
- AES-256 encryption
- Priority support
- Advanced analytics
Team
For development teams
- Everything in Pro
- Shared team memory
- Up to 25 seats
- Admin dashboard
- Slack integration
Enterprise
For organizations
- Everything in Team
- Unlimited seats
- SSO / SAML
- Custom SLA
- Dedicated support
- On-premise option
Need a custom plan? Contact us →
AI Has Amnesia
Context window limits mean your AI starts fresh every session. You re-explain your codebase, preferences, and decisions over and over. The workarounds (copy-paste, prompt files, manual context) are all fragile.
Wyrm Gives Your AI a Permanent Brain
Wyrm is a Model Context Protocol (MCP) server: the memory and intelligence layer that gives AI agents persistent, searchable memory. Projects, sessions, quests, skills, and arbitrary data, indexed with hybrid full-text and semantic recall and available across every conversation. Local-first and private. Your memory stays on your machine, built on the Wyrm Memory Protocol. The free tier needs no account and no login, and works fully offline. See how Wyrm compares to other AI-memory tools or read the docs.
$ npm install -g wyrm-mcp{
"mcpServers": {
"wyrm": {
"command": "wyrm-mcp"
}
}
}Up and running in under 60 seconds. Full setup guide →
Everything Your AI Needs to Remember
A frozen 32-verb typed surface over 137 tools: run-attributed fleet memory, fleet negative learning, hybrid recall (with optional NVIDIA NIM embeddings and reranking), the knowledge graph, Live Memory replication, provable writes with receipts, provenance trust lanes, and the render target. The load-bearing twelve are below; the full surface is in the docs.
Persistent Memory
Projects, sessions, quests, skills, ground truths: everything kept across every session, local-first.
Hybrid Recall
Full-text (FTS5) fused with semantic vectors and an optional cross-encoder rerank. Local nomic by default; opt into NVIDIA NIM embeddings + reranker for a sharp recall lift, nothing egresses unless you say so.
Benchmarked Firewall
Negative learning, measured: 100% recall / 100% precision blocking a repeated mistake, ~0.1 ms, deterministic. The benchmark ships in the repo.
Fleet Memory
Every write is attributed to the (agent_id, run_id) that produced it, so a swarm of agents shares one accountable memory bus.
Fleet Negative Learning
A failure recorded by one agent blocks the repeat for every sibling, enforced by the wyrm-guard hook before the doomed call runs.
Knowledge Graph
Entity/relationship graph with neighborhood traversal, path finding, and merge. Memory with structure, not just search.
Frozen 32-Verb Surface
A stable, typed verb surface (structuredContent + annotations) an orchestrator can pre-check. 32 advertised verbs; 137 tools all callable.
Portable Skills
SKILL.md content syncs end-to-end across every machine. Teach a skill once; the whole fleet inherits it.
Render Target
wyrm render compiles your memory straight into CLAUDE.md, AGENTS.md, Cursor and Copilot slots: byte-stable, budget-capped. Casual sessions cost zero MCP tokens.
Provable Memory
Every write returns and ledgers a receipt: stored, queued, merged, or dropped, and why. wyrm digest --writes reconstructs any day offline, so nothing is lost silently.
Provenance Trust Lanes
Every memory is marked by origin. Untrusted content is quarantined and categorically withheld from context briefs, so a planted instruction in a transcript can never be read back as a command.
Temporal & Graph Recall
Recall ranks on temporal, usefulness, and knowledge-graph signals beside recency, not just keyword and vector overlap.
Get Updates. Shape Wyrm.
New features, setup tips, and the occasional offer. No noise. Building with Wyrm? Join the dev circle for early access and exclusive perks.
Stop Losing Context. Start Remembering.
Give your AI the memory it deserves. Free to use, local-first, and ready in 60 seconds.
Built by Ghost Protocol, the studio that ships security tools that actually work.
Fleet-safe sync, prime that lands on the right project
The 8.5 line hardens the sovereign cloud fleet: a mis-keyed push is quarantined and skipped instead of wedging every device's sync, login fails loud (naming the fix) when a key cannot read the account, and the daemon keeps its configured interval across restarts. Session prime now resolves the project that owns the server's working directory, so a session on one project can never hijack another project's memory brief.
Injection quarantine hardening
A full garak red-team (622 real jailbreak payloads) fired at the memory-brief surface. Untrusted-origin content is now categorically withheld from context briefs, so a planted instruction in a transcript can never be read back as a command. The detector catches 80.7% of the corpus with zero false positives, and it ships as a CI gate.
PROOF: memory you can prove
The v8 major. Every write now returns and ledgers a receipt (stored, queued, merged, or dropped, and why), and wyrm digest --writes reconstructs any day offline. Harness memory files become compiled views of Wyrm via wyrm bridge, and wyrm metabolize supersedes duplicates and decays stale noise. Recall gained temporal, usefulness, and knowledge-graph legs beside recency, and provenance trust lanes mark and quarantine untrusted content.
NVIDIA NIM retrieval
Optional NVIDIA NIM embeddings + reranker as an explicit opt-in beside the local default. On LoCoMo, NIM embeddings lift recall@1 33% → 47% and the reranker again to 52%; nothing egresses unless you turn it on, and the per-recall receipt names any remote scorer that saw text.