Start here
Memory & Intelligence Layer · v7.3.2 · Local-First · Free to Use
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.
100% / 100% recall / precision firewall·Local-first · nothing egresses·Measured, not claimed
Or install the CLI: npm install -g wyrm-mcp
New here? 5-minute Getting Started
Works with
Claude Desktop·GitHub Copilot·Cursor·Windsurf·Any MCP Client
The Problem
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.
The Solution
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. Free tier (account required). See how Wyrm compares to other AI-memory tools or read the docs.
Projects & sessions that persist forever
Skills & patterns your AI learns once
Quests with priorities, tags, and status tracking
Data lake for any structured data
Full-text search across everything
32-verb typed surface: 137 tools, all callable
Knowledge graph with entity relationships
Memory artifacts recalled by confidence + relevance
Ground truths & scaffolds for every context build
Unified capture: one tool classifies anything
Session prime: full context in one call
Terminal CLI: browse memory without leaving your shell
Import from git history, PRs, and rules files
Staleness scoring: stale truths flagged automatically
Encrypted snapshots: sync across devices securely
Memory decay + prune: keeps your knowledge fresh
Install via npm, Homebrew, or Docker
Visual dashboard: memories, quests, truths, review queue in browser
Run-attributed fleet memory: every write tagged (agent_id, run_id)
Fleet negative learning: one agent's failure blocks the repeat for all
Portable skills: SKILL.md content synced across every machine
Benchmarked firewall: 100% recall / 100% precision on repeated mistakes
wyrm render: memory compiled into CLAUDE.md/AGENTS.md at zero MCP tokens
Install
$ npm install -g wyrm-mcpclaude_desktop_config.json
{
"mcpServers": {
"wyrm": {
"command": "wyrm-mcp"
}
}
}Up and running in under 60 seconds. Full setup guide →
Features
Everything Your AI Needs to Remember
A frozen 32-verb typed surface over 137 tools, covering run-attributed fleet memory, fleet negative learning, counter-pattern blocking, the OODA agent loop, knowledge graph, Live Memory + cross-device replication, hybrid recall, an auto-extraction review queue, the Wyrm Memory Protocol, and a live statusline daemon, designed for real-world AI workflows.
Persistent Memory
Projects, sessions, quests, skills: everything stored permanently across sessions.
Full-Text Search
FTS5-powered instant search across all stored data. Find anything in milliseconds.
Data Lake
Store and query any structured data with category-based organization.
Encryption
Optional AES-256-GCM encryption for sensitive data at rest. Per-install salt, no shared keys.
Analytics
Token usage tracking, cost estimation, and session metrics.
Zero Config
npm install, add to your MCP config, done. Under 60 seconds.
Knowledge Graph
Entity/relationship graph with neighborhood traversal, path finding, and merge: build structured memory.
Memory Artifacts
Store lessons, patterns, anti-patterns with confidence scoring. Recalled automatically on context build.
Intelligence Amplification
Ground truths + reasoning scaffolds injected into every context brief. AI that knows your project deeply.
Unified Capture
One tool classifies anything: task, truth, lesson, pattern. Confidence-gated so only high-signal data persists.
Session Prime
One call loads truths + scaffolds + recent memories + open quests. AI starts every session fully briefed.
Terminal CLI
wyrm search, ls, show, capture, import, prune, sync: browse and manage your memory directly from the terminal.
Staleness Detection
Ground truths with TTLs get staleness scores. Session prime flags outdated facts before they mislead your AI.
Encrypted Sync
Export your entire memory as an AES-256-GCM snapshot. Restore on any machine. Your data, fully portable.
Visual Dashboard
Run wyrm serve --ui to open a live browser dashboard: memories, quests kanban, truth staleness, and review queue.
Security Hardened
DEF CON-level pentest audit. SQL injection prevention, no error leakage, path traversal guards, rate limiting, and graceful shutdown.
Fleet Memory
Run-attributed memory: every write is tagged with the (agent_id, run_id) that produced it. Spin up a fleet of agents and they share one accountable memory bus.
Fleet Negative Learning
A failure recorded by one agent blocks the repeat for every sibling. The wyrm-guard PreToolUse hook enforces it before the doomed call ever runs.
Frozen 32-Verb Surface
A stable, typed verb surface (structuredContent + tool annotations) that an orchestrator can pre-check. 32 advertised verbs; all 150 legacy tools still callable.
Measured Token Economy
Real, benchmarked savings: −9% on a default working hour, up to −49% with opt-in WYRM_CHANNEL, and −39% on fleet runs. Memory that pays for its own context.
Portable Skills
SKILL.md content syncs end-to-end across every machine. Teach a skill once; the whole fleet inherits it, everywhere.
Benchmarked Firewall
Negative learning, measured: 100% recall / 100% precision blocking repeated mistakes, ~0.1 ms, deterministic. No recall-only memory can do this, and the benchmark ships in the repo.
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.
Pricing
Free Forever. Scale When Ready.
Local use is free with no limits. Paid plans add cloud sync, encryption, and team collaboration.
Free
Free
For individual developers
Local SQLite storage
All 137 tools · 32-verb surface
Fleet memory + negative learning
Live Memory + hybrid recall
Cross-project memory
Full-text + semantic search
Unlimited projects
Community support
Most Popular
Pro
$29/month
For power users
Everything in Free
Cloud sync across devices
AES-256 encryption
Priority support
Advanced analytics
Team
$199/month
For development teams
Everything in Pro
Shared team memory
Up to 25 seats
Admin dashboard
Slack integration
Enterprise
$499/month
For organizations
Everything in Team
Unlimited seats
SSO / SAML
Custom SLA
Dedicated support
On-premise option
Pro, Team, and Enterprise launching soon.
Early Access
Get Notified When Pro Launches
Early access, priority support, and founding member pricing. No spam, just one email when we're ready.
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.
Changelog
v7.3.2Latest
Receipts, the benchmarked firewall
The receipts behind the intelligence-layer claims, all measured on real LoCoMo and reproducible from the repo. Negative-learning benchmark: 100% recall / 100% precision blocking a repeated mistake (0 false blocks on novel actions), deterministic, ~0.1 ms, something no recall-only memory can do. A stopword fix lifts reworded-repeat recall 6.3% → 12.5% with precision held at 100%. The opt-in cross-encoder reranker lifts recall@1 33% → 53% (+19 pts). Plus an honest competitor comparison (COMPARISON.md, shipped in the npm tarball), retrieval recall@k vs the field's LLM-judged QA accuracy, never conflated, every figure cited.
v7.1.0
F4, render target + context economy
The BROOD arc completes. wyrm render deterministically compiles your memory straight into CLAUDE.md / AGENTS.md / Cursor / Copilot slots (hard 200-line budget, byte-stable, never silently overwrites a human edit), casual sessions cost ~zero MCP tokens. wyrm:// resources return big payloads as links, not inlined bytes. The reverse bridge harvests your manual edits back through the review queue. Retrieval ships a published two-tier benchmark with a CI-gated no-LLM floor. 104 suites / 1,732 tests green.