Your AI has amnesia. Fix it.
Persistent memory modeled on the human brain. Security on every LLM call. Expert agents that carry context. All local. Nothing leaves your machine.
Get Started Free GitHubYour architecture decisions, your coding patterns, your preferences — gone. Every session, you re-explain. Every session, your AI starts as a stranger. Context windows aren't memory. They're a junk drawer that empties when you close the tab.
GreyMatter fixes this.
GreyMatter doesn't just store text in a database. It models memory the way your brain does.
Spaced repetition (FSRS-6): Important knowledge gets reinforced. Stale knowledge decays naturally — following the same power-law forgetting curves neuroscientist Hermann Ebbinghaus documented in 1885. Your AI's memories age like yours do. The important ones get stronger.
Five memory types: Observations (raw notes), patterns (recurring behaviors), expertise (domain knowledge), corrections (mistakes learned from), and gotchas (traps to avoid). Each type has different decay rates and promotion paths.
Cross-session consolidation: After each session, GreyMatter reviews what was observed, detects contradictions with existing knowledge, and promotes stable observations into permanent expertise. Like sleep consolidation for your AI.
Review it on your phone. The GreyMatter mobile app lets you browse your AI's knowledge graph, confirm what matters, correct what's wrong, and strengthen fading memories — like Anki flashcards, but for your AI's brain.
Two search modes, both local. Full-text search (FTS5) finds anything by keyword, instantly. Semantic search (via local Ollama + nomic-embed-text) finds conceptually related knowledge even when the exact words don't match. Ask "how do we handle auth?" and find entries about JWT tokens, session cookies, and OAuth flows — even if those words never appear in your query.
One SQLite database. On your machine. Nothing sent to any server, ever. EU AI Act compliant by architecture — data sovereignty solved at the design level.
Observations become patterns. Patterns become expertise. Duplicates detected. Contradictions resolved. Your AI gets smarter the more you use it.
Full Claude Code integration via Model Context Protocol. Memory, work queue, security, agents — all accessible through natural conversation. Type /mcp to see them.
Source code is public. Read it, fork it, learn from it. Free for non-production use. Source converts to Apache 2.0 after the Change Date.
Every line of code your AI writes passes through an LLM. What else passes through? Your API keys. Your database credentials. Your customers' PII. GreyMatter scans inline — before your data ever reaches a model endpoint.
PII Detection: SSNs, credit cards, API keys, AWS credentials, private keys, phone numbers — caught and redacted before the model sees them.
Prompt Injection Guard: Instruction override, system prompt extraction, role hijacking, delimiter attacks — blocked at the gate.
Append-Only Audit Trail: Every scan logged. Every block recorded. Full forensic history in your database.
This isn't a plugin you install later. Security runs on every LLM interaction by default. You can't use GreyMatter without it — whether you're using Claude Code, the REST API, or the Python SDK.
SecureLLM is also available as a standalone enterprise product with compliance reporting, multi-tenant policies, and provider-level proxy controls. Learn more
An agent without context is just a fancy autocomplete. GreyMatter agents carry "souls" — persistent identity documents with domain expertise, coding standards, common patterns, and failure modes.
When GreyMatter dispatches a work item, it matches the task to the right expert using vector similarity. A Rust memory safety issue gets the Rust Systems expert. A React component gets the TypeScript Frontend expert. The right brain for every job.
Autonomous work queue: Submit tasks, set priorities, walk away. Agents execute in parallel, run quality gates, and generate follow-up work items. Your pipeline runs 24/7.
Create your own experts by dropping a markdown file in ~/.greymatter/soul/experts/. Your team, your specializations, your rules.
GreyMatter isn't just a tool — it's a platform. The SDK gives your applications access to the full knowledge graph, work queue, and security layer through clean, documented interfaces.
Three transports, one API:
MCP (Model Context Protocol) — 17 tools that integrate directly with Claude Code. Memory, work queue, security, agents — all through natural conversation. This is how most developers interact with GreyMatter daily.
REST API — Start a local API server and access GreyMatter from any web app, mobile app, or dashboard. Full CRUD on knowledge, work items, agents, and config. Every endpoint secured by API key + rate limiting.
Python SDK — Import GreyMatter directly into your Python applications. Six namespaces: gm.knowledge, gm.work, gm.agents, gm.security, gm.sync, gm.config.
Every API endpoint is secured from day one — not bolted on after launch.
API key authentication: Every request requires a valid key. Keys are hashed, scoped to specific operations, and revocable.
Rate limiting: Per-key request limits prevent abuse. Configurable per endpoint — tighter on writes, relaxed on reads.
SecureLLM inline: API calls that touch LLM endpoints pass through the same PII detection and prompt injection scanning as direct Claude interactions. The security layer doesn't care how you reach it.
Local by default: The API server binds to localhost. Expose it through Tailscale, a reverse proxy, or your own network — you control who connects.
GreyMatter's MCP integration is the deepest available for Claude Code. Every tool is designed for natural language interaction — you don't memorize commands, you just ask Claude.
Memory tools: gm_knowledge (search, create, list), gm_observe (record observations), gm_pattern (find recurring behaviors), gm_semantic (vector similarity search)
Work tools: gm_work (submit, assign, track tasks), gm_app_dispatch (dispatch to agents), gm_context (project-aware context retrieval)
Security tools: gm_security (scan content, view audit trail), gm_standards (quality standards enforcement)
System tools: gm_status (health), gm_soul (expert agents), gm_cluster (multi-node), gm_checkpoint (session saves)
curl -sSL https://greymatter.axiomworks.ai/install | bash
Checks for Claude Code (installs it if missing), downloads GreyMatter, sets up your database, configures the MCP plugin, and verifies everything. Two minutes.
With an Anthropic API key:
export ANTHROPIC_API_KEY=your-key-here
Or with a Claude Pro/Team subscription, start Claude and type:
/login
cd your-project-directory claude
Claude starts with GreyMatter's 17 tools loaded. Type /mcp to see them.
Ollama enables vector search — finding conceptually related knowledge. GreyMatter works without it (keyword search only).
curl -fsSL https://ollama.com/install.sh | sh ollama serve ollama pull nemotron-3-nano # reasoning ollama pull nomic-embed-text # embeddings
GreyMatter Solo works standalone. When you're ready to scale across machines or collaborate with other developers, toggle on Teams mode.
CRDT sync: Work offline. Changes merge automatically when you reconnect. No conflicts, ever. Built on conflict-free replicated data types — the same technology Linear and Figma use internally.
Multi-node orchestration: Distribute work across your MacBook, a server, a Linux box — all coordinated through a central control plane.
Privacy by design: Your personal observations never sync. Team knowledge is shared. You control the boundary. Up to 3 team connections per Solo instance.