Announcing Engraphis 0.1.0: the initial public release

Engraphis 0.1.0 is live: a self-hosted memory engine for AI agents built around Ebbinghaus decay, interaction-aware recall, bi-temporal facts, and background consolidation. It's local-first, everything runs on your machine, and you bring your own LLM only if you want optional chat or synthesis features.

What's in this release

This is the full product, not a preview.

• A full MCP server with 18 tools, available via the engraphis-mcp entry point, for Claude Code, Cursor, Cline, Zed, and Windsurf.
• A transport-agnostic MemoryService facade over the v2 engine: the same service backs the dashboard, the Inspector, and the MCP server.
• The Memory Inspector product UI (engraphis-inspector, port 8710), a focused, accessibility-first view with supersession-chain diffs.
• A dashboard rebuilt on the v2 engine, covering recall, governance, consolidation, and analytics.
• Team mode with PBKDF2 logins and viewer / member / admin roles with seat limits.
• Grounded recall with cited answers and an explicit abstain gate: no confident hallucinations.
• Sleep-time consolidation with compaction accounting, so you can see exactly what got distilled.
• A Personalized PageRank graph arm (HippoRAG-style) for graph-aware recall.
• Offline Ed25519-signed license keys: no phone-home required to activate Pro or Team.
• A Pro analytics dashboard and compliance export.
• A code-symbol graph via tree-sitter, with a regex fallback when tree-sitter isn't available.
• Docker and docker-compose deployment out of the box.
• 300+ tests, an eval harness, and an ablation suite.

Security, from day one

The engine ships with 100% parameterized SQL, constant-time auth, and PBKDF2-HMAC-SHA256 for password handling. Input is sanitized, workspaces are isolated, and CORS defaults to loopback-only. When served over HTTPS, cookies get the Secure flag automatically. See SECURITY.md in the repo for the full rundown.

Try it now

The Community tier, the full engine, MCP server, dashboard, Inspector, CLI, and library, is free and Apache-2.0, permanently. See the product page for what Pro and Team add, or head straight to the GitHub repo to read the source.

Give your agents a memory.