AI Intelligence Platform -- knowledge graph + semantic search + reasoning + learning in a single binary.
Engram is a self-hosted AI knowledge engine that combines graph storage, semantic search, logical reasoning, and continuous learning into a single Rust binary with a single .brain file. No cloud, no external dependencies.
- Single binary -- no runtime dependencies, no Docker, no cloud
- Single file -- one
.brainfile is your entire knowledge base. Copy = backup, move = migrate - No external database -- custom mmap storage, everything built in
- Hybrid search -- BM25 full-text + HNSW vector similarity + bitmap filtering
- Confidence lifecycle -- knowledge strengthens with confirmation, weakens with time, corrects on contradiction
- Inference engine -- forward/backward chaining, rule evaluation, transitive reasoning
- Ingest pipeline -- NER (GLiNER2 ONNX, GPU-accelerated), entity resolution, conflict detection, PDF/HTML/table extraction
- Multi-agent debate -- 7 analysis modes with War Room live dashboard and 3-layer synthesis
- Chat system -- 47 tools across 8 clusters (analysis, investigation, reporting, temporal, assessment)
- Assessment engine -- Bayesian confidence with living assessments and evidence boards
- Temporal facts -- valid_from / valid_to on edges with automatic extraction
- Contradiction detection -- automatic conflict detection with resolution workflows
- Knowledge mesh -- peer-to-peer sync with ed25519 identity and trust scoring
- Built-in web UI -- Leptos WASM frontend with 3D graph visualization, onboarding wizard, and SSE live updates
- Multiple APIs -- HTTP REST (230+ endpoints), MCP, gRPC, A2A, LLM tool-calling
As a backend memory layer -- integrate Engram into your AI pipeline via REST, MCP, or gRPC. Use the onboarding wizard once, then run headless.
As an intelligence workbench -- ingest documents, build knowledge graphs, run multi-agent debate, assess with Bayesian confidence. Full web UI with interactive 3D graph, chat, and War Room.
Download the latest release from Releases.
| Platform | Download |
|---|---|
| Windows x86_64 | engram-windows-x86_64.zip |
| Linux x86_64 | engram-linux-x86_64.zip |
| Linux aarch64 | engram-linux-aarch64.zip |
| macOS aarch64 | engram-macos-aarch64.zip |
Unzip and run. The web UI frontend is bundled inside the zip.
engram serve my.brain
# HTTP API + Web UI: http://localhost:3030Open http://localhost:3030 -- the onboarding wizard guides you through setup.
We recommend Gemma 4 as the LLM. Run it locally with Ollama:
ollama pull gemma4:e4bAny OpenAI-compatible LLM endpoint works (Ollama, vLLM, OpenAI, Azure, etc.).
Four sections accessible after login:
- Knowledge -- interactive 3D graph explorer, entity search, Knowledge Chat with 47 tools
- Insights -- knowledge stats, contradictions, documents, intelligence gaps
- Debate -- 7 AI analysis modes: Analyze, Red Team, Outcome Engineering, Scenario Forecast, Stakeholder Simulation, Pre-mortem, Decision Matrix
- System -- hardware, embeddings, NER, LLM config, web search providers, ingestion sources, domain taxonomy
| Command | Description |
|---|---|
engram create [path] |
Create a new .brain file |
engram store <label> [path] |
Store a node |
engram relate <from> <rel> <to> [path] |
Create a relationship |
engram query <label> [depth] [path] |
Query and traverse edges |
engram search <query> [path] |
Search (BM25, filters, boolean) |
engram serve [path] [addr] |
Start HTTP + gRPC server |
engram mcp [path] |
Start MCP server (stdio) |
engram reindex [path] |
Re-embed all nodes after model change |
engram stats [path] |
Show node and edge counts |
engram delete <label> [path] |
Soft-delete a node |
| Page | Description |
|---|---|
| Getting Started | Download, install, first brain, quick start |
| Configuration | Onboarding wizard, LLM setup, embeddings, SearXNG |
| HTTP API | Full REST API reference (230+ endpoints) |
| MCP Server | MCP tools for Claude, Cursor, Windsurf (24 tools) |
| Python Integration | EngramClient, bulk import, LangChain, auth, debate, chat |
| SearxNG Setup | Self-hosted web search: installation, engines, rate limits |
| Architecture | System design, layers, storage engine, compute |
| Use Cases | 13 end-to-end walkthroughs with Python demos |
| # | Use Case | Description |
|---|---|---|
| 1 | Wikipedia Import | Build a knowledge graph from Wikipedia summaries |
| 2 | Document Import | Ingest markdown/text with metadata and entity extraction |
| 3 | Inference & Reasoning | Vulnerability propagation and SLA mismatch detection |
| 4 | Support Knowledge Base | IT support error/cause/solution graphs |
| 5 | Threat Intelligence | Threat actor, malware, CVE, and TTP graphs |
| 6 | Learning Lifecycle | Full lifecycle: store, reinforce, correct, decay, archive |
| 7 | OSINT | Open Source Intelligence with multi-source correlation |
| 8 | Fact Checker | Multi-source claim verification |
| 9 | Web Search Import | Progressive knowledge building from web search |
| 10 | NER Entity Extraction | spaCy NER pipeline for entity extraction |
| 11 | Semantic Web | JSON-LD import/export for linked data |
| 12 | Codebase Understanding | AST analysis for codebase knowledge graphs |
| 13 | Intel Analyst | OSINT intelligence dashboard with real-time ingest and gap detection |
| Project | Description |
|---|---|
| Intel Analyst | OSINT intelligence dashboard powered by engram's knowledge graph, ingest pipeline, and gap detection engine |
Engram is free for personal use, research, education, and non-profit organizations.
Commercial use requires a paid license. Contact [email protected] for commercial licensing.
See LICENSE for full terms.