Stop losing knowledge in Slack threads and Notion docs. xbrain gives your team a shared, persistent AI memory layer — searchable, scoped by team, verified by truth level.
Every team using AI hits the same wall — three distinct failure modes that compound over time.
Every developer reinvents their own prompts. Architecture decisions made in a Slack thread disappear within weeks. The next hire asks the same questions all over again.
ChatGPT, Claude, Cursor — each lives in its own bubble with zero shared memory. Your most useful AI conversations are trapped inside one person's chat history.
Is that architecture decision still valid? Was it a brainstorm or a final call? There's no way to know if information is fresh, validated, or just someone's idea from last quarter.
xbrain is a self-hosted memory layer that sits between your team and your AI tools. Every conversation, every extracted fact, every validated decision lands in one place — tagged by team, by project, by truth level.
LibreChat ──┐
Open WebUI ─┤──► memory-api ──► PostgreSQL 17
Chrome Ext ─┤ │ Qdrant v1.17.1
Agents ─────┘ │ Neo4j (graph)
│
7 mandatory tags:
· team_scope
· truth_level
· project_scope
· visibility
· confidence
· source
· validation_status
Six integrated capabilities, all open-source, all running in your own infrastructure.
Every conversation, document, and agent output is stored with 7 mandatory tags: team scope, truth level, project, visibility, confidence, source, and validation status. Nothing gets lost.
Facts climb the truth ladder through human-approved workflows. Direct truth_level patches return 405 — promotion is always explicit.
LibreChat and Open WebUI connect to the same memory layer. Use Claude, GPT-4, or Grok in the same team — every conversation enriches the shared brain regardless of which frontend your teammates prefer.
Scraper, Google Drive reader, Calendar, Deck generator — all registered via a central MCP gateway. Every tool call is audited with team_scope and user_id. Extend with your own MCP servers.
Add brain.yaml to any repo. Push to GitHub and xbrain auto-indexes the content at your chosen truth level — architecture docs, runbooks, decisions.
Graphiti extracts facts from conversations. Neo4j stores entity relationships and lineage with timestamps. Query "what depends on entity X?" or "what changed last week?" directly from LibreChat.
No cloud setup. No SaaS account. One Docker Compose command.
Clone the repo and run one command. All 25 containers — frontends, memory layer, vector store, graph DB — start together.
Google OAuth and GitHub Org verification out of the box. Set team scopes, assign roles via RBAC. Each member gets their own data isolation.
Every chat, every Drive doc, every agent output is auto-tagged and stored. Graphiti extracts facts. The shared knowledge base compounds over time.
No custom databases, no proprietary runtimes. Every component is a battle-tested open-source project you can run anywhere.
All components are open-source. No vendor lock-in. Deploy on any machine with Docker.
Read the deployment guide