BrainOS - The knowledge layer for AI agents.

BrainOS solves the core blocker to enterprise AI automation: models are capable, but they lack company-specific operational knowledge. Critical context lives across Slack threads, PDFs, screenshots, diagrams, runbooks, support tickets, and tribal knowledge inside employees’ heads. AI agents cannot reliably operate on fragmented information. BrainOS is a multi-agent, multi-modal system that ingests company knowledge from text, PDFs, screenshots, architecture diagrams, and live Slack conversations, then builds a living reconciled knowledge graph. Incoming information is continuously reconciled against existing facts: stale ownership is superseded, duplicates are removed, and conflicting claims are marked Disputed when unresolved. The system exports portable SKILLS.md context files consumable by Claude Code, Cursor, OpenAI GPTs, Aider, and other AI agents. Four specialized agents run on AMD MI300X through vLLM: an Ingestion Agent (text + vision), Structuring Agent (extraction + reconciliation), Execution Agent (graph-aware retrieval + grounded generation), and Feedback Agent (groundedness auditing). MI300X’s 192GB HBM3 enables co-resident 70B text and 7B vision models on a single GPU without swapping, enabling sub-second multimodal ingestion and per-task model routing. BrainOS includes a production-ready Slack integration with thread ingestion, channel ingestion, semantic search-then-ingest, and /brainos slash command support for grounded in-channel Q&A with confidence labels. Knowledge can also sync into Slack Canvas for department-level operational memory. Beyond standard RAG, BrainOS adds conflict detection, provenance tracing to source quotes, confidence scoring, knowledge-gap analysis, security-aware segmentation, and department-scoped memory isolation. Self-hostable, vendor-neutral, and open-source, BrainOS continuously evolves its organizational understanding while surfacing what knowledge is now disputed.