Organizational knowledge is valuable because of its sources and boundaries.
Documents, meeting notes, decisions, risks, blockers, and actions often live across disconnected files and workflows. A useful intelligence layer must make that material searchable without weakening tenancy, source traceability, or human accountability.
EchoMind AI is being built to connect organizational evidence through local embeddings and grounded retrieval while keeping PostgreSQL authoritative and isolating private material by organization.
Significant foundations are implemented and stabilized, but the product is actively evolving and is not presented as production-ready.
A tenant-aware application with local intelligence.
System topology
The foundations already cover knowledge and operational workflows.
Identity and tenancy
Secure accounts, organization memberships, role policies, and audit events establish organizational access boundaries.
Document intelligence
PDF, DOCX, and TXT ingestion, asynchronous extraction, local embeddings, document scope, and ChromaDB semantic retrieval.
Grounded assistance
Source-cited answers, prompt-injection defenses, local Ollama inference, and grounded RAG keep responses tied to evidence.
Meeting intelligence
Extracts decisions, risks, blockers, and actions, with human-reviewed task conversion, notifications, and authenticated real-time updates.
Authority, derivation, and review are deliberately separated.
- The organization is the tenant boundary.
- Private organizational evidence is isolated.
- Answers remain grounded in retrievable sources.
- PostgreSQL is authoritative; vectors are derived and rebuildable.
- AI-generated decisions and actions require human review.
- Prompt-injection defenses are part of the document workflow.
A layered application and retrieval stack.
Frontend
Next.js · React · TypeScript · Tailwind CSS
Backend
Django · Django REST Framework · Channels · Celery
Data
PostgreSQL · Redis · ChromaDB
AI
Ollama · nomic-embed-text · Mistral · Grounded RAG