Memories & RAG
AI that remembers what matters — RAG & persistent memory
Upload documents, connect knowledge bases, and let Synaplan find the right context for every answer. True semantic vector search — multilingual, self-hosted, and shared across web chat, widget, WhatsApp, email, and MCP. Your AI stops hallucinating — and starts knowing.
- 01
Any Document In
Upload PDF, Word, Excel, PowerPoint, HTML, Markdown — even scans and audio recordings. Apache Tika extracts the text, OCR reads images, and Whisper transcribes speech.
- 02
Vector Search, Your Choice of Engine
Multilingual bge-m3 embeddings (1024 dimensions) find answers by meaning, not keywords. Run search on MariaDB's native VECTOR type or on Qdrant — switchable, both self-hosted.
- 03
Cross-Session Memory
Remember user preferences, past decisions, and key facts across conversations without manual summarisation.
- 04
Organized & Shareable Knowledge
Group documents per team or topic, tune similarity thresholds, and publish share links with expiry dates. Marketing uses brand guidelines; support uses product docs.
- 05
Reduces Hallucinations
Grounding AI responses in your actual documents significantly reduces hallucinations. The same knowledge base answers in the web app, chat widget, WhatsApp, email, and via MCP.
- 06
Open Source & Self-Hosted
Your documents never leave your server — embeddings can even run locally via Ollama. Synaplan, MariaDB, and Qdrant all deploy with a single Docker Compose.
T-00:00:10