- NeuralGen
- Atlas
Product
Atlas
Your knowledge. One intelligent interface. A provider-independent platform that lets people ask questions across all their organisation's content.
Organisations keep knowledge in many separate systems. Atlas connects to them, indexes the content and lets people ask in plain language. It is a secure discovery layer on top of what you already have, and you choose the AI behind it: local, private or cloud.
Not another chatbot. Every answer is grounded in your documents and can be traced and checked.
Why Atlas
Three ideas behind it
Any model
All AI traffic goes through one AI Gateway. Use local runtimes or cloud APIs, and switch without changing code.
Grounded answers
Hybrid retrieval plus an agent that plans its own steps. Every answer cites its sources.
Enterprise-ready
Login by default, SSO, roles, encrypted secrets, audit log, tenant isolation and rate limiting.
How it works
From scattered files to a cited answer
- 1
Connect
Link your document sources. Atlas detects changes and keeps the index fresh.
- 2
Index
Content is parsed, chunked and embedded with the model you choose.
- 3
Ask
Hybrid retrieval and an optional planning agent assemble the evidence.
- 4
Verify
Every answer links back to the original documents it came from.
Features
Everything in one workspace
Ask Atlas
Streaming chat in Grounded and Agent modes, with inline citations and separate history per user.
Explore
Browse everything that is indexed. Filter by source, type, owner and date.
Sources
Connect, authenticate, pick folders or libraries and watch sync health.
Knowledge
Document inventory, chunk statistics and indexing status at a glance.
AI Models
Local and cloud providers, with separate chat and embedding models.
Users & access
Local accounts, roles, and Google, Microsoft or OIDC single sign-on.
Integrations
Connectors and model providers
Sources
- Local folders
- Mounted read-only into the container
- SMB / network shares
- Credentials, with share discovery
- Google Drive
- OAuth
- OneDrive
- OAuth via Microsoft Graph
- SharePoint
- OAuth via Microsoft Graph
- GitHub
- OAuth or personal access token
Model providers
- Ollama
- LM Studio
- vLLM
- SGLang
- LocalAI
- OpenAI
- Azure OpenAI
- Anthropic
- Google Gemini
- Mistral
- Groq
- OpenRouter
- Any OpenAI-compatible endpoint
Chat and embedding models are configured separately, per provider, in the UI.
Security
Authorisation is enforced by Atlas, never by the model
- Login required by default, with local accounts or SSO
- Role-based access control on every route
- Document permissions enforced inside the vector query, so restricted content is never retrieved
- Provider credentials encrypted at rest
- Audit log, rate limiting and prompt-injection filtering
- Optional retention policies and GDPR erasure
- Internet lookup off by default, with admin and per-user controls
Technology
- Frontend
- Next.js, React, TypeScript, Tailwind CSS
- Backend
- Python 3.12, FastAPI, async SQLAlchemy
- Data
- PostgreSQL for metadata and auth, Qdrant for vectors
- Parsing
- PDF, DOCX, XLSX, PPTX, Markdown and text
- Deployment
- Docker Compose with health checks and resource limits

See Atlas on your own content
We will run a short demo and, if it fits, a pilot against a small set of your real documents.