Platform Overview
StellarBase is the modular platform on which you build your organization's AI workflows. This page gives you the map — what's inside, how the pieces fit, and where to read next.
The pieces
The platform breaks down into a small number of primitives that combine in many ways:
| Primitive | Role | Docs |
|---|---|---|
| Knowledge base | Your unified data layer across every source and format | Knowledge Base |
| Data sources | Connectors that pull data in continuously | Data Sources |
| StellarOCR | Composite document-processing engine — text, layout, tables, math, images | StellarOCR |
| DSM | Dynamic Semantic Module — the engine that links documents, entities, concepts | DSM |
| Agents | Scoped AI specialists with your playbook encoded as a prompt | Agents |
| Workflows | Deterministic pipelines that orchestrate agents, tools, and data | Workflows |
| Chat | Conversational interface over your knowledge base and agents | Chat |
| Search | Natural-language retrieval across every connected source | Search |
| Collaboration | Shared Bases, comments, approvals, audit log | Collaboration |
How they fit together
The typical flow:
- A data source connects an external system (Drive, SharePoint, a database). Documents flow in.
- StellarOCR parses each document — text, layout, tables, figures, math — preserving structure.
- The DSM engine indexes everything, resolves entities, builds the semantic graph.
- Your agents query that graph. Their prompts encode your playbook; their tools include anything you’ve registered.
- Workflows orchestrate agents into repeatable pipelines that produce structured outputs.
- Users interact via chat and search; colleagues collaborate on outputs in shared Bases.
What’s unique
Modular by design
Every primitive is composable. An agent can call another agent. A workflow can trigger another workflow. A knowledge base can be scoped to a subset for one agent and the full set for another. You don’t get trapped in pre-baked scenarios — you assemble the platform you actually need.
Bring your own everything
Your ML models plug in as tools. Your APIs plug in as connectors. Your file formats are read by StellarOCR. Your business rules become agent prompts. The platform bends to your stack, not the other way around.
Multilingual from the ground up
Ingest in 24+ languages, query in any of them. The DSM normalizes across languages — a search in Czech finds relevant German and French sources. See Multilingual.
Cited and auditable
Every AI output traces back to source passages. Every workflow run produces a reproducible trace. Every permission decision is logged. This is what makes StellarBase suitable for regulated work.
What’s next
- If you’re starting fresh: read Knowledge Base, then Agents, then Workflows in that order.
- If you already know what you’re building: skim the primitive you need and work outwards.
- For production deployment: see Deployment.