Broad recall
Full-text search
Scans the complete text of your indexed files to surface relevant passages, phrases, and surrounding detail across the workspace.
Context is the advantage
Three ways to find what matters.
One stronger context.
Cortex does not trust a single retrieval method to understand your work. It combines complementary search signals so the model receives relevant passages, precise terms, and meaningfully related material together.
Exact precision
Exact keyword search
Locks onto names, identifiers, product codes, clauses, and specialist terminology where the exact words matter.
Semantic relevance
Vector search
Finds conceptually related content even when your question and the source use different language to describe the same idea.
Combined by Cortex
Superior context for model reasoning
Cortex brings these results together into a focused evidence set for the model. Better source material in means more grounded analysis, fewer missed facts, and answers that reflect the knowledge in your workspace.
Not a hard-coded chatbot
Behavior assembled from configuration, not baked in
The WebUI is just the visible control surface. Beneath it, the Matbot runtime loads whatever your active workspace says it should — and nothing more.
Local chat
A fast, familiar conversation surface in your browser — powered entirely by the runtime on your machine.
Provider selection
Switch between OpenAI-compatible providers per workspace, per conversation. Local models welcome.
Durable memory
Say “remember my name” once. Remembered facts, skills, and Mem0-backed memory persist per workspace.
Controlled file access
Local services expose safe file indexing and safe file access — governed, auditable, never a free-for-all.
Hybrid retrieval
Full-text, exact keyword, and vector search work together to build stronger context from workspace files.
Expert panel
A tool-based panel of domain experts that answer independently and synthesize a cited decision.
Workspace isolation
Every project gets its own world
Each workspace carries its own providers, plugins, memories, skills, files, and RAG indexes — persisted independently. Your legal review workspace never sees your product research, and compliance can audit each one on its own terms.
- Per-workspace matbot.yaml defines exactly what loads
- New workspaces copy the Default shape, then diverge freely
- Durable memory and indexes stay local, per workspace
D
DefaultOpenAI-compatible providers · workspace RAG · expert panel
isolatedL
Legal reviewLegal expert · contract index · strict file scope
isolatedR
Product researchLocal provider · markdown RAG · research memory
isolatedSelectionMode: Parallel
✓ Legal Expert
✓ Security Expert
✓ Compliance Expert
✓ Finance Expert
✓ Operations Expert
✓ Synthesize decision — one answer, citations included
Expert panel
Ask once. Hear from the whole panel.
The expert-panel tool runs selected domain experts independently — legal, security, finance, compliance, operations, and more — then synthesizes their opinions into one decision, with citations. Run them in parallel, pick who’s on the panel per question.
How it fits together
A composed runtime, one control surface
WebUI
Conversations, files, providers, experts, workspaces, and settings — the visible control surface.
Matbot runtime
Loads providers, plugins, tools, stores, sessions, and hooks from the active workspace configuration.
Local services
Safe file indexing, safe file access, and Mem0 — plus per-workspace persistent data and RAG indexes.
See Cortex on your own files, on your own machine
A 30-minute demo with your team — we'll bring the workspaces.
Request a demo