Explicit contracts
Extensions use public SDK methods. Hidden imports and private coupling make systems harder to reason about.
Project principles
Agentic software should stay inspectable as it becomes useful.
Democr.ai is built around a simple position: generated behavior needs platform boundaries, runtime visibility, and recoverable state if it is going to ship inside real applications.
Principles
Make the runtime boundaries visible.
The platform is useful only if developers can understand where behavior lives and why it is allowed to run.
Contractflow stage
Scopeflow stage
Traceflow stage
Reviewflow stage
UI as runtime output
The interface can be generated and streamed, but it still needs stable components, clear actions, and renderer contracts.
Context with scope
Messages, documents, embeddings, and graph nodes need user, organization, module, and thread metadata.
What this avoids
The target is not novelty; it is control over moving parts.
Agentic applications fail operationally when runtime work becomes invisible or spread across unrelated layers.
Opaque glue
Business logic spread across scripts, frontend state, provider clients, and storage adapters.
Untracked work
Background tasks and agent steps that fail without surfacing state to users or operators.
Unsafe shortcuts
Feature modules reaching directly into core internals because the platform boundary is missing.
Engineering stance
Keep product code readable after the prototype phase.
The project assumes that speed is useful only when the resulting application can still be debugged, migrated, and reviewed.
Small modules
Feature packages should be understandable without reading the entire runtime.
Traceable context
Requests should carry enough metadata to explain why a model, tool, attachment, or retrieval result was used.
Replaceable providers
Provider choice should be configuration and capability driven rather than hardcoded in feature code.
Next
Related pages
Use these pages to move from the concept to adjacent parts of the runtime.