Skip to content
CCinder
Concepts Overview

Concepts

Concepts Overview

Understand the semantic model behind tags, graph building, local resolution, dispatch, speech, and bounded fallback.

Read this before the deeper pages

Cinder makes the most sense when you treat it as a semantic runtime for websites you control, not as a chatbot shell or a speech-only layer.

This section explains the mental model that makes the SDK coherent.

The shortest mental model

Cinder works like this:

  1. semantic tags describe the app
  2. scan turns the tagged DOM into structured records
  3. graph building turns those records into scoped application state
  4. local planners use graph + route/context awareness to resolve commands
  5. dispatch executes the result through routes, DOM behavior, adapters, or bindings
  6. fallback is optional and bounded, not the default path

Core ideas

Semantic tags

Meaning is declared directly in the UI through route, context, field, filter, action, output, and data tags.

Scoped application graph

The runtime turns scan results into a graph with scoped identity, route/context relationships, indexes, active state, and selections.

Graph-aware local resolution

Known commands should resolve through planners, graph queries, route/context state, and optional runtime context before fallback is considered.

Real app execution

Dispatch uses route activation, standard DOM behavior, adapters, bindings, and composite steps to act on the actual app surface.

Bounded fallback

Fallback exists for unknown or weak local understanding, with timeout, classification, normalization, and telemetry built in.

Honest failure handling

Ambiguous duplicates, callback failures, and invalid fallback responses should degrade clearly instead of silently choosing wrong targets.

  1. Semantic Tags
  2. Application Graph
  3. Local Resolution
  4. Tier 2: Embeddings
  5. Dispatch and Feedback
  6. AI Fallback
  7. Speech Input

Continue reading