Chat Interface & State

Relevant source files

The following files were used as context for generating this wiki page:

• README.md
• packages/conduit_core/test/fixtures/chat_blobs/01_minimal_new_chat.json
• packages/conduit_core/test/fixtures/chat_blobs/04_file_image_attachments.json
• packages/conduit_core/test/fixtures/chat_blobs/05_web_search_status_sources.json
• packages/conduit_core/test/fixtures/chat_blobs/06_code_executions.json
• packages/conduit_core/test/fixtures/chat_blobs/11_embeds.json
• assets/release_notes/en.json

Conduit provides a native mobile interface for interacting with various AI backends including Open WebUI, direct model endpoints (Ollama, OpenAI), and local Apple Intelligence models. The system manages chat state through a structured "chat blob" format that synchronizes between the local device and remote servers, supporting token-by-token streaming, multimodal attachments, and complex execution statuses.

Sources: README.md:1-50, packages/conduit_core/test/fixtures/chat_blobs/01_minimal_new_chat.json

Chat State Architecture

The application stores chat state in a comprehensive JSON structure referred to as a "blob". This blob contains the conversation history, metadata (envelopes), and extended features like code execution results or web search sources.

Chat Blob Schema

A chat object consists of several core components:

• Envelope: High-level metadata for indexing and listing (ID, title, folder, pinned status).
• History: A graph-based structure mapping message IDs to message objects, tracking the currentId of the active branch.
• Messages: A linear array of message objects for legacy compatibility or rendering.
• Metadata: Lists of models used, parameters, and global file collections.

Sources: packages/conduit_core/test/fixtures/chat_blobs/01_minimal_new_chat.json:3-40, packages/conduit_core/test/fixtures/chat_blobs/04_file_image_attachments.json:3-15

The diagram shows the hierarchical relationship of components within a single chat state object.
Sources: packages/conduit_core/test/fixtures/chat_blobs/05_web_search_status_sources.json, packages/conduit_core/test/fixtures/chat_blobs/06_code_executions.json

Message Composition & Content

Messages in Conduit support multiple roles and rich content types. The system handles standard text, Markdown, LaTeX, and various status indicators.

Message Attributes

Field	Type	Description
id	UUID	Unique identifier for the message.
role	String	user, assistant, or system.
content	String	The actual message text, often containing Markdown or HTML.
done	Boolean	Indicates if a streaming response is complete.
timestamp	Integer	Epoch seconds for the message creation.
parentId	UUID	ID of the preceding message in the conversation branch.

Sources: packages/conduit_core/test/fixtures/chat_blobs/01_minimal_new_chat.json:20-35, packages/conduit_core/test/fixtures/chat_blobs/02_linear_multi_turn.json

Multimodal Attachments

Conduit supports images and files within message objects. Files are serialized with metadata including hashes and content summaries.

• User Uploads: Included in the files array of a user message.
• Assistant Generations: Generated images or processed files returned in assistant replies.
• Embeds: Assistant messages can carry an embeds array containing raw HTML or URLs for interactive charts.

Sources: packages/conduit_core/test/fixtures/chat_blobs/04_file_image_attachments.json:1-65, packages/conduit_core/test/fixtures/chat_blobs/11_embeds.json:1-15

Execution & Status Tracking

Assistant messages track internal model actions through statusHistory and code_executions fields. This allows the UI to render progress for long-running tasks like web searches or code interpretation.

Web Search Flow

When a model performs a web search, the state updates through multiple actions:

1. web_search_queries_generated: Lists the search terms.
2. web_search: Records searched URLs and snippets.
3. sources_retrieved: Finalizes the count of reference documents.

Sources: packages/conduit_core/test/fixtures/chat_blobs/05_web_search_status_sources.json:45-90

The flowchart illustrates the sequential status updates stored in the message state during a web-search turn.
Sources: packages/conduit_core/test/fixtures/chat_blobs/05_web_search_status_sources.json

Code Execution

The code_executions array stores the code snippet, language (e.g., Python), and the resulting output or generated files (e.g., charts stored in /cache/code/executions/).

Sources: packages/conduit_core/test/fixtures/chat_blobs/06_code_executions.json:45-75

User Interface & Controls

The interface supports advanced chat features such as branching, model comparison, and reasoning visibility.

• Reasoning Blocks: The app renders <details type="reasoning"> tags natively, showing the model's thought process with duration timers.
• Branching: The state tracks childrenIds and currentBranchHints to allow users to switch between different paths in the conversation.
• Regenerated Answers: Users can switch between multiple model attempts for the same prompt using UI arrows that navigate the state's message graph.

Sources: README.md:120-150, assets/release_notes/en.json:4.1.9, packages/conduit_core/test/fixtures/chat_blobs/08_unknown_future_keys.json:45-60

Connection Types

The chat interface adapts its features based on the active backend:

• Open WebUI: Full features (tools, web search, folders).
• Direct: Model-specific endpoints (Ollama, OpenRouter).
• Apple On-Device: Local offline inference on iOS.
• Hermes: Agent-centric interaction with tool approval steps.

Sources: README.md:55-100

Conduit centralizes chat logic in a robust state blob that facilitates native rendering of AI artifacts. By managing hierarchical message history and detailed execution statuses, the system maintains consistency across diverse model backends and mobile platforms.