Chat Overview

Learn the core Starfire AI conversation experience.

Chat & reasoning

Chat is the interactive core of Starfire AI. It is where a user works directly with a model, tools, files, project context, and—when enabled—research or knowledge systems.

A Starfire conversation can combine:

• prior messages in the conversation
• the selected or routed model
• reasoning configuration
• account and plan entitlements
• project context
• attached files and images
• knowledge retrieval
• tools such as search or calculation
• organization policies
• runtime and usage limits

Request lifecycle

At a high level, a chat request moves through several layers:

Your message
    ↓
Conversation + project context
    ↓
Entitlements / policy / feature availability
    ↓
Model selection or routing
    ↓
Tool and context preparation
    ↓
Provider execution
    ↓
Streaming response
    ↓
Usage + credits + operational telemetry

The exact internal routing logic is proprietary and can evolve without changing the user-facing contract.

Conversation context

Starfire keeps earlier messages available to the active model within the limits of the model and platform. Very long chats can become less efficient because more context must be selected, summarized, or processed.

Use a new conversation when the goal changes substantially. Use a Project when many related conversations should share durable context.

Models

Starfire is provider-independent at the product layer. Models can differ in reasoning, context, tool support, vision, output limits, latency, and credit consumption.

Depending on the account, Starfire can expose direct model selection, platform-managed routing, or both.

Models & routing

Understand availability, capabilities, fallback behavior, and how Starfire chooses a route.

Tools

When enabled, a model can receive tools that extend what it can do beyond generating text. Examples in the Starfire platform architecture include search, URL reading, workspace retrieval, calculation, build filesystem operations, validation, and artifact packaging.

Tool availability can depend on the active model, plan, workflow, organization policy, and administrator configuration.

Files and images

Files can provide task-specific context directly to a conversation, while persistent Knowledge or Project sources are better for material that should remain reusable across many chats.

Research

For a quick factual lookup, ordinary chat with search can be enough. For work that requires multiple sources, source comparison, claim verification, or a structured report, use the research workflow when available.

Failures and fallback

Requests can fail because of provider errors, timeouts, rate limits, unsupported capabilities, invalid context, or platform restrictions. Starfire can use configured retry and fallback behavior where appropriate, while still surfacing meaningful failures.

A model returning an answer is not the same thing as a fact being verified. For important external claims, use search, research, citations, or source inspection.

Next steps

Prompting effectively

Give Starfire a clear objective, context, constraints, and output contract.

Files & context

Understand temporary attachments versus durable project and knowledge context.

Search & research

Know when to use web search, citations, and deeper research workflows.

Projects

Move long-running work into a persistent workspace.