Reasoning
Deliberate
Planning, proofs, architecture, and difficult debugging.
Automatic routing
Auto / Lite Mode: let the workspace route the task.
The model directory calls this Auto Mode; the user guide calls the automatic routing entry Lite mode. Choose the automatic option shown in your client, then switch models when the assignment needs it.
Floatboat · Auto / Lite Mode
Current model
Auto / Lite Mode
What do you want to finish?
Plan a difficult projectSummarize a long documentInspect this screenshotRefactor this code
Describe the outcome or add project files…
- Access
- Default client option
- Availability
- Available to everyone
- Routing signals
- Task, context, modality, availability
- API key
- Not required
A dispatcher that removes model management from the job.
Auto Mode is for people who want to describe the work once and let Floatboat decide how much reasoning, context, and speed it needs.
Routes without extra setup
Initiative
The request itself supplies the routing signal.
Keeps one workspace
Continuity
Files, rules, and prior output stay with the project.
Supports fallback
Resilience
A suitable fallback can keep work moving when a route is unavailable.
Manual choice remains available
Control
Switch to a fixed model when repeatability matters.
Different work calls for different strengths.
Reasoning depth, context length, media input, and response speed can point to different execution routes. This overview explains the task demands considered by Auto Mode rather than presenting routing as a preference for one permanent model.
Long context
Large input
Repositories, research collections, and long documents.
Automatic routing
Vision & media
Multimodal
Screenshots, diagrams, images, and mixed inputs.
Automatic routing
Fast work
Everyday
Drafting, classification, and lightweight edits.
Automatic routing
Use Cases for Auto / Lite Mode
Research, implementation, and daily edits place different demands on an execution route. These scenarios examine continuity across changing requests and interruptions, with results checked against the same project constraints after a route changes.
From research to implementation
A project may move from reading sources to choosing a design and implementing it. Keep the research, open decisions, and approved requirements in the same context as the work progresses. Check that implementation and verification still follow the original constraints and that conclusions remain linked to their sources.
Mixed daily queue
A daily queue can mix summaries, local edits, and fault investigations. Treat each request according to its own requirements while retaining the project context needed by adjacent steps. Review the summary, diff, or diagnosis with the evidence appropriate to that task, rather than applying one acceptance rule to everything.
Continue through disruption
An unavailable model route can interrupt a task before its work is complete. Retain the current state and continue through available system routes where possible, then recheck fallback results against the original requirements. Identify any steps that remain blocked instead of treating a changed route as proof of completion.
Keep text and visual evidence connected
A task may combine written requirements with screenshots or other supported visual input. Identify what each part needs to establish and keep them in the same project context while using an available route that fits the input. Review the conclusion against both sources, and leave unsupported media steps open instead of assuming every modality shares one path.
Routing decisions that normally interrupt the work.
Automatic routing should adapt execution without replacing the project being worked on. Examine how task complexity, retained context, input modality, and fallback affect model selection while the original material and expected outcome remain the reference for review.
Task-complexity routing
Distinguishes lightweight requests from work that deserves deeper reasoning.
Context-aware selection
Accounts for repositories, document sets, and long-running conversations.
Modality fit
Considers whether the task includes text, images, or mixed material.
Automatic fallback
Can move to a suitable alternative when the preferred route is unavailable.
Use Auto when the work changes shape.
Choose Auto Mode when
- You want a reliable default without tracking every model release.
- One project mixes research, reasoning, editing, and execution.
- Continuity matters more than fixing one model for every request.
Choose a model directly when
- A repeatable workflow requires the same model behavior every time.
- You are testing or comparing a specific model.
- A task has a known requirement that only one listed route satisfies.
How Auto Mode routes a request
- Read the task
- Floatboat identifies the required depth, context, and modality.
- Choose a route
- It selects a suitable model and can fall back if that route is unavailable.
- Keep continuity
- Files, rules, and output remain attached to the same project.
Explore Other Model Families
Compare model families by the kind of work they support. Availability differs by model and entry point.
Auto / Lite Mode FAQ
When should I use Auto Mode instead of choosing a model?
Use Auto Mode when the task matters more than maintaining a fixed model choice, especially when one project moves between research, reasoning, editing, and execution.
Can I override Auto Mode?
Yes. Select a model directly when repeatability or explicit control matters.
Does one request go to every provider?
No. Auto Mode selects a route; it does not broadcast the request to every model.
What happens when a route is unavailable?
Auto Mode can choose a suitable fallback so the task can continue.
What signals does an AI model router use?
Auto Mode evaluates task complexity, context size, input modality, response speed, and route availability before selecting a model.
How can I compare Auto Mode with a fixed-model run fairly?
Use the same brief, files, tools, and acceptance checks, and record the routing information available in the client. Compare the accepted deliverables and revisions; two runs can differ even when the visible request is identical.
Route the next task without picking a model
Start with the real brief and files. Auto Mode chooses among available models and falls back when needed, so you can review the result.