Real files and tools stay in reach
Environment
Work with approved local folders, project materials, browser context, and tools instead of repeatedly copying them into a conversation.
Give agents a workspace, not another chat window.
Floatboat brings agents, local files, models, tools, project rules, and deliverables into one persistent desktop environment—so work can continue from context instead of restarting with every prompt.
- Local-first project access
- Multi-agent collaboration
- Human-controlled execution
FLOATBOAT WORKSPACERunning locally
Work queue
F
Floatboat Agent
Approved for this task
Open a bounded project
Every step stays inspectableReady for approval
Keep the project, rules, and results together
Agents can work from the same approved files and tools, leave inspectable outputs, and continue from what the project has already learned.
The environment around the agent is part of the product.
A persistent workspace connects local material with agent execution and editable outputs. Compare how project context, working rules, and reviewable deliverables support ongoing work without requiring every task to restart in a new conversation.
Context survives the individual task
Continuity
Keep workspace rules, folder-specific guidance, run history, and prior outputs available when the next piece of work begins.
People remain inside the execution loop
Control
Review artifacts, redirect agents, change models or instructions, and approve actions that require human judgment.
Use Cases for Agent Workspace
Keep research, repository changes, and editable deliverables connected in one project. These scenarios show how an Agent Workspace preserves sources and decisions as work passes between roles and review stages.
Coordinate many roles from one desk
Research, operations, content, and client work can share one project without losing their separate responsibilities. Keep the source material, task assignments, and outputs together, and identify unanswered questions when work changes hands. Each handoff should retain the decisions and sources the next role needs.
Work with repositories and project rules
Start repository work with the codebase, project rules, and a bounded requirement. Read the relevant interfaces before editing, use the project's existing tools to check the change, and deliver the diff with verification records. Reviewers can then assess both behavior and consistency with repository conventions.
Produce work that can be examined
A report, slide deck, or analysis needs more than a finished-looking draft. Organize approved files around the delivery format and review criteria, retain source attribution, and record questions that still need a decision. The resulting editable work gives reviewers a clear route back to its evidence.
Resume an approved project
An accepted draft, review notes, and a new request can form the starting point for the next task. Keep the relevant files and decisions in the project, identify what changed, and ask the agent to continue within that boundary. Review the new result against both the fresh request and the requirements already agreed.
Run a project in Agent Workspace
Setting up a project means giving the agent relevant material, defining the work, and reviewing what returns. The steps below keep those decisions connected with execution records so accepted results remain available for the next task.
Open a bounded project
Choose the project folder, relevant files, tools, and rules; state the intended deliverable and access limits.
Assign and supervise the work
Give Agents distinct research, drafting, checking, or creation tasks. Review intermediate output and approve actions that need a decision.
Review and retain the handoff
Compare artifacts with the original brief, request corrections, and retain useful context and accepted results for the next task.
Compare project execution with question-and-answer chat
This compares a persistent project workflow with a chat used mainly for questions and answers. Other products may add tools, files, or automation of their own.
Scroll horizontally to compare all columns; dimensions stay visible on the left.
| Comparison dimension | Floatboat Agent Workspace | Question-and-answer chat |
|---|---|---|
| Working context | Persistent project environment | Conversation history |
| Files and tools | Scoped where work lives | Often uploaded or copied in |
| Execution | Agents, tools, triggers, and review | Prompt-and-response turns |
| Outcome | Editable project deliverables | Primarily a message |
Explore More Floatboat Features
See how related capabilities fit into a complete Work Agent workflow.
Agent Workspace questions
What is an AI Agent Workspace?
It is a persistent work environment where agents can use approved files, tools, models, rules, and prior outputs to complete and continue real projects—not just answer isolated prompts.
How is this different from the Floatboat Coworker?
The Agent Workspace is the environment that holds context and resources over time. The Coworker is a Work Agent you can give a bounded task to and receive a deliverable from.
Does local-first mean nothing ever leaves my computer?
No. Local-first means work begins from your real desktop environment and files. Data handling still depends on the cloud models and connectors you choose, so each external path should be evaluated separately.
Can more than one agent work in the same workspace?
Yes. Floatboat supports multiple agents working from shared project context, with human review, redirection, and approval remaining part of the process.
What should stay with a project between agent runs?
Keep the brief, source files, project rules, accepted deliverables, and unresolved questions together. A transcript alone may omit why a decision was accepted or which file is authoritative.
When should I start a separate workspace?
Use a separate project when the client, access permissions, source material, or acceptance criteria change. This makes the intended context boundary explicit; still check which files and tools each task can access.
Continue the same agent project tomorrow
Reopen the same files, rules, agent decisions, and draft deliverable instead of rebuilding context in a new chat.