Agent Modes

[English] | 简体中文 · Home

Different work deserves different behavior. BitFun provides purpose-built modes so the agent’s approach matches the job instead of forcing every request through the same loop.

Quick chooser

Mode	Best for	How it behaves
Agentic	Building, changing, and completing work	Explores, edits, runs tools, checks results, and drives toward a finished outcome
Plan	Ambiguous or high-impact work	Clarifies goals, studies context, and proposes an inspectable plan before execution
Debug	Bugs with an unclear cause	Forms evidence-based hypotheses, gathers runtime evidence, and narrows the cause before changing things
Multitask	Several independent workstreams	Separates orthogonal tasks and delegates them to focused subagents
Cowork	Research, writing, analysis, and office work	Collaborates conversationally and turns findings into polished deliverables
Computer Use	Visual desktop workflows	Uses supported screen, pointer, keyboard, and application interactions with explicit permission
Deep Research	Broad questions that need sources	Runs parallel research, compares evidence, and produces a sourced synthesis
Team	End-to-end product or engineering initiatives	Coordinates virtual roles such as product, engineering, design, QA, security, and release
Claw	Ongoing personal assistance	Works inside a persistent assistant identity with dedicated context, memory, tools, and schedules

Agentic

Choose Agentic when you have a concrete outcome and want BitFun to carry the work across discovery, implementation, and verification. It is the general-purpose choice for most coding and action-oriented tasks.

Good prompts: “Add this feature,” “modernize this screen,” “prepare the release notes,” or “fix the test and verify the behavior.”

Plan

Choose Plan when the route matters as much as the destination. BitFun investigates the project, asks only the questions that materially change the result, and produces a staged plan you can review before implementation begins.

Good prompts: “Plan a migration,” “design a permission system,” or “show me the safest way to split this project.”

Debug

Choose Debug when symptoms are visible but the cause is not. The mode favors hypotheses, instrumentation, observations, and reproduction evidence over speculative edits.

Good prompts: “Why does this fail only after reconnecting?” or “Find the cause of this intermittent crash.”

Multitask

Choose Multitask when the request contains genuinely independent workstreams. BitFun keeps a coordinating view while specialized subagents explore or execute bounded parts in parallel.

Good prompts: “Audit the backend, frontend, and documentation at the same time,” or “Compare three approaches independently.”

Cowork

Choose Cowork for less code-centric collaboration: research, writing, planning, meeting notes, travel plans, analysis, documents, slide decks, spreadsheets, PDFs, and visual communication.

Good prompts: “Turn these notes into a proposal,” “research the market and make a deck,” or “analyze this spreadsheet and present the findings.”

Computer Use

Choose Computer Use when the work must happen through a visual interface. BitFun can interact with supported desktop and browser environments while keeping permissions and visible actions in the loop.

Good prompts: “Organize these downloaded files,” “walk through this web workflow,” or “check how this page behaves in the browser.”

Deep Research

Choose Deep Research for questions that require breadth, source comparison, and a defensible synthesis. It can divide the research into parallel lines of inquiry and bring them back into one report.

Good prompts: “Map this emerging market,” “compare the current policies across regions,” or “write a sourced technical landscape.”

Team

Choose Team for initiatives that benefit from several professional perspectives. Virtual roles can challenge requirements, architecture, UX, quality, security, and release readiness as the work moves through a sprint-like flow.

Good prompts: “Take this idea from brief to release candidate,” or “run a cross-functional review of this product change.”

Claw

Choose Claw when continuity matters. A Claw assistant can keep its own identity, role, preferred tools, Skills, model choices, memory, and scheduled work—useful for recurring personal or professional responsibilities.

Good prompts: “Be my weekly planning assistant,” “maintain a research watchlist,” or “help me run this project over time.”

You can change the shape of the agent

Modes are starting points, not cages. You can also:

• create custom Agents and subagents;
• decide which tools, Skills, and MCP connections each one may use;
• choose a primary, fast, inherited, or automatic model strategy;
• mark an agent as read-only or review-focused;
• organize reusable capability groups for different jobs;
• apply personal and project-level permission rules.

Next: see the coding workspace, explore Cowork, or build a personal assistant.