Customize and Extend

[English] | 简体中文 · Home

BitFun is designed to become more personal and more specialized over time. You can shape agent behavior, connect external capabilities, package repeatable know-how, and create purpose-built interfaces—without requiring every workflow to become part of the core product.

Custom Agents

Create Agents and subagents for specific roles:

• define their instructions and working style;
• choose allowed tools and Skills;
• connect selected MCP services;
• select automatic, inherited, primary, fast, or specific model behavior;
• keep an agent read-only or review-focused;
• organize personal and project-level agents separately.

External agent definitions can also be discovered and reviewed before use where supported.

Skills

Skills package repeatable knowledge and workflow guidance. They can teach an agent how to produce a particular artifact, follow a specialized process, use an organization’s conventions, or coordinate a sequence of tools.

BitFun includes Skills for browser work, documents, PDFs, presentations, spreadsheets, visual canvases, Mini App creation, review workflows, and more. You can add your own or install compatible community Skills.

MCP

Connect Model Context Protocol services to give approved agents access to additional tools, resources, and prompts. BitFun supports local and remote MCP connections, connection controls, authentication options, and per-agent selection.

Treat every MCP service as an extension of the agent’s authority: understand what it can access before enabling it.

Tool and Skill groups

Reusable groups make it easier to create coherent capability sets. A research group might combine web and citation abilities; an office group might combine documents, slides, spreadsheets, and PDFs; a review group might remain read-only and evidence-focused.

Lifecycle hooks

Advanced users can run compatible hooks around agent lifecycle events to apply local policy, record activity, add checks, or block an unsafe transition. BitFun follows the Codex hook contract for these user-controlled integrations.

Hooks execute commands with the authority of their environment. Only enable hook configurations you understand and trust. Project-provided hooks are not something to approve casually.

External Sources — Beta

BitFun can selectively discover compatible capabilities from OpenCode, Claude Code, and Codex environments. You decide whether a source is disabled, discoverable, approval-gated, or automatically allowed, and conflicts can be reviewed.

This is selective compatibility, not a promise that every plugin or behavior from another product works identically. Support varies by source and capability, and external code-based tools may require a local runtime plus explicit approval.

External agents through ACP

Advanced users can configure supported agents through the Agent Client Protocol. ACP is useful when an external agent should participate through a defined interface rather than being rebuilt as a BitFun-native agent.

Model choice

BitFun works with several popular model API families and custom endpoints. Depending on the provider, you can use API credentials or supported subscription sign-in, then assign models to primary, fast, automatic, vision, or speech roles.

Provider catalogs and model names change quickly, so the application’s model settings are the authoritative current list.

Mini Apps

When a workflow needs an interface rather than another prompt, turn it into a Mini App. AI-assisted creation, live draft preview, permission review, versions, and rollback make it possible to evolve small tools alongside the work they support.

Explore Mini Apps and Pages →

Product customization

Because BitFun is open source under the MIT License, teams can also create differentiated distributions and experiences. Community contributors have already explored specialized versions for media and short-form drama workflows.

That freedom comes with responsibility: a customized distribution should preserve clear identity, permissions, updates, security expectations, and user trust.

A practical extension ladder

Start with the smallest thing that solves the need:

1. Adjust an Agent’s instructions and permissions.
2. Add or create a Skill.
3. Connect an MCP service.
4. Build a Mini App for a dedicated interface.
5. Add trusted lifecycle hooks.
6. Use External Sources or ACP for cross-ecosystem capability.
7. Customize the open-source product only when the experience truly needs to diverge.

Next: review security and privacy, explore Agent modes, or get started.