BitFun

A desktop AI agent that turns every task into an app you can open.

Stars
★ 3.5k
Last updated
today
License
MIT
Primary language
Rust

At a glance

How it runs
Desktop appCLISelf-hosted service
Works with
Universal · cross-platform
Cost
Free software; you pay for model usage
Setup effort
Medium · a few setup steps
You'll need
Node.js 22.12+pnpm 10.15.0Rust toolchainTauri prerequisitesShell / CLINetwork accessLocal filesystemMCP Server
Typical use
Developers handling complex coding tasks in real repositories, using modes like Agentic, Plan, Debug, and Deep Review to move from code to merge-ready results.
Not a fit if
  • Teams unwilling to deploy their own relay server for multi-device sync
  • Users with legacy 0.2.x data who won't run the migrator

What does this agent do, and when should you use it?

BitFun is a desktop AI agent combining a high-performance Rust runtime with a polished desktop application. It functions as a Code Agent for real Git repositories, handling planning, editing, testing, and committing, while also supporting office tasks like research, writing, presentations, and reports. Its signature Agentic Mini Apps give each task its own interface—charts, boards, forms, panels—with the conversation bound to that interface's live state. BitFun offers self-hosted multi-device control through a relay you deploy, ensuring zero-knowledge privacy. With four tiers of customization (Custom Agents, MCP/Skills/Hooks, Mini Apps, source-level changes), it's highly adaptable. Performance features include a 98.67% average KV cache hit rate and flashgrep for fast searches in massive codebases. Cross-platform support spans Windows, macOS, and Linux, and it is model-agnostic.

BitFun performs end-to-end tasks within a desktop app: it reads local files and Git repositories, executes terminal commands, controls browsers and desktop applications, and accesses remote workspaces. Its Rust-based runtime handles agent execution, while flashgrep provides fast, cross-turn code search. For coding, it plans, edits, tests, and commits code. For office work, it researches, summarizes, writes, and generates presentations. Mini Apps generate task-specific UIs and bind conversations to their live state. It supports MCP, Skills, and Codex-compatible Hooks for extension, and offers self-hosted login and device sync via a relay. Installation is via downloaded installers or from source with pnpm run desktop:dev.

  1. Developers handling complex coding tasks in real repositories, using modes like Agentic, Plan, Debug, and Deep Review to move from code to merge-ready results.
  2. Knowledge workers transforming research materials into structured documents, reports, presentations, and meeting notes.
  3. Users who need cross-device workflows, syncing sessions and settings across devices via a self-hosted relay and controlling one device from another.
  4. Power users who want to customize agent behavior through Custom Agents, MCP, Skills, Hooks, or Mini Apps.
  5. Maintainers of large codebases using flashgrep to search Chromium-scale projects efficiently.
  6. Teams conscious of API costs, leveraging high KV cache hit rates to reduce expenses.

How do you install or deploy this agent?

Download the latest installer for macOS, Windows, or Linux from GitHub Releases. Alternatively, run from source: install Node.js 22.12+, pnpm 10.15.0 (via Corepack), the Rust toolchain, and Tauri prerequisites, then run pnpm install and pnpm run desktop:dev.

How do you use this agent?

On first launch, click Open on the Welcome tab and select a project folder. Then navigate to Settings > Models > Create First Configuration, choose a provider, enter API key, select models, and save. Return to the Session tab, type a concrete task, and press Enter or click Send.

What are this agent's strengths and limitations?

Pros
  • Agentic Mini Apps provide task-specific interfaces instead of a one-size-fits-all chat box.
  • Self-hosted zero-knowledge relay for multi-device control without vendor cloud dependency.
  • Proprietary flashgrep enables fast search in massive codebases.
  • High KV cache hit rate (98.67%) reduces API costs.
  • Model-agnostic and cross-platform (Windows, macOS, Linux).
Limitations
  • Early-stage project; initial benchmark results are single-run and may fluctuate.
  • Requires user configuration of model providers and API keys.
  • Source-based setup involves multiple dependencies (Node.js, pnpm, Rust, Tauri).
  • Self-hosted relay deployment requires additional server infrastructure.
  • Documentation does not detail enterprise-grade security certifications.

How does this agent compare with similar options?

Compared to other agent frameworks like OpenClaw, BitFun emphasizes its desktop app with Mini Apps and the privacy benefits of self-hosted relay. It also highlights performance metrics like KV cache hit rate and flashgrep.

Key facts side by side with the most closely related agents.

Agent Source review Form / cost Stars Updated Language Full support on
BitFun This agent 40 · Major gaps Desktop appFree + model costs ★ 3.5k today Rust —
Devo Desktop Coding Agent 58 · Major gaps Desktop appFree + model costs ★ 345 17d ago Rust OpenAI API · Claude API
Poracode 55 · Major gaps Desktop appFree + model costs ★ 121 today TypeScript ChatGPT · Claude Code
Pi Agent Desktop 52 · Major gaps — ★ 323 13d ago TypeScript OpenAI API

How does FollowAgents rate this agent?

FollowAgents source review · FARS-2.1
Major gaps
40/ 100 5-point scale 2.0 / 5
This review covers the source as of Aug 09, 2026. The project has shipped 14 stable release(s) since, which have not been re-reviewed. The score and verdict describe the reviewed version only. Recent releases ↓
Trust 8/29
Reliability 6/14
Adaptability 9/18
Convention 8/18
Effectiveness 6/13
Verifiability 3/8
Why each dimension lost points
Trust8 / 29 · 1.4/5

Evidence shows: README mentions self-hosted relay, zero-knowledge, local key derivation; SECURITY.md has security policy and vulnerability reporting; Cargo.toml includes encryption dependencies (AES-GCM, Argon2id). However, no explicit least-privilege design, user confirmation mechanisms, data flow transparency, sensitive data handling details, dependency security audit, external effects control, or rollback mechanisms were found. Deductions: insufficient evidence or absence in these areas.

Reliability6 / 14 · 2.1/5

Evidence shows: Cargo.toml has workspace dependency management, CI includes tests and build checks, README has performance metrics. However, dependency availability not verified (e.g., git dependencies), failure messages not documented. Deductions: insufficient evidence for dependency availability and failure messages.

Adaptability9 / 18 · 2.5/5

Evidence shows: README clearly identifies target users (developers, office workers) and scenarios (coding, office work), has customization tiers (L1-L4), supports multiple platforms. However, capability boundaries not explicit, trigger precision (e.g., command triggers) not described, environment fit partially mentioned. Deductions: insufficient evidence for capability boundaries and trigger precision.

Convention8 / 18 · 2.2/5

Evidence shows: README has clear structure, installation instructions, license (MIT), version number (0.2.16). However, missing changelog, known limitations, naming stability, and maintenance responsibility not explicit. Deductions: insufficient evidence in these areas.

Effectiveness6 / 13 · 2.3/5

Evidence shows: README claims output usability (e.g., generating documents, code), has performance metrics (KV cache hit rate, flashgrep speedup), but no cost-benefit analysis. Deductions: insufficient evidence for output usability and cost-benefit.

Verifiability3 / 8 · 1.9/5

Evidence shows: README has benchmark data (SWE-Bench-Pro), but no reproduction method or raw data, facts and inferences not clearly separated. Deductions: insufficient evidence for verifiability.

Risks and how to mitigate them
  • Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
  • Publisher identity unverified; exercise caution regarding supply chain risks.
  • Dependencies include git dependencies (tauri-runtime etc.); ensure supply chain security.
  • Performance metrics in README (e.g., 98.67% KV cache hit rate) are single-run results and may fluctuate; do not treat as fixed promises.
  • No explicit least-privilege or user confirmation mechanisms found; be cautious when granting permissions.
Evidence confidence: Low Reviewed Aug 09, 2026 Reviewed revision c6badb77888f New commits since this review; the score may not cover them
See the full review method →

Recent releases

Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.

  1. v1.0.9 OpenBitFun 1.0.9
    Release notes excerpt
    OpenBitFun 1.0.9 桌面端更新 - 多设备工作台:可在侧边栏、会话标签和 Agent companion 中浏览其他设备的工作区与会话,支持跨设备打开项目、创建项目、重命名设备和管理会话。 - 远程工作区与控制:SSH 支持 ProxyCommand 与预认证隧道,改进 SFTP 恢复、远程搜索和终端链接;CLI 可提供文件、Git、会话用量及远程控制能力,并支持向其他设备派发任务。 - 移动端控制:新增设备模型角色设置、命令面板、会话用量查看和子 Agent 实时任务卡片;支持一键发送下一轮建议,并改进会话与设备切换体验。 - Generative UI:集成 OpenUI Lang 流式交互界面,提供组件库、HTML 渲染、交互状态与能力调用,跟随 OpenBitFun 明暗主题和语义颜色,并改进渲染失败后的修复体验。 - Agent 与扩展:项目 Hooks 可在远程工作区运行,Python 程序化工具调用可用,Claw 与 Cowork 默认支持 Generative UI,并扩展 Pages 管理能力。 - 图片生成与 ACP:ChatGPT 订阅使用独立图片请求;外部 ACP 会话创建时显示启动进度,失败时提供错误提示与重试,减少初始化元数据的重复读取。 …
    View on GitHub ↗
  2. v1.0.8 OpenBitFun 1.0.8
    Release notes excerpt
    What's Changed * feat: add threshold-based model selection (default 0.7) by @xuanaozhang1-coder in https://github.com/GCWing/OpenBitFun/pull/2884 * feat(router): align token-aware context compression by @xuanaozhang1-coder in https://github.com/GCWing/OpenBitFun/pull/2945 * fix: improve DeepSeek Harness history imports by @kev1n77 in https://github.com/GCWing/OpenBitFun/pull/3501 * docs: refresh README media and community links by @GCWing in https://github.com/GCWing/OpenBitFun/pull/3510 …
    View on GitHub ↗
  3. v1.0.7 OpenBitFun 1.0.7
    Release notes excerpt
    下载 / Downloads 桌面端 / Desktop - Windows x64: [安装包](https://github.com/GCWing/OpenBitFun/releases/download/v1.0.7/OpenBitFun_1.0.7_windows-x86_64-installer.exe) - macOS: [Apple Silicon](https://github.com/GCWing/OpenBitFun/releases/download/v1.0.7/OpenBitFun_1.0.7_aarch64.dmg) | [Intel](https://github.com/GCWing/OpenBitFun/releases/download/v1.0.7/OpenBitFun_1.0.7_x64.dmg) …
    View on GitHub ↗

FAQ

Can I use my own model provider?
Yes, you can create multiple model configurations in Settings, selecting different providers and models.
How does multi-device control work?
A self-hosted relay is used; clients derive keys locally, and the server only stores hashes and encrypted material, ensuring zero-knowledge.
How do I create Mini Apps?
You can generate task-specific interfaces via custom agents, MCP, skills, or by writing Mini App code directly.
Does BitFun work offline?
Basic functions require network to call model APIs; multi-device sync needs your self-hosted relay, which must be deployed by you.
How can I contribute?
Pull requests to the main branch are welcome, following the guidelines in CONTRIBUTING.md.
View on GitHub ↗ Install ↓

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related agents