GCWing/OpenBitFun
AI EngineeringA desktop agent workspace with a Rust runtime, multiple harness modes, and phone-based remote control for long-running tasks.
What it is
OpenBitFun is an open-source agent workspace that pairs a Rust Agent Runtime with a desktop application. The runtime manages session state, context, and tool execution, with persistent sessions, interruption recovery, long-term memory, context compression, and cache reuse for longer runs. The app targets people who want a general-purpose agent for work beyond coding, not only an IDE assistant.
Four Agent Harness modes change how a task is handled: Minimal for direct collaboration, Standard for multi-step everyday work, Ultimate for complex problems split across agents, and Creative for building Mini Apps or customizing the workspace Four working modes.
Main features
- Cross-device control: switch between connected computers inside the desktop app, and send instructions or approvals from a phone or browser. Tasks continue on the selected host when the controller disconnects. The default path uses the official Relay with the same OpenBitFun account; a local Relay is available for same-network control connection FAQ.
- Mini Apps: describe a need and have the agent build a dedicated interface backed by an agent conversation; built apps can be installed and reused, and there is a Mini App Marketplace.
- Customization: define specialist agents with a chosen model, role, tools, and workflows; connect tools via MCP; turn repeatable processes into Skills; use Hooks to change task execution. Skins are installable, and the UI and runtime source can be modified agent management, Hooks.
- Remote workspaces: SSH into a remote host, jump host, or container; file, terminal, and workspace commands target that environment while the Agent Runtime stays on the OpenBitFun host remote workspaces.
The README reports a DeepSWE v1.1 evaluation with GLM-5.3-Flash at a 64.6% pass rate and 42.9 minutes median runtime (P50), and DeepSeek-V4-Flash at 56.6% and 19.6 minutes. These are the project's own reported measurements, not independently verified here.
Getting started
For normal use, download the Windows, macOS, or Linux package from the official download page, install it, configure a model, and open a workspace. The user guide covers usage, and download verification describes package signature checks. Mobile builds are listed under mobile releases.
To run from source, install Node.js 22.12+, pnpm 10.15.0, the Rust toolchain, and the Tauri prerequisites, then in the cloned repository run:
pnpm install
pnpm run desktop:dev
Desktop development supports frontend hot reload and automatic Rust rebuilds. A separate CLI guide covers the command-line workflow.
Constraints and caveats
- Version 1.0.0 is not data-compatible with legacy 0.2.x releases. Migration uses the OpenBitFun Data Tools, currently only for stable releases 0.2.17–0.2.19.
- Same-account device control relies on the online OpenBitFun identity service for sign-in; LAN mode is not an offline-login option. Private Relay deployments require a matching client build deployment guide.
- Device-to-device payloads are described as end-to-end encrypted, but Relay still handles account, device, and routing metadata. Model-provider traffic, messaging bots, and published Pages are outside that boundary privacy boundaries.
- Core code is MIT-licensed; third-party dependencies and assets keep their own licenses license. Security reports follow the security policy.
HotRepo's take
On the 2026-11-08 Rust daily list, OpenBitFun ranked 4th with 3,966 stars, adding 50 stars that day. It is a single recorded snapshot rather than a sustained series, so it supports that day's position but not a trend. The project's scope is broader than a coding assistant: the Rust runtime, the four harness modes, and device handoff are aimed at keeping long-running work moving across machines and phones, which is the part worth watching as the project matures.
Written by AI from the project's README, updated . The project's own docs are the source of truth.