Software development
Work in a real repository to understand the codebase, plan and implement changes, debug, test, and review them, and carry the work through to a deliverable result.

An open, capable agent system built for real work.BitFun combines a high-performance agent runtime written in Rust with a polished desktop application. It pairs the depth of a Code Agent with open, general-purpose capabilities for work beyond software development.


A cross-platform agent workspace with an expanding range of device experiences.
Give BitFun a real codebase, source materials, or a clear objective. It carries the task from understanding and execution through verification to delivery.
Work in a real repository to understand the codebase, plan and implement changes, debug, test, and review them, and carry the work through to a deliverable result.

Research source materials, synthesize findings, and produce presentations, documents, spreadsheets, PDFs, meeting notes, and reports.

Extend BitFun around your domain knowledge, data, and processes to support additional professional use cases.

BitFun is being developed around continuous execution, cross-device collaboration, open customization, efficiency, and interaction quality.
Continuous execution
Plan work during the day, let tasks run on a server overnight, and review the results in the morning.

Cross-device collaboration
Extend work from desktop and browser to mobile, wearables, and other devices so tasks remain accessible and continuous across devices.

Open customization
Extend BitFun with custom Agents, MCP integrations, Skills, Mini Apps, or source-level changes to create workflows for your own needs. Community contributors have already built versions for short-form drama, media production, and other specialized uses.

Efficiency, quality & cost
Optimize for task efficiency, result quality, and cost together.

Visually polished and easy to use
BitFun continually refines visual hierarchy, motion feedback, and interaction flows so its desktop GUI remains clear, responsive, and easy to use.

BitFun provides a runtime, context management, tools, permissions, execution environments, product interfaces, and cross-platform infrastructure. Builders can add domain data, specialized tools, and workflows to create agent products for specific use cases.

Rust runtime · Cross-platform support · Local and remote environments
Context · Tools · Permissions · Execution · Sessions
Agents · MCP · Skills · Hooks · Mini Apps
Data · Decision logic · Workflows · Product experience
Community contributors have built specialized versions of BitFun

Interfaces, interaction, and product experience

Generative content production

Audiovisual production workflows

Additional versions continue to emerge
A composable agent runtime for multiple product surfaces, with local execution, relay-mediated peer sessions, and SSH-based remote workspaces.

Windows · macOS · Linux

Browser-based remote access

CLI / TUI

Mobile Web · Messaging

Watch · Glasses · In development
A shared access layer that connects product surfaces to runtime capabilities.
Turn requirements into reliable software
Design · build · debug · verify
Create office deliverables from source material
Slides · documents · spreadsheets · PDFs
Investigate complex questions
Search · verify · analyze · synthesize
Continue toward a defined objective
Plan · execute · verify · adapt
Support additional specialized tasks
In development

Maintains the context needed to understand and continue work across sessions

Coordinates task orchestration, scheduling, and resources during execution

Executes plans through governed tools and returns structured results

High performance · Memory safety
Rust kernel
Model access · Protocol compatibility
Request orchestration · Multi-model support
Files · Events · Gateways
Shared capability services
Isolation · Permission controls
Logs · Metrics · Audit trailsCompletion benchmarks, KV cache reuse, and large-codebase search measurements provide separate views of task results, token economy, and retrieval speed.
Measured in the same SWE-Bench-Pro round; a higher hit rate indicates greater context reuse but does not by itself quantify end-to-end cost.
Measured with flashgrep on repositories at Chromium scale; the same evaluation reported an average speedup of about 36.1×.
The completion scores are initial, single-run results using DeepSeek-V4-Pro and may vary with task sampling, model versions, runtime environments, and single-run variance. They are an early capability signal, not a fixed ranking. The cache and search figures reflect the specific evaluations described above; fuller cost and latency reporting is still planned.
Explore the source, product website, and community discussions to follow progress or share feedback.