Run coding and research AI agents in your terminal and browser without subscriptions or API keys
It enters your project folder, locates the necessary files, modifies the code, and runs tests in a single process. Freebuff offers five products tailored for coding, app building, and research, utilizing free models supported by text ads. You can start without subscriptions, credits, or API keys, making it suitable for individual developers and learners.
Five product configurations by environment
Freebuff is divided into five forms based on the usage environment. Desktop runs multiple agents in parallel locally, while CLI performs coding tasks directly in the terminal. Web is used for building and deploying full-stack apps, and Cloud connects to GitHub repositories to run agents. Chat focuses on researching and thinking with AI. To use the CLI, install it with the npm install -g freebuff command, then run freebuff in your project folder and describe the desired task. Freebuff then finds relevant files, applies changes, and runs validations appropriate for the project.
Free model catalog and access permissions
Freebuff provides various AI models, each with an optimized use case. GLM 5.3 Flash and DeepSeek V4.1 Flash are the default models, available without limits, with strengths in deep reasoning and fast coding, respectively. GPT-6 Luna is an all-rounder model capable of image generation, while Solar Mini 4 and Solar Pro 4 are Upstage models that support a 524K context. Some models are accessible only with paid plans or in specific regions, and using a VPN restricts you to a limited set of models. All models are paid for using an internal currency called Freebucks, with free users receiving a base of 25 Freebucks per day (20 when using a VPN).
Task handling via specialized agents
Freebuff does not process all tasks with a single model and prompt; instead, it deploys specialized agents suited to the purpose. For coding tasks, a file exploration agent first maps the relevant parts of the project, an implementation agent performs the changes, and a review agent checks the results. When research or browser usage is required, agents investigate documents or test apps in an actual browser. In the Desktop environment, concurrently running agents are isolated in separate workspaces to prevent conflicts. Web and Cloud provide sandboxes, previews, terminals, and deployment workflows to perform tasks in hosted environments.
Data usage scope and privacy policy
User data is used for AI training only when a model or feature explicitly permits it. Freebuff uses prompts, messages, agent trace records, code, files, and repository data to provide its service and may analyze prompts and messages for ad personalization. However, separately uploaded files or connected repositories are not shared with advertisers. Limited partners may evaluate connected Cloud repositories but do not use them for other purposes, share them broadly, or use them for training. You can find more details in the privacy policy.
Contribution methods and technical foundation
Freebuff is a monorepo built with TypeScript and Bun, welcoming contributions to products, agents, tools, documentation, and runtimes. This project is built on Codebuff, an open-source multi-agent framework, so you should refer to the Codebuff documentation and @codebuff/sdk if you want to create custom agents or integrate them into other applications.