CodebuffAI/freebuff

Freebuff is worth a look if you want an agentic coding assistant without provisioning a model account first, but read the data-use notice before pointing it at a private repository.

The problem

A developer wants an agent that can map a codebase, edit files and run project checks from the terminal. The normal route means signing up with a model provider, creating an API key, and wiring it into a coding tool before the first useful prompt. Freebuff collapses that setup into one global npm install and a launch command in the project directory project instructions. What changes is the provisioning step, not the asking: you still describe the goal, and the agent still finds files, makes edits and runs checks.

What's different

The mechanism is a curated model catalog whose usage is covered by in-app text ads. There are no subscriptions, credits or API keys, and access is organized around daily session or hourly balances rather than per-token billing README. On top of that, Freebuff routes work through specialized agents for file finding, planning, implementation, review and research instead of one prompt against one model.

The practical benefit is trying agentic coding quickly and switching between several models without separate accounts, including Flash-class models that report no session cost at full access. The trade-off is that the free tier is shaped by advertising and capacity: supported regions get full access while other regions and VPN users get limited access, and the unmetered models are capped by a daily hour allowance, 5 hours normally or 4 behind a VPN free access terms. The README also states that data may train models where a model or feature says so, and that usage data may personalize ads and train ad models, so this is a different bargain from a paid API plan privacy policy.

Who should try it

Good fit: a developer or small team that wants to evaluate an agentic CLI on side projects or open-source work and does not want to manage model credentials. The terminal path is one command, and Freebuff Desktop can also run locally installed Claude Code and Codex agents through an existing provider account, which helps people who already have access desktop. Poor fit: teams that need contractual no-training guarantees for proprietary code, or projects that depend on a specific model that free access does not cover. The README explicitly ties availability to region, tier and current capacity, so a workflow built around a particular model can change under you.

Requirements and limitations

The CLI needs Node.js and npm for a global install, while local development of the project itself additionally needs Bun, Docker and a configured .env.local because it is a Bun monorepo with separate services contributing guide. The Desktop, Web, Cloud and Chat products are hosted entry points, so the CLI is not the only way in web.

Ongoing upkeep means watching what your tier includes: balances are measured in hours with per-model multipliers, earned hours are spent after the daily allowance, and the model picker can change — DeepSeek V4 Pro was retired from the catalog, for example. Some flagships are limited to one session a day or restricted by region. Before adopting it for real work, check whether the model you would rely on has a data-use notice that permits your code to be used for training and ad-related purposes.

HotRepo's take

The only recorded HotRepo observation is a single daily-board appearance on 2026-10-26 at 13,764 stars, with +26 that day. One snapshot cannot establish whether interest persists, so treat the trend as unproven rather than as momentum. The decision-relevant check is the model-specific data-use notice attached to the model you intend to use, since that is what separates Freebuff from a conventional paid coding agent.