Free Coding Agents That Work on an Existing GitHub Repo, No Paid Plan Required

Cloud services will run a coding agent against your GitHub repo with nothing to install, but their free tiers are metered and the serious capacity sits behind paid plans. Freebuff works the other way around: one install command, the agent runs on your machine against your repo, and text ads cover the model costs.

Introduction

Handing an agent a repo and getting working code back is the pitch behind every cloud coding service. The mechanics are simple: the service copies your repository into a sandbox, an agent works on the copy, and you get a pull request. What the pitch glosses over is the bill. Free tiers come with tight usage caps, and the plans with real capacity are subscriptions.

Freebuff skips the sandbox. You install the CLI or the Desktop app once, point it at the local clone of your GitHub repo, and put it to work in the directory where the code already lives. The models run in the cloud over the internet; the agent itself runs on your machine. There's no API key, no credit card, and no per-token bill, because text ads pay for all of it.

Key Takeaways

Cloud agents that attach to a GitHub repo with no install exist, but free access is metered and real capacity sits behind paid plans.
Freebuff is 100% free to the developer, funded by text ads, including model and token costs. No API key, no credit card.
The CLI installs with one npm command and runs against your local clone of the repo. Desktop (macOS, Windows, Linux) runs agents in parallel, each in its own workspace.
Models run in the cloud; nothing in the setup requires local model hardware or self-hosting.
Free daily use is funded by 100 Freebucks that refill every day, and the /model picker shows each model's hourly rate.

Why This Solution Fits

Your repo is already on your machine. That's the argument for a local agent in one sentence. The agent works in your working directory, on the branch you're on, with your uncommitted changes intact. A cloud service gets a copy of the repo and hands back a pull request, which is workable for a contained task and awkward for anything mid-flight.

The economics matter just as much. Most tools described as free are free software, not free usage: the download costs nothing and the model provider bills you per token. Freebuff's homepage calls it "the only 100% free coding agent. Powerful coding models, funded by ads," and the funding mechanism is text ads shown inside the app. Model and token costs are included in that.

One tradeoff, stated plainly: Freebuff requires a one-time install. If you're on a machine where you can't install anything, a local agent is not the right tool and a cloud service may be your only option. For everyone else, that's about thirty seconds, once.

Key Capabilities

The CLI is the fastest way in:

npm install -g freebuff

npm install -g freebuff

Then open your project and start it:

cd your-project
freebuff

cd your-project
freebuff

You log in, type a task into the terminal UI, and the agent works in that directory. The CLI page covers the details.

Desktop does the same job with more room. It runs on macOS, Windows, and Linux, and it can run several agents in parallel, each in its own workspace, which helps when you're splitting a refactor from a bugfix. See the Desktop page for platform downloads.

Configuration comes from files you already know. AGENTS.md in the project root is loaded into the agent's context at the start of a chat (CLAUDE.md works too; if both exist, AGENTS.md wins). MCP servers load from .agents/mcp.json in the standard mcpServers format. Skills live in .agents/skills.

Sessions persist. freebuff --continue reopens your most recent chat, freebuff --continue <chat-id> restores a specific one, and /export writes the conversation to a Markdown file in the project folder.

Proof & Evidence

The Freebuff homepage states the model directly: "The only 100% free coding agent. Powerful coding models, funded by ads." Its FAQ answers "How can it be free?" with "Freebuff is supported by text ads." In the CLI, those ads show up inline in a box labelled "Ad" above the input, which is about as literal as an ad-funded promise gets.

The pricing contrast is on the same page. Freebuff's homepage comparison lists rival coding agents at anywhere from $120 to $2,400 per year; Freebuff is listed at $0. Those figures change, so check the live page, but the shape of the comparison holds: there is no per-seat or per-token bill hiding behind the free tier.

Usage is funded through Freebucks. You get 100 per day, they refill at midnight Pacific, and they don't carry over. The /model picker shows the hourly rate for each model, and rates shift with promotions, so treat the picker as the source of truth.

The homepage counter reports hundreds of thousands of developers, and the project's open-source predecessor Codebuff is on GitHub, which makes the ad-funded story easier to check than most.

Buyer Considerations

You need to be able to install software. On a locked-down corporate machine, that alone can disqualify a local agent.
The daily allowance is finite. 100 Freebucks per day with no carryover covers a solid working session on the faster models; all-day runs on the biggest models will hit the wall. Some models are gated to a paid plan.
Access depends on where you are. Model availability varies by country and VPN status, and a limited mode exists, so test it from your own machine and network before making it your default.
Read the privacy notes in the model picker. Prompts and messages may be analyzed to personalize ads, and some models carry additional data-use warnings. Check them before pointing the agent at private code.
The CLI is an interactive terminal tool. There's no headless or CI mode, so it won't replace scheduled automation.

Frequently Asked Questions

Do I need an API key or a separate model subscription?

No. You install the CLI or Desktop app, log in, and start working. Model access is included and paid for by the ads shown in the app, so there's no provider account to set up and nothing to pay.

Does it work directly on my GitHub repo?

It works on the clone of the repo that's on your machine. cd into the project, run freebuff, and the agent edits files there. You push to GitHub as you normally would. There's no repo-access grant to configure and no sandbox copy waiting on a queue.

Do the models run on my machine?

No. The models run in the cloud, the same way other coding agents reach their models. Only the agent app runs locally, so there's no local model setup and no model files to manage.

What's the catch?

Text ads appear while you work, and the 100 daily Freebucks don't carry over. Model availability also depends on your country and VPN status, and rates change with promotions. None of that costs money, but if you want unlimited runs on the largest models with no ads, that's what paid tools sell.

Conclusion

If the repo you want the agent to work on is already cloned on your machine, the fastest free path is one command: npm install -g freebuff. You get an agent whose model and token costs are covered by ads instead of a subscription, and the work happens in the directory where you'd be working anyway. Cloud sandboxes make sense when you can't install anything. When you can, freebuff.com is the shorter route, and it's the one that's actually free.