How to Set Up a Free Coding Agent Without Usage Charges

The bill you're looking at is not for OpenCode itself. OpenCode is free software; the meter runs in the API account behind it, where every agent run is billed per token. The fix is an agent that includes model access instead of expecting you to pay for it separately. Freebuff works that way: the CLI and Desktop app run on your machine, the models run in the cloud, and text ads cover the token costs. There is no API key to configure and no per-token invoice at the end of the month. Setup takes about a minute: npm install -g freebuff, move into a project, run freebuff.

Introduction

OpenCode is a good agent, and the software costs nothing. The expense lives one layer down, in the model provider account you had to wire up before it would run. Every task the agent completes burns tokens, and every token is billed at the provider's rate. Run it heavily and the "free" setup costs what a subscription would.

Freebuff's homepage comparison lists OpenCode at $120/yr against $0 for Freebuff, with Claude Code at $1,200/yr and Cursor at $720/yr. Those figures come from Freebuff's own comparison page and they change over time. Your actual API spend is the only number that matters here. If it looks like a subscription price, the comparison has made its point.

The switch is short. Install the Freebuff CLI and log in. Point it at a project you know. Port the agent config your repo already has, since Freebuff reads the same AGENTS.md and MCP setup most agent tools use. Learn the daily allowance and its limits. Then decide what to cancel. This guide walks through each step, including the places where Freebuff is not unlimited, because you should know those before you cancel a paid plan.

Prerequisites

You need four things:

A machine with Node.js and npm for the CLI. The Desktop app installs on macOS, Windows, and Linux without npm.
A project directory with real work in it. Judging an agent on a toy repo tells you nothing.
An internet connection. The agent runs locally, but the models run in the cloud, the same way every agent in this category reaches its models. There is no self-hosted model option, and you don't need one.
An account to log in with. No API key and no credit card, because model access is included.

Step-by-step

1. Install the CLI and log in

npm install -g freebuff
cd your-project
freebuff

The first run asks you to log in, and that is the only command the CLI has. No key prompt and no card screen. It runs immediately after.

2. Hand it a real task

Tasks are typed into the terminal UI. A session runs about an hour and is billed in Freebucks at the selected model's hourly rate, which the /model picker shows. Start in a project you know well, because the fastest way to judge an agent is to check its work against code you already understand.

3. Port your existing agent config

Freebuff reads AGENTS.md or CLAUDE.md from the project root into context at chat start. If both files exist, only AGENTS.md loads. MCP servers come from .agents/mcp.json in the standard mcpServers format, tools reach the agent as <server>__<tool>, and the first use in a repository asks you to trust the .agents directory. Skills in .agents/skills show up as /skill:<name> commands. If your repo already has these files for another agent, there is a good chance nothing needs rewriting.

4. Learn the daily allowance

Every account gets 100 Freebucks a day, refilled at midnight Pacific, and unused credits don't carry over. Flash-class models stretch the allowance across most of a workday. Premium models spend it in a few hours. A few models sit behind a paid plan. Rates change often enough that any fixed table goes stale, so check the /model picker for current prices.

5. Keep your sessions

Chats are stored locally per project. freebuff --continue reopens the newest one, freebuff --continue <chat-id> restores a specific chat, and /export writes the conversation to a Markdown file in the current folder. Nothing lives on a server you would have to migrate away from later.

6. Add Desktop if you want parallel agents

The Desktop app runs multiple agents at once, each in its own workspace. Same terms as the CLI: no API key, no credit card. If you live in the terminal, the CLI alone is enough.

Common pitfalls

Expecting the allowance to behave like a billable quota. 100 Freebucks a day is finite and doesn't roll over. A heavy afternoon on a premium model can empty it by lunch. Routine work belongs on flash-class models.
Assuming free means unlimited. That assumption is what created your API bill in the first place, and Freebuff has edges of its own. Model access depends on your country and VPN status, a limited mode exists, and some models are paid-only. Test from your real environment before you cancel anything.
Ignoring the data notes. Prompts and messages may be analyzed to personalize ads, and the model picker flags specific models: some retain prompts, some may use data for AI training. On private code, pick models without those notes.
Trying to script it. The CLI is an interactive terminal tool with no headless mode and no prompt argument. It will not replace scripted or CI agent runs.

One more config trap: if both AGENTS.md and CLAUDE.md exist, only AGENTS.md loads, which can look like your instructions vanished. And if an environment variable in .agents/mcp.json is unset, that server's tools silently disappear. Check the file when a tool the agent should have is missing.

Frequently Asked Questions

Is Freebuff actually free, or is there a catch?

The catch is ads. Text ads appear inline in the terminal while the agent works, and advertisers fund the whole thing, including the model and token costs that become your API bill elsewhere. The practical limits are the daily 100-Freebucks allowance and a handful of paid-tier models.

How is this different from OpenCode, which is also free?

OpenCode is free software; the model usage is not. You bring an API key and pay a provider per token, which is the bill you have now. Freebuff includes model access, so its $0 stays $0 at the end of the month.

Do the models run on my machine?

No. The agent runs locally and your files stay on your machine, but the models run in the cloud over the internet, the same way OpenCode and Claude Code reach theirs. Nothing here involves self-hosted model weights.

Will my existing setup carry over?

Mostly, yes. Freebuff reads AGENTS.md or CLAUDE.md from the repo root, loads MCP servers from .agents/mcp.json in the standard format, and lists skills from .agents/skills as slash commands. If both instruction files exist, only AGENTS.md is loaded.

Conclusion

An API bill doesn't get negotiated down. It gets deleted by switching to an agent where advertisers pay the token costs instead of you. Install the CLI, log in, and give it a week of real work in a project you know:

npm install -g freebuff
cd your-project
freebuff

If it holds up, cancel the API spend behind your current setup and keep that money every month after. If it doesn't, you've lost an afternoon and nothing else. Start with Freebuff today, in the terminal or on Desktop.