Supported agents

opub integrates with leading agents for coding and repository maintenance. Each consumes compute through a capped key and surfaces a linked MCP session. The right choice depends on what you want the donated compute to do.

The opub CLI is open source.

Coding agents

Claude Code, Codex, Vibe, OpenCode, and Continue are agentic coding systems. Use any of them when the work is writing, editing, debugging, reviewing code, or running source checks in a local development context.

opub setup claude --project owner/repo --compute-key-id ck_...
opub setup codex --project owner/repo --compute-key-id ck_...
opub setup vibe --project owner/repo --compute-key-id ck_...
opub setup opencode --project owner/repo --compute-key-id ck_...
opub setup continue --project owner/repo --compute-key-id ck_...

They run inside your existing development environment. Choose based on the workflow and model provider you prefer or already use. Continue CLI is especially useful for headless PR review, bug checks, security notes, and documentation checks from the command line or a script. See Available models for the full list of supported models and recommended picks per provider.

Repository management

GitHub Copilot CLI extends the agent with built-in GitHub repository tooling. It can close issues, review pull requests, triage discussions, and interact with other GitHub resources directly — work that goes beyond editing files locally.

opub setup copilot --project owner/repo --compute-key-id ck_...

Copilot CLI is the default maintainer agent and the strongest fit when donated compute should go toward keeping the repository itself healthy rather than local development.

Running a session

Once configured, start a funded session with opub run <agent>:

opub run claude
opub run codex
opub run vibe
opub run opencode
opub run copilot
opub run continue -- -p "Review staged changes for obvious bugs" --silent

Each session injects credentials for that process, refreshes the MCP session state with runtime launch proof, and loads the bundled skill, plugin, or generated agent config. See the CLI guide for install and key setup.

Using a key directly

An agent is not required. A compute key is a capped provider credential for opub's allowed model providers. You can paste it into a compatible model playground, wire it into your own tooling, or configure it manually in a tool that accepts an OpenRouter key.

Watchlist tools

Other CLIs are worth watching. Cursor CLI supports automation and code review workflows, but its official CLI authentication uses Cursor credentials. Gemini CLI uses Google login, a Gemini API key, or Vertex AI credentials. Those are useful agent frameworks, but they are not opub compute-key compatible unless a supported OpenRouter-compatible path is verified.

The difference is session linking. When you run through opub run, the CLI creates a local MCP session and gives the launched process runtime proof that ties it to a verified project and compute key. That verified session signal lets donors see that their compute went toward the project they funded — not as proof of specific work, but as a signal that the funded key was active on the right project.

If you use the key directly, spend still accrues to the project, but the local MCP session link is not present. That is often fine. It matters most when your project has donors who care about that signal.