Compatibility
Rivet is an alpha. Use this page to check your environment and understand where testing is still incomplete.
Runtime and installation
| Environment | Support and testing |
|---|---|
| macOS and Linux, Node 22 or 24 | CI covers the regression suite and installed-package lifecycle. |
| Native Windows | Spawned CLI execution is unsupported. |
| WSL | Not yet tested as a complete environment. |
| Yarn Classic 1.22 | Registry, source and tarball installation tested on macOS. Modern Yarn does not provide yarn global. |
| npm | Registry installation tested on macOS; installed-package lifecycle tested in CI. |
| pnpm | Registry global installation and CLI help tested on macOS with pnpm 11.18.0. |
| Project-only runtime | Installed-runtime checks cover setup and local verification without a global Rivet command. |
| Registry package | Published on npm as @agilno-tech/rivet; the default latest tag contains alpha software. See installation. |
Node runs Rivet, not your application. Python/Django, Node.js and other projects use their own configured checks and dependency procedures. Automatic discovery does not recognize every framework or workspace layout; review setup and provide explicit commands when necessary.
Coding harnesses
| Entry point | Tested scope |
|---|---|
| Claude Code CLI | Small spawned tasks on macOS, including Claude 2.1.280. |
| Codex CLI | Small spawned tasks on macOS with Codex 0.155.0-alpha.16. |
| Claude and Codex desktop | Host command support exists; complete desktop workflows have not been validated. |
| Other coding harnesses | Require local file/terminal tools, Rivet instructions and permission to operate on the project. End-to-end compatibility must be checked for the harness. |
CLI versions above record test evidence, not a version lock. Rivet probes required capabilities and stops when an installation cannot satisfy them. See harness compatibility.
Local trials with both CLIs completed documentation tasks in a Node project with build, 40 tests and type-check, and a Docker Django project with a system check. The trials used prepared disposable environments; the Django plans were operator-supplied and the Claude runs used reviewed larger token allowances. They do not establish full application testing or unassisted first-user setup. Full active-host and desktop workflows still need testing.
Integrations, models and delivery
Supported commands and their restrictions are described in integrations, models, repository inspection and delivery.
Automated coverage exists for these interfaces. Live validation across provider accounts is incomplete. Test your configured connection in a suitable project before relying on it. In particular:
- Claude Worker budgets use native token counters, including cached context. Codex direct-result usage is model-reported and can understate consumption; do not rely on it as an enforced usage or billing limit. See task budgets.
- API/local text models provide advice, not autonomous implementation with coding tools. They do not enforce a monetary cap.
- GitHub and GitLab merges support specific repository policies. Unsupported or unreadable policies block merging.
- Bitbucket merging is manual.
- GitHub Actions deployment requires an explicitly configured workflow and separate approval.
- Shared cross-user memory is not implemented.
For installation problems or failed tasks, use troubleshooting.