Integrations
Use smol machines from tools you already run. Each integration below starts ordinary smol machines, through the smolmachines SDK, the smolvm CLI or the Cloud API: each one its own Linux guest, with branching where the integration uses it.
Most of these live in another project’s repository. The Code column says where, and that project’s issue tracker is the place to report problems with the integration itself.
Testing
| Integration | What it does | Code | Runs on |
|---|---|---|---|
| e2e by TesterArmy | Runs each web test attempt in its own branch of a warm, signed-in Chromium machine | @e2e-dev/smol, in TesterArmy’s repository, written by smol machines | Local engine |
Agent evaluation and RL
| Integration | What it does | Code | Runs on |
|---|---|---|---|
| Harbor | -e smol runs each trial in a machine branched from one warm machine per task environment | environments/smol.py, built into Harbor, written by smol machines | Local engine or smol cloud |
| AgentEnv Framework by Scale AI | A sandbox provider that runs AgentEnv agents and environments in local machines | smol_vm, in Scale AI’s repository, written by smol machines | Local engine |
| NeMo Gym | A sandbox provider: install smolmachines[nemo-gym] and select smol in NeMo Gym’s sandbox configuration. An image mapped in checkpoints forks a prepared, running machine | smol.nemo_gym, in the smol Python SDK | Local engine or smol cloud |
| verifiers by Prime Intellect | smol.verifiers.SmolRuntime, a runtime where each rollout runs in its own machine, or in a fork of a running golden machine. Install smolmachines[verifiers] | smol.verifiers, in the smol Python SDK | Local engine or smol cloud |
Agents and orchestration
| Integration | What it does | Code | Runs on |
|---|---|---|---|
| exo | --sandbox-provider smolvm runs an exo agent’s sandbox as a microVM | sandbox_provider/smolvm, in exo’s repository, written by smol machines | Local engine |
| Ray Sandboxes | A backend for ray.experimental.sandbox, with live branching of a running sandbox. Install smolmachines[ray] and call ray.init(runtime_env=ray_sandbox.use_smol()) | smol.ray_sandbox, in the smol Python SDK | Local engine or smol cloud |
| Sandbox orchestration harness by the Temporal community | A smol cloud compute provider, with checkpoints and restores | sdk/compute/smolcloud, in the harness repository, written by smol machines | smol cloud |
Ray’s sandbox API is experimental in Ray itself, so its shape can change between Ray releases.
Agent tools
| Integration | What it does | Code | Runs on |
|---|---|---|---|
| MCP server | npx -y smolmachines-mcp gives any MCP client tools to create a machine with a project copied in, run commands, branch, checkpoint, restore and export | smolmachines-mcp, in the smol repository | Local engine, or smol cloud with SMOL_CLOUD_TOKEN |
Add an integration
To propose an integration or report one that is missing from this page, open an issue on smol-machines/docs.