Is this a separate built-in template?

This is a workload recipe based on the devbox profile. It adds preparation and execution guidance, not a separate built-in image or a new backend runtime.

Can I run the command now?

The command runs in a Linux environment where the listed dependencies and input files are prepared. cpuOS SDK examples describe a proposed contract; confirm runtime access and package versions before integration.

How should I choose CPU and memory?

Use the starting profile to run a representative fixture, measure peak memory and elapsed time, and add room for package installation, worker processes and larger inputs. Enforce a task timeout separately.

Coding agent sandbox recipe

A Dev box recipe for agent-driven repository edits: baseline tests, bounded command execution and patches that a human can review before publishing.

git

Repository toolchain

Test runner

When to use this profile

Give a coding agent a task, a source checkout and a bounded set of tools. Run baseline checks before edits so existing failures are distinguishable from regressions. After each meaningful change, return the relevant test output to the model. Export the final diff for review; publication is a separate action with separate authorization.

Prepare a repeatable environment

Record the base commit and remove unrelated working-tree changes from the input archive.
Set a command timeout, output budget and maximum number of agent iterations.
Install the project's pinned dependencies and record baseline failures.

Command inside a prepared Linux guest

git -C /work/project diff --stat

Return results the agent can use

A unified diff and a list of files changed.
Validation results for the relevant tests and build.
A concise explanation linking each change to the task.

Use a seeded bug with an existing regression test. Confirm the agent fixes it, preserves unrelated files and returns the patch without pushing it.

Resources and boundaries

Start with 4 vCPU and 8 GB of RAM, then measure peak memory and task duration on a representative fixture. These are workload planning values, not a benchmark or a provisioned configuration. Use the sandbox cost calculator to estimate running time and retained snapshots.

An agent reading repository instructions can be steered into running unsafe commands.
Repository access should be read-only unless the workflow explicitly authorizes publication.
An unchanged test suite is evidence about tested behavior, not proof that every change is correct.

Keep model inference separate from this execution profile. A hosted model or gpuOS can decide the next action while the CPU environment runs it. The quickstart describes the account workflow and the proposed runtime contract.

Questions

Is this a separate built-in template?: This is a workload recipe based on the devbox profile. It adds preparation and execution guidance, not a separate built-in image or a new backend runtime.
Can I run the command now?: The command runs in a Linux environment where the listed dependencies and input files are prepared. cpuOS SDK examples describe a proposed contract; confirm runtime access and package versions before integration.
How should I choose CPU and memory?: Use the starting profile to run a representative fixture, measure peak memory and elapsed time, and add room for package installation, worker processes and larger inputs. Enforce a task timeout separately.

Plan your coding agent task

Create a workspace and choose a profile. Connect an execution backend before running code.

Create account See pricing

gpuOS · where models think

Need the model too? Run it on gpuOS

gpuOS serves open models on your own GPUs behind one OpenAI-compatible API. The model reasons on gpuOS, the agent acts in a cpuOS sandbox.

gpuos.si