Cloud coding agents: one prompt, one pull request
Run coding agents in sandboxed cloud environments instead of on your laptop — provider-aware (Codex, Claude Code, or our runner), skill-attachable, triggerable by schedules and webhooks.
A coding agent is only as useful as the machine it runs on. Run it on your laptop and it competes with your editor, dies when you close the lid, and can't be triggered by a webhook. app.nz runs coding agents in the cloud: you give a repo and a prompt, we provision a sandboxed environment, run the agent to completion, and open a pull request.
One prompt, one PR
app agent run --prompt "add a /health endpoint with a test" --repo me/apiThat launches a cloud coding agent against me/api. It clones the repo into an isolated workspace, runs the model with file, shell and search tools, makes the change, runs the tests, and pushes a branch with a pull request you can review like any other. No local setup, no long-running process on your machine.
Provider-aware: bring the agent you like
Different agents are good at different things, so app.nz is provider-aware rather than locked to one. The Agent SDK can dispatch the same task to Codex, Claude Code, or our own runner:
app agents-sdk run "refactor the auth module" \
--source openpaths --repo me/api --skills jj,visualbench--skills attaches reusable skills — self-contained capabilities like driving a browser via visualbench or working a GitHub flow — so the agent starts with the right tools already loaded instead of rediscovering them.
Why cloud beats local for agents
- Parallelism. Launch ten agents on ten repos at once; your laptop runs one.
- Triggers. A cloud agent can be started by a schedule or a webhook, not just by you sitting at a terminal.
- Isolation. Each run is sandboxed — an agent running
rmor installing dependencies can't touch your machine. - Reproducibility. The environment is defined, not "whatever was on my path that day."
- Metering. Every run is billed against prepaid credits you can cap, so an agent loop can't surprise you.
Watch it work
Agents stream their progress — tool calls, file diffs, test output — to the /agent studio while they run, and the final diff is a normal PR. List your runs from the CLI:
app agent listIt's the same toolchain the rest of app.nz is built with. Pair it with the MCP server and your agent can reach the whole model gateway through one connector, then ship the result as a pull request.