Durable artifacts for coding agents
Why app.nz stores agent diffs, logs, previews, screenshots, and generated files as artifacts so a run remains auditable after the worker shuts down.
Listen to this article
On-device voiceUses the voice built into your browser; no article text leaves this page.
Audio narration is not supported by this browser.
Coding agents produce more than a final message. They produce diffs, logs, previews, screenshots, generated files, and sometimes deployed sites. app.nz treats those as artifacts so the result can outlive the worker that created it.
That is the difference between an agent transcript and an engineering workflow.
Why artifacts matter
If an agent edits code and then the worker shuts down, the user still needs to inspect what happened. A durable artifact lets the UI show:
- the patch,
- changed files,
- command output,
- test results,
- generated screenshots,
- preview URL,
- final summary.
Without artifacts, the platform is forced to replay logs or ask the user to trust a short text summary.
R2-backed when configured
app.nz can store agent diff artifacts and generated outputs in an R2-backed public or signed object path. In local development and tests, the code can no-op or use inline data paths so the rest of the flow still works.
That fallback is valuable. Artifact plumbing should be testable without a cloud bucket.
Artifacts versus logs
Logs are chronological. Artifacts are named results. A test log tells you what happened. A diff artifact tells you what changed. A screenshot artifact tells you what the UI looked like. They answer different questions.
A good agent UI needs both.
Preview sites
When an agent builds a frontend, the natural final artifact is a hosted static site. app.nz already has the sites primitive, so an agent can build, upload the directory, and attach the resulting URL to the task.
That is why static hosting and coding agents belong in the same platform. The output of one subsystem becomes the input to another.
Security and retention
Artifacts can contain sensitive data. The platform needs explicit choices about:
- public versus private URLs,
- retention windows,
- redaction,
- user ownership,
- deletion when a task is deleted.
The easy mistake is to make every artifact public forever because it simplifies sharing. The better model is to store ownership metadata and choose visibility by artifact type.
The principle
An agent run should be auditable. If it claims it fixed a bug, app.nz should show the diff and the test evidence. Durable artifacts turn "the model said it worked" into "here is the work product."