Can't I just share the repl?

You can, and then your reviewer is looking at an IDE with your code, your secrets pane and your run button, in order to read a document. Sharing the workspace to share the output grants far more than you meant to.

What about deploying it?

A deployment publishes an app, which is the right move when the deliverable is an app. When the deliverable is a report, a deployment means a running service, an ongoing cost and a URL that breaks when the deployment sleeps or is removed.

How does the agent publish?

One POST to /api/v1/reports with the HTML, using a token from the repl's secrets. It returns a URL.

Does the reader need a Replit account?

No. A published report is a page with its own access setting — link-shared, password-protected or restricted to people you name.

What token scope is needed?

`reports:write`, stored in the repl's secrets rather than in the code.

Share what Replit Agent produced

Replit Agent is quick at the jobs that end in a document: read this dataset and tell me what is odd about it, compare these two APIs, write up what the scraper actually collects. The result lands in a repl.

Sharing a repl to share a document is a bad trade. The reader gets an IDE, your source, your secrets pane and a run button, and has to work out which pane the answer is in. Making the repl public is worse — the code goes with it.

Publish the document, keep the workspace

Put a token in the repl's secrets (from API tokens) and let the agent publish:

import os, requests

html = open("report.html").read()
r = requests.post(
    "https://commareports.com/api/v1/reports",
    headers={"Authorization": f"Bearer {os.environ['COMMA_API_TOKEN']}"},
    json={"title": "Scraper coverage review", "html": html},
    timeout=30,
)
print(r.json()["url"])

Tell the agent that this is what finishing looks like — write the HTML, run the publish, reply with the URL — and the repl stops being the delivery mechanism. Full request shape in the API docs.

Deployment is for apps, not documents

A deployment is a running process: it costs money while it runs, it sleeps, and the URL dies when you tear it down. That is a reasonable price for an app and an unreasonable one for a page of findings that will never change again. See hosting an HTML file for the general version of this argument.

What you get instead

Its own access setting, separate from the workspace.
Faithful rendering in a sandboxed iframe — charts, tables, the lot.
Anchored comments on the specific number someone disputes — see commenting on HTML.
A stable URL across revisions. Re-run, update, same link.

Worth knowing

reports:write only — see scoped tokens.
5 MB of HTML per report body.
Nothing re-executes, so a report opened by forty people costs the same as one opened by nobody.

Try it

Free — unlimited reports, commenters and revisions.

Read the API docs →