Add session.fetch for HTTP requests from the page - #943

Preview deployment for your docs. Learn more about Mintlify Previews.

💡 Tip: Enable Automations to automatically generate PRs for you.

No actionable comments were generated in the recent review. 🎉

Walkthrough

The browser and remote session fetch APIs now return standard requests.Response objects. Response bodies use base64 transport to preserve binary data and declared character encodings. HTTP errors remain available through raise_for_status(), while network failures remain JavaScript errors. Documentation adds fetch examples, navigation, SDK references, and a Response page. OpenAPI generation now reuses cached sections by default and supports explicit refreshes.

Estimated code review effort: 4 (Complex) | ~45 minutes

Merge Risk: ⚪ Minimal · up to d88d8

This adds browser-context fetch APIs returning standard response objects, with request validation and byte-preserving response handling. No concrete merge-blocking risk remains.

• Fix all pre-merge checks with AI

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

_{Comment @coderabbitai help to get the list of available commands.}

Confidence Score: 5/5

The PR appears safe to merge; the previous findings are resolved and no new actionable defects remain.

Query parameters are now inserted before URL fragments, GET and HEAD bodies are rejected before browser execution, and the follow-up preserves response bytes while selecting an appropriate text encoding. The feature also includes both unit and browser integration tests.

Important Files Changed

_{Reviews (5): Last reviewed commit: "fix(fetch): preserve response bytes and ..." | Re-trigger Greptile}

Addressed both review findings in 677102d:

• Query parameters are now inserted before any URL fragment (/items#results + params becomes /items?page=2#results), using urlsplit/urlunsplit so an existing query string is preserved too.
• A body on a GET or HEAD request now raises ValueError before the script runs, instead of letting the browser reject it.

Both have unit tests. 307a063 adjusts one existing test that relied on a GET with a JSON body.

@greptileai review

Switched the return type to a standard requests.Response and dropped the custom FetchResponse and FetchStatusError.

• status_code, ok, reason, headers (case-insensitive), text, content, url, json() and raise_for_status() all behave as in requests, and raise_for_status() raises requests.HTTPError.
• requests was already a dependency of notte-core, so nothing new is pulled in.
• Tests updated accordingly; docs regenerated.

@greptileai review

Fixed the failing pre-commit job in this PR.

The docs-sdk-generate hook ran make docs-llms, which rebuilds the API section of llms.txt from the live OpenAPI spec at api.notte.cc. Any API deploy between a local run and CI made the hook rewrite the file and fail, on any PR. Now:

• generate_llms.py keeps the API section between <!-- openapi:begin ... --> / <!-- openapi:end --> markers and reuses it verbatim by default, so make docs-llms and the hook are offline and deterministic.
• --refresh-openapi (new make docs-llms-refresh) fetches the live spec and rebuilds the section. The daily refresh-llms workflow now calls that, so its drift detection and degraded-spec guards keep working unchanged.
• llms.txt is refreshed to the current spec in this commit (session-scoped file endpoints replace the old storage ones).
• Tests cover the offline reuse, the refresh path, the fallback when no cached section exists, and the stub line the workflow greps for.

@greptileai review

Follow-up on the review threads:

• The response body now crosses the page boundary as base64 of response.arrayBuffer() instead of response.text(), so content is the exact bytes the server sent and binary responses survive. encoding comes from the charset parameter of Content-Type, falls back to utf-8 when the bytes are valid utf-8, and is otherwise left to requests' detection. Unit tests cover a binary body and a latin-1 page; a browser test fetches a PNG and checks the magic bytes.
• The misc/response.mdx page is generated by sphinx-mintlify from requests.Response's own docstring; the generator has no way to exclude a referenced third-party class, so its truncated description and the json(kwargs) rendering are artifacts of that tool rather than something to edit by hand here.

@greptileai review