FAQ

Why not just use WebFetch or curl?

Use them first: they are free and send nothing to a third party. fetchwright is for public pages that refuse them, with a 403, a block page, a login wall on a public page, or an empty JavaScript shell. Reddit threads refuse plain clients.

Why did my page come back as a summary?

Pages default to a digest because Gemini blocks long word-for-word copies as recitation (HTTP 400). --verbatim asks for the exact text and falls back to a digest when blocked. See How pages are read.

Why is my Reddit thread "not read"?

Google has to have the thread in its index. A brand-new thread may not be there yet. Check the retrieval line: if no URL shows success, Google did not get the page, and fetchwright discards whatever the model wrote.

Why did it drop a comment that was one word?

Gemini sometimes returns a commenter's username as the comment's text, and a single word cannot be told apart from a username. One word gives a note nothing to use. See How pages are read.

Can it read deleted comments?

No. In every test, deleted and removed comments came back as placeholders or empty, and fetchwright drops placeholders. If Gemini ever returns text from a deleted post, please open an issue.

gemini-3.8-flash keeps failing. Is my key broken?

Probably not. On 2026-10-15 it answered check but returned 503 or dropped the connection on URL context reads, while gemini-3.5-flash-lite read everything. fetchwright falls back by itself. A key problem shows as exit 4.

How many reads can I do a day?

Google does not publish fixed free-tier numbers on its rate-limits page; AI Studio shows your project's limits after you sign in. On 2026-10-15 a 429 said the free tier allows gemini-3.8-flash 20 requests a day; gemini-3.5-flash-lite's limit is unverified (about 20 reads in one session drew no 429). Since 0.2.0, fetchwright moves to the next model when one model's daily quota runs out. Limits are per project, and daily quotas reset at midnight Pacific time.

Is it legal?

Nobody has tested it. Terms and privacy says what Google's terms say and what is untested.

Does it work with Google Search grounding?

No, on purpose. Google's terms forbid passing grounded Search results to another model and using the search to find pages to crawl. The code refuses that tool.

Why is my JSON cut off?

Stdout is cut at 25,000 characters by default, and that includes --json. Use --out FILE or --max-chars 0. See Commands.

Does it need the agent skills?

No. The CLI runs on its own with Python 3.10 or newer and a key. The skills tell an agent when to use it and how to set the key up.