Commands

Two commands, check and read. Outputs below come from fetchwright 0.1.0 run against a stand-in for Google's API, with sample content (see Home).

Usage

python fetchwright.py --help

usage: fetchwright [-h] [--version] {check,read} ...

Read pages other fetchers cannot, through Gemini's URL context tool.

positional arguments:
  {check,read}
    check       is a key set, and which models answer (one tiny request per
                model)
    read        read one or more public pages

options:
  -h, --help    show this help message and exit
  --version     show program's version number and exit

python fetchwright.py --version

fetchwright 0.1.0

check

python fetchwright.py check --help

usage: fetchwright check [-h] [--models MODELS] [--no-call]
                         [--timeout TIMEOUT] [--json]

options:
  -h, --help         show this help message and exit
  --models MODELS    comma-separated, first is tried first (default
                     gemini-3.8-flash,gemini-3.5-flash-lite or
                     FETCHWRIGHT_MODELS)
  --no-call          only look for the key; send nothing
  --timeout TIMEOUT
  --json

• Without --no-call it sends one request per model ("Reply with the word OK.", no URL tool) and reports which answer. It stops early on a refused key or a 429.
• --timeout defaults to 90 seconds for check.
• Exit 0 when a key is found and, with calls, at least one model answers. Exit 4 with no key, or when every model refused the key. Exit 5 when every model was unavailable.

With no key set:

python fetchwright.py check --no-call

fetchwright 0.1.0
key: not set (GEMINI_API_KEY)

That exits 4. With a key, see Getting started.

read

python fetchwright.py read --help

usage: fetchwright read [-h] [--thread | --page] [--verbatim | --digest]
                        [--max-comments MAX_COMMENTS] [--models MODELS]
                        [--timeout TIMEOUT] [--gap GAP] [--json] [--out OUT]
                        [--max-chars MAX_CHARS] [--save-raw DIR]
                        URL [URL ...]

positional arguments:
  URL

options:
  -h, --help            show this help message and exit
  --thread              read as a discussion thread: title, post, comments
  --page                read as a page: title and main text
  --verbatim            ask for the text word for word (threads do by default;
                        long pages are often blocked as recitation, then
                        fetchwright retries as a digest)
  --digest              ask for a faithful account instead of quotes (pages do
                        by default)
  --max-comments MAX_COMMENTS
                        ask for at most N comments (cuts output tokens)
  --models MODELS       comma-separated model list; the next is tried on 5xx,
                        404 or a dropped connection
  --timeout TIMEOUT     seconds per request (default 120)
  --gap GAP             seconds between URLs (default 5)
  --json
  --out OUT             write the output to this file instead of stdout
  --max-chars MAX_CHARS
                        cut stdout here (0 for no limit)
  --save-raw DIR        keep each raw response here (page text: keep it out of
                        repositories)

option	what it does
--thread, --page	Force the shape. Without either, a reddit.com/.../comments/... URL or a Hacker News item URL reads as a thread, anything else as a page.
--verbatim, --digest	Word for word, or a faithful account. Threads default to word for word, pages to a digest. See How pages are read.
--max-comments N	Asks the model for at most N comments. Fewer comments means fewer output tokens, which are most of the cost of a thread.
--models a,b	The model list, tried in order. Default gemini-3.8-flash,gemini-3.5-flash-lite, or FETCHWRIGHT_MODELS.
--timeout S	Seconds per request, default 120.
--gap S	Seconds between URLs, default 5. One request per URL.
--json	One JSON record per URL (a list when there are several). The fields are on How pages are read.
--out FILE	Write everything to FILE (UTF-8, LF) and print one line saying so. Not cut.
--max-chars N	Cut what goes to stdout at N characters, default 25,000; 0 for no limit.
--save-raw DIR	Keep each raw API response as JSON in DIR. They hold page text: keep them out of repositories.

The 25,000-character cut applies to --json on stdout too, so a long thread printed as JSON can end mid-record and fail to parse. For JSON a program reads, use --out FILE or --max-chars 0:

python fetchwright.py read https://www.reddit.com/r/examplesub/comments/1abc234/ --json --max-chars 900

ends with:

[cut at 900 characters of 1357; use --max-chars 0 or --json --out FILE]

A page that was not read

python fetchwright.py read https://example.org/article

When Google could not retrieve the page:

# https://example.org/article
url: https://example.org/article
model: gemini-3.8-flash  mode: page  style: digest  (the model's account, not quotes)
retrieval: error https://example.org/article
tokens: page 300, output 410, total 900
NOT READ: Not able to browse the provided URL.

The style line names the style that was asked for, even when nothing was read.

Messages and streams

The page text, check's report and the wrote FILE line go to stdout. Errors that end the run go to stderr:

python fetchwright.py read https://www.reddit.com/r/examplesub/comments/1abc234/

with no key set:

No Gemini API key: set GEMINI_API_KEY (the fetchwright-setup skill walks through it).

On a 429 that names one model's own quota (Google's message says "per day" or "for model"), fetchwright moves to the next model and skips the spent one for the rest of the run (since 0.2.0). Any other 429, or a quota 429 when every model's quota is spent, stops the run at once, even with more URLs to go, and prints the records it already has:

rate limited (429): Resource has been exhausted. Stopped: wait before trying again, and check the quota in AI Studio.

A URL that is not http or https:

python fetchwright.py read file:///etc/hosts

not an http(s) URL: file:///etc/hosts

Exit codes

code	meaning
0	every URL was read
1	at least one page was not read (the record says why)
2	bad arguments, or a URL that is not http or https
3	rate limited (HTTP 429), or every model's daily quota spent: the run stopped
4	no key, the key was refused (400 "API key not valid", 401, 403), or payment needed (402)
5	every model was unavailable (5xx, 404 or a dropped connection)

Environment variables

variable	used for
GEMINI_API_KEY	the key (preferred)
GOOGLE_API_KEY	the key, when GEMINI_API_KEY is not set
FETCHWRIGHT_MODELS	the default model list, comma-separated

On Windows, when neither key variable is in the environment, fetchwright reads them from the user environment in the registry (HKCU\Environment). That finds a key set with setx after the shell started.

fetchwright writes no files of its own: only --out and --save-raw write, and only where you say.