perplexed launched pplxthe official command line shopper for the Search API. The device returns all well-founded search outcomes and extracted web page textual content as JSON. In line with the documentation, it targets people and coding brokers alike. It isn’t a chat shopper. There aren’t any dialog modes, no mannequin choice, and no synthesized solutions.
2 surfaces, 1 output contract
This device exposes precisely two work planes. pplx search net Run a dwell net search. pplx content material fetch Will get the URL and returns the cleaned web page textual content.
The contract concerning them is the attention-grabbing half. In line with the official pplx-cli agent skillssuccess means an exit code of 0 and just one JSON object on stdout. Search outcomes {hits: [{url, title, domain, snippet, ...}], complete, saved_to?}.
All failures exit with empty commonplace output 1. A single JSON error object is shipped to stderr and formatted. {"error":{"code","message","command","trace"?}}. Documented code contains: AUTHENTICATION, UNKNOWN_ARGUMENT, ARGUMENT_ERROR and BAD_REQUEST. The talent factors out that the listing shouldn’t be exhaustive, so the caller ought to department primarily based on the next objects: error.code.
Set up path and platform assist
Set up is a single shell command to which you pipe a script. sh:
curl -fsSL https://github.com/perplexityai/perplexity-cli/releases/newest/obtain/set up.sh | shlearn install.sh Present what it really does. Obtain manifest.json Extract tags and variations from the newest launch. Then pin all remaining downloads to that tag to keep away from concurrent publishing conflicts.
it will get SHA256SUMS and platform binaries, confirm checksums, and set up within the following location: ~/.native/bin/pplx. sudo shouldn’t be required. The receipt is written as ~/.config/pplx/pplx-receipt.jsonNonetheless, it solely executes after the put in binaries have been efficiently executed.
Platform protection is restricted to a few targets: macOS on Apple Silicon, Linux x86_64, and Linux arm64. The rest will exit with an error. There aren’t any Home windows or Intel macOS builds.
Context window economic system is a first-class design concern
Probably the most agent-specific function is token budgeting. --output-dir Writes the whole outcome set to a JSON file. --stdout-preview[=<CHARS>] Truncates and appends lengthy string fields on commonplace output ...<truncated> marker.
This talent speaks frankly about traps. --stdout-preview Nothing will likely be performed if there is no such thing as a save listing. It’s going to solely be truncated if the outcome can also be saved. --output-dir or $PPLX_OUTPUT_DIR. When used alone, it returns full-sized output, and every hit might be a number of KB.
Saved search outcomes will seem in {dir}/net/{rand}.json and fetch {dir}/fetch/{rand}.json. The file will likely be written solely after a profitable request. PPLX_OUTPUT_DIR There is no such thing as a must repeat the flag because it units the workspace default.
For content material fetches, the talent provides a correctness verify fairly than a value. verify error and is_paywall Within the output earlier than trusting content material. --html add raw_html Fields are retrieved dwell through the crawler, --no-cache Drive dwell fetch.
Necessary factors
pplxis a single verified binary that exposes two instructions.pplx search netandpplx content material fetch.- Success is exit 0 and one JSON object on stdout. On failure, a single JSON error object is positioned on the usual error output.
pplx auth loginis a TTY solely, so brokers and CIs should be exportedPERPLEXITY_API_KEY.--stdout-previewis truncated solely when mixed with--output-diror$PPLX_OUTPUT_DIR.- Search API billing is $5.00 per 1,000 requests, capped at 50 QPS at every utilization stage.
supply: perplexityai/perplexity-cli, pplx-cli SKILL.md, API platform developer, Perplexity API pricing, Rate limits and usage levelsand Search API quickstart
Michal Sutter is a knowledge science knowledgeable with a grasp’s diploma in knowledge science from the College of Padova. With a robust basis in statistical evaluation, machine studying, and knowledge engineering, Michal excels at remodeling complicated datasets into actionable insights.
