Perplexity Released pplxThe official client is a command-line interface for the Search API. This tool will return the extracted text from a page and search results as JSON. It targets both humans and coders equally, according to the documentation. It’s not a chat application. It does not have a conversational mode. There are no models to choose from and there is no synthesized response.
One output, two surfaces
This tool has two exact working surfaces. pplx search web Searches the web in real time. The pplx file content is fetch Pulls URL and cleans page text.
It is interesting to see the contract that surrounds them. As per the official pplx-cli Agent SkillSuccess means that there is exactly one JSON on stdout and exit code 0. Results of a search {hits: [{url, title, domain, snippet, ...}], total, saved_to?}.
Each failure is followed by an empty stdout. The JSON object is sent to the stderr. {"error":{"code","message","command","hint"?}}. Documented codes are AUTHENTICATION, UNKNOWN_ARGUMENT, ARGUMENT_ERROR You can also find out more about the following: BAD_REQUEST. This list of skills isn’t exhaustive. Therefore, callers can branch out. error.code.
Support for paths and platforms
The installation is done by a simple shell command which pipes the script into You can also find out more about the following::
curl -fsSL https://github.com/perplexityai/perplexity-cli/releases/latest/download/install.sh | sh
Reading install.sh What it really does. Downloads manifest.json The tag and version are extracted from the latest released. This pins each remaining download on that tag in order to avoid a competing publish.
The snare is a good example. SHA256SUMS The platform binary verifies checksum and installs ~/.local/bin/pplx. It is not necessary to sudo. You will receive a receipt. ~/.config/pplx/pplx-receipt.jsonOnly after the binary has been installed and run successfully.
Coverage is limited to just three targets: macOS x86_64 on Apple Silicon; Linux arm64; and Linux x86_64. All other builds will exit with an error. Windows builds and Intel macOS versions are not available.
Designing context-windows is a top-class concern
Most agents will use token budgeting. --output-dir The full set of results is written to a JSON-formatted file. --stdout-preview[= Adds a truncation to long strings in the stdout. ... markers.
It is a skill that is blatant about traps: --stdout-preview This is impossible without saving the result. The result will only be truncated if it is saved using --output-dir The following are some examples of how to use $PPLX_OUTPUT_DIR. When used alone, it produces full-size output. Hits can range from multiple KB to a few MB.
Search results saved at {dir}/web/{rand}.json At the moment, it is not possible to fetch at {dir}/fetch/{rand}.json. The files are only written after successful requests. PPLX_OUTPUT_DIR This option sets the workspace as default, so you don’t have to repeat it.
The skill for fetching content adds an accuracy check instead of a cost. Verify You can't help but be confused when you see this. You can also find out more about the following: is_paywall Before you trust, check the results Content. --html Adds to a raw_html Field fetched via crawler in real time. --no-cache Forces a live fetch.
What you need to know
pplxThis binary is verified and reveals two commands.pplx search webYou can also find out more about the following:The pplx file content is fetch.- Failures are indicated by a JSON error message on the stderr.
pplx auth loginTTY is the only option, and agents must also exportPERPLEXITY_API_KEY.--stdout-previewOnly truncates if paired with--output-dirThe following are some examples of how to use$PPLX_OUTPUT_DIR.- Each usage tier is capped at 50 requests per second.
Sources: perplexityai/perplexity-cli, pplx-cli SKILL.md, api-platform-developers, Perplexity API pricing, Rate limits and usage tiers” Search API quickstart

