Perplexity Releases pplx, a Single-Binary CLI That Puts Its Search API in the Terminal for Coding Agents

Perplexity just dropped a command line tool called pplx, and it's refreshingly narrow in what it tries to do. There's no chatbot personality here, no model picker, no synthesized answer stitched together from sources. It does exactly two things: pplx search web runs a live web search, and pplx content fetch grabs a URL and hands back cleaned page text. Everything comes out as JSON, which tells you immediately who this is really for.

The design leans hard into machine-readability. A successful run means exit code 0 and one JSON object on stdout, nothing more. A failure means exit code 1, empty stdout, and a JSON error object on stderr with a code like AUTHENTICATION, UNKNOWN_ARGUMENT, ARGUMENT_ERROR, or BAD_REQUEST. Perplexity's own documentation admits that list isn't exhaustive, so anyone building on top of this needs to branch on error.code rather than hardcode a handful of cases. That's a small detail, but it signals the tool was built by people who've actually had to debug flaky agent pipelines at 2 a.m.

Installation is a single curl-to-shell command, and the install script is more careful than most. It pulls a manifest, pins the exact release tag so it can't get caught mid-publish, verifies a SHA256 checksum against the binary, and drops the result into ~/.local/bin/pplx without needing sudo. A receipt file only gets written after the binary actually runs successfully — a nice touch for anyone auditing what landed on their machine. Platform support is deliberately narrow, though: Apple Silicon Macs, Linux on x86_64, and Linux on arm64. No Windows, no Intel Macs.

The part that really gives away the target audience is the token budgeting. Coding agents live and die by context window space, and Perplexity clearly knows it. The --stdout-preview flag truncates long fields, but only if paired with --output-dir or the PPLX_OUTPUT_DIR environment variable — used alone, it's a no-op that dumps full results, some of which run multiple kilobytes per hit. Saved files land in predictable web/ and fetch/ subfolders, written only after a request actually succeeds. For content fetch, there's also a trust check baked in: agents are told to verify the error and is_paywall fields before treating fetched text as reliable, with --html and --no-cache available for edge cases needing raw markup or a live pull instead of a cache.

Auth is TTY-only for the interactive login flow, so anything automated — CI, an agent runner — has to export PERPLEXITY_API_KEY directly. Pricing on the underlying Search API sits at $5.00 per 1,000 requests, with a rate cap of 50 QPS that applies across every usage tier, no exceptions for bigger spenders.