Perplexity CLI toolkit¶
NOT for pay-as-go API
Perplexity conversation archive & interactive query toolkit (pplx-export / pplx-ask).
It talks directly to the Perplexity REST/GraphQL API using your browser's logged-in cookies, and archives conversations — steps, citations, deep-research reports, computer-mode assets, and sub-agent workflows — as local Markdown + JSON. Successful exports retain the raw API responses alongside the rendered artifacts, so rendering can be re-run offline at any time.
1. Introduction¶
Beyond archiving historical conversations, this project primarily aims to give local agents — closer to the data and often backed by more compute — a measure of Perplexity Computer's capabilities. By letting them access reports produced by Perplexity Deep Research directly within their working loop, they can use high-quality information to tune key parameters in code more precisely while making fuller use of an existing Perplexity Max subscription.
As of July 20, Perplexity did not offer an official CLI for Unix-like environments. On July 23, Perplexity publicly released the
pplxtool used in Computer mode, but that tool still uses pay-as-you-go billing.
It is not a full replacement for Computer mode. Two abilities remain out of reach:
- the Deep Research skill's freedom to pick an arbitrary model;
- the Council skill's ability to run deep research with several user-specified models, produce reports, and compare them side by side.
The _platform_context/
directory includes archived system prompts and operating rules that may help
approximate parts of the Computer workflow locally, including Deep Research mode
selection and sub-agent model selection.
2. Features¶
2.1 pplx-export — archive your library¶
- Library index and space index
- Single-thread / batch export with incremental early-stop + resumable checkpoints
- Asset backfill and credit-usage backfill
- Conversation relation graph
- Offline re-render (
re-render, zero network) - Periodic-incremental cron snippet
2.2 pplx-ask — query Perplexity from the shell¶
- SSE streaming queries in four modes: search / deep-research / council / study
- Auto-move into a BOT space on completion, read receipts
- Automatic archiving of every thread it creates — built so other agents can call it to fetch real-time information
Per-mode artifact boundaries (citations / reports / assets / sub-agents): see Modes; rendering-fidelity principles: see Export pipeline.
Documentation provenance
Most pages in this MkDocs site were generated or reconstructed from the current code and tests. Some pages also retain design context, observations, and decisions from earlier discussions with agents. When a documentation statement and the implementation differ, treat the current code and tests as the source of truth.
3. Where to next¶
- Use the tools — follow the task-oriented User guide.
- Understand the implementation — use the Architecture reading map.
- Work with the observed web interface — consult the Web API reference.
- Change the project safely — follow the Maintainer guide.