Subscribe to Our Newsletter

Success! Now Check Your Email

To complete Subscribe, click the confirmation link in your inbox. If it doesn’t arrive within 3 minutes, check your spam folder.

Ok, Thanks

Perplexity switches on content filtering by default for API users and pledges zero data retention

The AI search company has published a developer FAQ clarifying how its API handles privacy, billing and model behaviour

Defused News Writer profile image
by Defused News Writer
Perplexity switches on content filtering by default for API users and pledges zero data retention
Photo by Possessed Photography / Unsplash

Perplexity, the AI-powered search company, has confirmed that its application programming interface (API) applies SafeSearch content filtering by default and does not retain or train on any query data submitted by developers.

The company published an FAQ setting out how its API differs from its consumer web interface, covering data handling, model behaviour, billing and support.

On privacy, Perplexity states: "We do not retain any query data sent through the API and do not train on any of your data," with zero-day retention applying as the default.

The API runs on Amazon Web Services infrastructure located in North America.

SafeSearch, a filtering system that restricts explicit or harmful content from appearing in results, is enabled by default for all API users, the company said.

The FAQ also addresses a technical quirk in the sonar-reasoning-pro model, which emits a block of reasoning tokens immediately before producing a valid JSON object, meaning the response_format parameter will not strip those tokens automatically.

Perplexity recommends that developers use a custom parser to isolate the valid JSON portion of the response.

On billing, the API operates a tiered credit system with upgrade thresholds at $50, $250, $500, $1,000 and $5,000 in cumulative purchases.

Developers can track spending per API key by navigating to Settings, then View Dashboard, then Invoice history.

A 401 authentication error, which typically signals a rejected or unrecognised credential, usually indicates an invalid key or exhausted credits and can be resolved by purchasing additional credits or enabling auto-top-up, the FAQ states.

Perplexity also notes that its API is broadly compatible with OpenAI's chat completions endpoint, though response structures differ between the two.

Bug reports should be filed in the Developer Community's dedicated Bug Reports category, while feature requests can be submitted via the project's GitHub page.

The recap

  • Perplexity enables SafeSearch filtering for its API by default.
  • Models run on Amazon Web Services in North America.
  • Tier upgrades require all-time credit purchases ranging $50–$5,000.
Defused News Writer profile image
by Defused News Writer