dalao.ai

Perplexity

An AI answer engine with cited sources, plus the Computer agent, the Comet browser and a developer API.Reference profile

About Perplexity

Perplexity is an AI answer engine built around cited sources, extended into the Computer agent, the Comet browser and a developer API platform. The consumer service is closed; parts of the infrastructure are open source, including the Apache-2.0 SPACE sandbox platform. As of 2026-10-26, Pro costs $17 a month billed annually, Max $167 a month billed annually and Enterprise Pro $34 per seat per month billed annually; the enterprise page cites 50,000+ organizations (as reported by the vendor). The API bills per token and per tool call; the Search API costs $5 per 1,000 requests.

EditorialUpdated 10 min read

What it is

At its core Perplexity is an "answer engine": a user asks a question in natural language, the system searches the web and a large language model writes an answer with inline citations, each pointing back to the source page. The homepage lists the answer engine, Computer, the Comet browser and the API platform among its products.

Beyond answers, Computer is a multi-step agent launched in February 2026 that researches, analyzes, writes documents, builds websites and code, and runs the whole process as an automated workflow. Comet is the company's own browser, with an assistant that asks, navigates and acts on web pages. The API platform offers the same search and agent capabilities to developers.

Computer gained several core features in the second half of 2026. Brain arrived on July 13, letting Computer remember past work, switch models mid-task, publish websites and research private companies. July 27 added role-based access controls for enterprises and a custom API credential vault; August 4 added Projects with shared files, Personal Computer for Windows and Model Council mode; September 21 added Effort Mode, the Skills Marketplace for reusable skills and hybrid cloud-and-local compute on Mac.

Perplexity does not rely on a single model vendor. The Pro page lists models from OpenAI, Google, Anthropic, Moonshot AI, Z.ai and xAI for users to pick from, and Max adds each vendor's flagship; the company says the system routes each task to a suitable model (as reported by the vendor).

The contracting entity is Perplexity AI, Inc. The terms of service set a minimum age of 13, and minors need a parent or guardian to accept the terms. The company blog says it was founded in August 2022 (as reported by the vendor).

Agent governance is an engineering problem, and open source is essential to solving it.

Perplexity blog, "Open-sourcing SPACE, the sandbox platform behind Computer"

Core capabilities

Cited answers. Every answer carries inline citations, and Pro and enterprise plans also cite premium data such as PitchBook and Statista. This presentation is the main difference from most general-purpose chatbots.

The Computer agent. Given a task in plain language, Computer plans, calls models and tools, runs code in isolated sandboxes and delivers documents, spreadsheets, websites or apps. Automations, launched on September 29, 2026, start work on a schedule or on events in Slack, Gmail, Outlook, Linear or GitHub, and remember earlier runs. The October 19 update added an Ultrafast mode for Max subscribers that is faster but uses more credits.

Local execution. Since October 20, Computer can run fully on Apple silicon Macs, with model, orchestration and execution on the device, using Qwen 3.6 or PPLX Pro, a post-trained version of it; with permission it can escalate to 15+ frontier models in the cloud. The same day Perplexity announced a partnership with Valarian to deploy Computer on institutions' own servers, including air-gapped networks.

The SPACE sandbox. Since July 2026 Computer has executed code in the in-house SPACE platform: each sandbox is a small virtual machine with its own operating system, while credentials and network access are managed outside it. The company says it creates millions of sandboxes a day (as reported by the vendor), and released SPACE under Apache-2.0 on October 15.

Developer platform. The API covers the Agent API, Search API, Embeddings API and Decisions API. The Agent API calls models from several vendors and includes built-in tools for web search, URL fetching, finance data, people search and a code sandbox.

Who it's for / Not for

A fit for: researchers, consultants, finance and legal professionals who need fast answers with sources. Each claim links to its source, and paid plans add PitchBook and Statista data.

A fit for: individuals who use several model vendors and would rather not pay for each. One Pro subscription switches between models from OpenAI, Anthropic, Google, xAI and others.

A fit for: teams that run research, summaries and drafting as recurring workflows. Automations in Computer run on schedules or events, and Enterprise plans add SSO, SCIM, audit logs and data retention settings.

A fit for: developers who need web search in their applications. The Search API bills per request, and the Agent API packages search, fetching and a code sandbox as tools, with token and per-call charges listed separately.

Not a fit for: users under 13, or consumer-plan users who object to their data being used for model training. The consumer privacy notice lists improving or creating services, including AI models, among its purposes; the enterprise and API terms state that customer data is not used for training.

Not a fit for: offline writing or pure code completion, tasks in which web search and citations add little. The terms of service restrict use in some regions and for some purposes; see the terms for details.

How it compares

Against Google's AI plans: as of 2026-10-26, Google AI Pro costs $19.99 a month and bundles cloud storage and Gemini inside Google apps; Perplexity Pro costs $17 a month billed annually and centres on switching between vendors' models, cited answers and Computer credits. Both answer questions with web search; the differences lie in model choice and ecosystem integration.

Against other search APIs: the Perplexity Search API costs $5 per 1,000 successful requests, or $1 with Fast Search (as of 2026-10-26). Tavily's docs list one credit per basic search at $0.008 per credit on usage-based billing, plus 1,000 free credits a month (as of 2026-10-26). At list prices Tavily basic search works out to about $8 per 1,000; Perplexity is cheaper per request, while Tavily offers a monthly free allowance.

Against cloud-only agents: Computer can run fully on a Mac or on an institution's own servers, a direction Perplexity pushed in the second half of 2026, and a fit for users who need data to stay local.

Getting started

Consumer sign-up is free, and the free plan answers questions right away; Pro and Max work across the web, desktop and mobile apps. Local Computer requires the Perplexity Mac app on an Apple silicon Mac, with 36 GB of memory or more listed as the target configuration.

Developers create a project in the API console, add credits and generate a key. Rate limits follow cumulative purchases across six tiers: new accounts start at Tier 0, and cumulative purchases of $50, $250, $500, $1,000 and $5,000 unlock higher tiers, which are never downgraded. Agent API requests per second rise from 1 at Tier 0 to 33 at Tiers 4 and 5.

Support for the old Sonar Chat Completions API ended on September 27, 2026. Synchronous and streaming requests keep working as they are gradually reformulated into Agent API requests; asynchronous Sonar requests are no longer supported, and the migration guide points them to Agent API background mode. The Agent API follows the Open Responses standard.

Official tooling includes the Python SDK perplexity-py, a TypeScript SDK and a Model Context Protocol server that connects Perplexity search to tools such as Claude Code and Cursor.

Pricing & license

Individual plans (as of 2026-10-26): Free costs nothing. Pro is $17 a month billed annually, with 4,000 bonus credits, deep research, model selection and higher usage limits. Max is $167 a month billed annually, with 35,000 bonus credits plus 10,000 credits a month, frontier models and the highest usage limits. The pages list only annual-billing equivalents.

Enterprise plans: Enterprise Pro is $34 per seat per month billed annually, with 500 Computer credits per seat each month. Enterprise Max is $271 per seat per month billed annually, with 15,000 Computer credits per seat each month plus SCIM, audit logs and data retention settings. A custom-priced tier adds custom roles, a dedicated domain, volume discounts and forward-deployed support.

API pricing: the Agent API bills each model's per-token rate plus tool fees: web search $2.50 per 1,000 calls ($1 with Fast Search), URL fetching $0.0005 per call, people and finance search $5 per 1,000 calls, and $0.03 per sandbox session. The Search API is $5 per 1,000 requests; embeddings run $0.004 to $0.05 per million tokens; the Decisions API bills input tokens only, at $0.01 to $0.02 per million, with free output. Rate-limited requests are not billed.

Search API billing details: one request may carry up to 5 queries and is billed once; failed requests are free. Embeddings come in pplx-embed-v1 0.6B and 4B sizes plus contextual versions, at $0.004, $0.03, $0.008 and $0.05 per million tokens. The input price of the Decisions API model pplx-decider-v1.2 fell from $0.04 to $0.02 per million tokens in October 2026; the flash version costs $0.01. The Agent API also offers a flex tier at 0.5x and a priority tier at 2x. API credits can also be bought through AWS Marketplace.

Enterprise usage differences: per the pricing page, Enterprise Max allows 100x file uploads against 30x for Enterprise Pro (relative to Free), and 20x against 2x for asset generation; one organization can mix Enterprise Pro and Enterprise Max seats.

Subscription rules: the terms state that subscriptions renew automatically and are non-refundable, with no credit for partial periods, unless the terms or a country-specific billing page say otherwise.

Licensing: the consumer service and API platform are closed; SPACE and the Python SDK are Apache-2.0, and the MCP server is MIT.

Risks & limits

Consumer data use. The consumer privacy notice, updated October 23, 2026, lists improving or creating AI models among its purposes. The enterprise pricing page states that Enterprise Pro and Max data is not used for training, and the API terms say the same for customer content. The two kinds of account follow different data rules.

Compliance wording differs. The enterprise and enterprise pricing pages say "HIPAA compliant", while the security page says "HIPAA-aligned safeguards"; all three list SOC 2 Type II (as reported by the vendor). The API terms of service add that customers may not process protected health information, as defined under HIPAA, through the services unless they and Perplexity have executed a Business Associate Agreement (BAA).

API migration. Sonar Chat Completions support has ended, and synchronous requests are gradually reformulated as Agent API requests, so response behaviour may change; asynchronous Sonar requests no longer work.

Disputes and subscriptions. The terms include binding individual arbitration and a class-action waiver, with a written opt-out within a set period; subscription fees are generally non-refundable. The terms of service restrict use in some regions and for some purposes; see the terms for details.

Advertising. The cookie notice on the site says Perplexity does not sell third-party ads on its services and uses tracking technologies to measure its own ads on partner sites.

Shopping. The terms include a separate ProShop section: products Pro subscribers order through Perplexity are sold directly by third-party merchants, Perplexity is not a party to the transaction, and delivery, refunds and returns follow each merchant's own policy.

Data export. The privacy notice says users can export or delete their data in account settings; the enterprise pricing page lists data retention settings for Enterprise Max.

How it got here

Perplexity was founded in August 2022, raised a $3.1 million seed round that September and released its answer engine, Ask, in December. A blog post on March 28, 2023 announced a $25.6 million Series A and an iPhone app, and said monthly active users reached 2 million within four months (as reported by the vendor).

On December 18, 2024 the company blog announced the acquisition of Carbon, a retrieval engine that connects external data to language models, to bring apps such as Notion and Google Docs into Perplexity. The perplexityai GitHub organization was created on October 18, 2024.

Computer dominated 2026: launched in February, Personal Computer for Mac in May, the switch to the in-house SPACE sandbox in July, Portable Computer for user-owned hardware in August, then Hybrid Compute on Mac, Portable Computer on AMD and Windows machines, and Automations in September. SPACE was open-sourced on October 15, and fully local Mac execution and the Valarian partnership followed on October 20.

On partnerships, HP announced on September 15, 2026 that it would preload the Perplexity Windows app on Windows PCs, and on October 1 Perplexity and American Express released 10 business skills for Computer covering cash flow forecasting, marketing and more.

On the API side, Sonar Chat Completions support ended on September 27, 2026; October brought the Decisions API and renamed the Agent API preset parameter to effort.

In open source, SPACE has about 1,951 GitHub stars and about 17 contributors with GitHub accounts (2026-10-26), with its latest push at October 26, 2026 01:04 (UTC+8); the MCP server has about 2,589 stars and about 18 contributors with GitHub accounts (2026-10-26).

Our verdict

Inference: a fit for individuals and teams that use web search as a daily starting point and need a source behind every claim; Pro is priced in line with mainstream AI subscriptions and switches between several vendors' models.

Inference: a fit for institutions that need control over data location; local and private deployment of Computer is uncommon among similar products, and the enterprise plans exclude customer data from training.

Inference: a fit for developers who need search, since the Search API's $5 per 1,000 list price is below some comparable search APIs and the Agent API bills search and sandboxes together.

Inference: Max costs about ten times Pro on annual billing in exchange for 10,000 monthly credits, flagship models and Ultrafast mode, a fit for individuals who run Computer tasks heavily every day, and not a fit for occasional users, whose usage falls within Free and Pro limits.

Inference: not a fit for consumer-plan users who object to training on their data, or for projects that depend on the old asynchronous Sonar API and cannot migrate soon.

Why it matters

Perplexity combines search with language models and cites its sources in its answers.

Sources

Official

  1. Perplexity | AI for the Curiousperplexity.ai · Accessed Oct 26, 2026
  2. Perplexity Properplexity.ai · Accessed Oct 26, 2026
  3. Perplexity Maxperplexity.ai · Accessed Oct 26, 2026
  4. Perplexity Enterpriseperplexity.ai · Accessed Oct 26, 2026
  5. Perplexity Pricing: Plans for Individuals and Enterpriseperplexity.ai · Accessed Oct 26, 2026
  6. Perplexity Securityperplexity.ai · Accessed Oct 26, 2026
  7. Perplexity Terms of Serviceperplexity.ai · Accessed Oct 26, 2026
  8. Perplexity Privacy Noticeperplexity.ai · Accessed Oct 26, 2026

Updates & reading

No articles or updates yet. Claim it and connect a feed.