Perplexity
Also known as: Perplexity AI, perplexity.ai, Perplexity Computer, Perplexity Enterprise, Perplexity Enterprise Pro, Perplexity Enterprise Max, Perplexity Agent API, Perplexity Search API, Perplexity Personal Computer, Perplexity Pro, Perplexity Max
Perplexity pairs its cited answer engine with Computer, an agent that turns a goal into a plan, hands the pieces to subagents in an isolated cloud sandbox, and acts across more than 400 apps, from Gmail and Slack to Salesforce and Snowflake. Computer runs on schedules or app events, keeps a per user memory called Brain, and lets people or admins pick models from Anthropic, OpenAI, Google and xAI. Enterprise plans add SSO, SCIM, custom roles and audit logs, and the Agent API opens the same runtime to developers.
Perplexity began as an answer engine that searches the web and cites every source, and that search is still free. The company now sells three things around it. Computer is a general purpose digital worker. Perplexity Enterprise brings search and Computer to organizations under admin control. An API platform serves developers building their own agents. Comet, Perplexity's agentic browser, has a separate profile.
A person describes an outcome and Computer breaks it into tasks, creating subagents for web research, document and app building, data processing and calls to connected services, and running dozens of sessions in parallel for hours or across days. Each task gets its own Firecracker microVM with a real filesystem, a real browser and code execution, and credentials reach it only for the task at hand, through short lived proxy tokens.
Work arrives from the omnibar, Slack, Microsoft Teams and email, and Automations start runs on a schedule or on events in Slack, Gmail, Outlook, Linear and GitHub, with the person deciding which actions run alone and which wait for review. Brain, a research preview on Max and Enterprise Max, builds a per user wiki of projects, people and preferences that its owner can browse, edit or delete. Effort mode picks a model and reasoning level for each job, or a user can choose a specific model from Anthropic, OpenAI, Google or xAI. Plugins bundle skills, connectors and files, and a Discover directory lists ready plugins from Perplexity. On a Mac, Personal Computer also operates native apps such as Mail, Finder and Notes.
The free plan covers search with citations. Pro costs $17 a month and Max $167 a month billed annually, and Max includes 10,000 Computer credits a month. Enterprise Pro is $40 a seat a month or $400 a year with 500 credits a seat, and Enterprise Max is $325 a seat a month or $3,250 a year with 15,000.
Enterprise adds SOC 2 Type II and HIPAA compliance, SAML single sign on, SCIM, custom roles on annual contracts, admin control of connectors and models, customer managed encryption keys and audit logs that stream every Computer action to a SIEM. Since October 2026, regulated institutions can run Computer on their own servers or in air gapped networks through Valarian, and Portable Computer runs locally on Windows and Linux. The Agent API bills per token and per tool call, and the Search API per thousand requests.
Perplexity fits knowledge workers, analysts and teams in finance, legal and consulting who want research, analysis and finished deliverables handed off end to end, and organizations that want search, an agent and a browser under one set of enterprise controls. Developers get the same search and agent runtime behind one API key. Teams that need formal test suites for agent changes will find quality checks limited to source verification, and heavy Computer use draws down credits at a rate that depends on how complex each task is.
Vendor details
Canonical URL
https://www.perplexity.ai
Category
Enterprise operations agent
Subcategory
General purpose digital worker that plans and runs multistep work across 400+ apps, with enterprise search and an agent API for developers
Funding status
Independent, venture backed company.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
More than 400 app connectors, among them Gmail, Outlook, Slack, Microsoft Teams, Notion, GitHub, Linear, Salesforce, HubSpot, Snowflake and MySQL, plus file connectors for Google Drive, OneDrive, SharePoint, Dropbox and Box. Custom remote MCP connectors and custom API credentials reach tools without a built in connector, and plugins bundle connectors with skills. Legal work adds Box, Docusign and NetDocuments with Midpage case law, premium data sources include PitchBook, Statista and S&P Capital IQ, and audit logs stream to Splunk and other SIEMs over webhooks.
In practice
A private equity associate asks Computer for a diligence tracker on a target company. Subagents split the web research, data pulls and drafting, and by the end of the day the associate has a tracker and a memo with cited sources.
A revenue operations lead sets an Automation for every Monday at 8 a.m. that pulls pipeline data from Salesforce, builds a dashboard and posts it to a Slack channel, with the post held for review until the lead approves the format.
A developer adds the Agent API to a support product, calling web_search and fetch_url with an OpenAI model and a fallback chain that moves to an Anthropic model when the first is unavailable.
Sources & related URLs
Related / legacy domains
Research sources
Agentic Index coverage score
13.5 / 14 capabilities · 96%
| Integrations & Tool Calling | Full |
|---|---|
|
Computer acts in outside systems through more than 400 app connectors, with authenticated integrations for Gmail, Outlook, GitHub, Linear, Slack, Notion, Salesforce, HubSpot, Snowflake and MySQL. Organizations add their own tools as custom remote MCP connectors over HTTPS, and for a service with no connector, Computer can use custom API credentials added for one task, saved to a personal vault or shared with a project, with the raw secret kept out of the sandbox. Admins choose which connectors the organization can use and what each may do. Sourceperplexity.ai/hub/blog/computer-for-enterpriseread 2026-10-24 |
|
| Workflow Orchestration | Full |
|
A request becomes a set of tasks and subtasks, and Computer creates subagents to carry them out, one researching the web, another generating documents, another processing data or calling a connected service, coordinating them asynchronously until the deliverable is done. When it gets stuck, it spins up further subagents to find what it needs or to write code. Dozens of Computer sessions can run in parallel, a task can carry on for hours or across days, and sessions can be forked, renamed, pinned and archived. Sourceperplexity.ai/hub/blog/introducing-perplexity-computerread 2026-10-24 |
|
| Knowledge Grounding & RAG | Full |
|
Files synced from Google Drive, OneDrive, SharePoint, Dropbox or Box, or uploaded to a Project, persist as copies that sessions draw on, with Perplexity selecting the relevant sections, and importing again refreshes them. Standard search can also query a connected service live at the moment of the question without copying anything. Personal synced files are searchable only by the person who synced them, shared Project files by anyone given access, and full contents stay behind the permissions of the source service. Sourceperplexity.ai/help-center/en/articles/10672063-introduction-to-file-connectors-for-enterprise-organizationsread 2026-10-24 |
|
| Human Oversight & Guardrails | Full |
|
Setting up an Automation means choosing which actions Computer may take on its own and which must wait for human review, and admins can limit what each connector is allowed to do. Reusing a saved API credential needs the user's approval. On a Mac, the privacy gate in Personal Computer can mask sensitive details, keep a step local, refuse it or ask for consent before protected file information reaches the cloud, and Portable Computer asks permission before anything leaves the device. Computer in Email has no review step of its own, and its replies go only to the verified sender. Sourceperplexity.ai/hub/blog/computer-adds-automations-for-ongoing-workread 2026-10-24 |
|
| Security, Identity & Governance | Full |
|
Enterprise plans carry SOC 2 Type II and HIPAA compliance, SAML single sign on, including Okta, and SCIM provisioning, and annual contracts add custom roles with granular permissions mapped to identity provider groups, each with its own monthly credit limit. Admins can switch Computer off or limit it to chosen members, restrict which models it may use, turn connectors on or off and enforce incognito mode. Enterprise data is never used for model training, and customers can bring their own encryption keys, so revoking a key locks the agent out of the protected data. Sourceperplexity.ai/enterpriseread 2026-10-24 |
|
| Observability & Auditability | Full |
|
Audit logs record every Computer task start, automated agent action, completion and error, along with queries, answers, file uploads and downloads, shared and deleted threads, connector changes and role changes. They stream in real time to an HTTPS webhook, formatted for Splunk's HTTP Event Collector or for Slack, and need 50 seats or at least one Enterprise Max seat. Computer Analytics charts daily credit use by model, top connectors, skills and projects and average task duration, with every panel exportable and an analytics API for enterprises. Sourceperplexity.ai/help-center/en/articles/11652747-audit-logsread 2026-10-24 |
|
| Memory & State Persistence | Full |
|
Brain is a self improving memory that background agents write from a user's own sessions, connected tools, files and corrections into a wiki of concepts, entities and workstreams, each entry linked to its source. It is scoped to one user and leaves incognito sessions out, and its owner can browse, edit or delete any entry on the Memory page or turn it off, while enterprise admins can disable it for members. Brain is a research preview for Max and Enterprise Max subscribers using Computer, and account memory keeps preferences until the user deletes them. Sourceperplexity.ai/help-center/en/articles/19700001-what-is-brainread 2026-10-24 |
|
| Deployment & Data Residency | Full |
|
Computer usually runs in Perplexity's cloud, each task in its own sandbox. From October 20, 2026, regulated institutions can run it on servers they own, in clouds they control or in air gapped networks with no link to outside model APIs or the public web, through a partnership with the UK sovereign infrastructure company Valarian, arranged with Perplexity's sovereign AI team. On a Mac, hybrid mode keeps private files and sensitive steps on a local model, and Portable Computer runs the model and agent on Windows and Linux devices, reaching the cloud only with permission. Sourceperplexity.ai/hub/blog/perplexity-brings-computer-to-enterprise-infrastructure-with-valarianread 2026-10-24 |
|
| Prebuilt Agents / Templates / Packs | Full |
|
Plugins bundle skills, connectors and files, up to 32 skills and 15 remote MCP servers in one package, and the Discover tab lists plugins, connectors and skills published by Perplexity by category, installed in one step and called with a slash. Teams can also write their own skills, import them as markdown or upload a plugin from a ZIP or a public GitHub repository, with admins deciding who may publish for the organization. Computer for Counsel adds a legal bundle with Midpage case law, Deel compliance data and connectors for Box, Docusign and NetDocuments. Sourceperplexity.ai/help-center/en/articles/2026100702-find-install-and-remove-pluginsread 2026-10-24 |
|
| Triggers & Channel Coverage | Full |
|
Automations start Computer on a schedule, such as every Monday at 8 a.m., or when something happens in Slack, Gmail, Outlook, Linear or GitHub, and a condition can narrow an event trigger to, for example, emails from one sender that ask for a decision. Watching for a trigger uses no credits. People also hand Computer work by forwarding or copying computer@perplexity.com on an email, by message in Slack or Microsoft Teams, and from the web, iOS and Android apps, and each Automation keeps a run history and can be paused or edited. Sourceperplexity.ai/hub/blog/computer-adds-automations-for-ongoing-workread 2026-10-24 |
|
| Model Flexibility & Routing | Full |
|
Effort mode offers Light, Standard, High and Ultra settings, with Perplexity picking the model and reasoning level for each and cheaper models on the lower settings, and anyone who wants control can choose a specific model and reasoning level in custom controls. Computer draws on about 20 models from Anthropic, OpenAI, Google and xAI, assigning them to subtasks such as deep research, images, video and quick lookups. Admins can restrict which models Computer may use, and analytics break credit spend down by model. Sourceperplexity.ai/hub/blog/computer-adds-effort-mode-for-model-selectionread 2026-10-24 |
|
| APIs / SDKs / MCP Extensibility | Full |
|
The Agent API exposes Perplexity's managed agent runtime at POST /v1/agent on api.perplexity.ai, with a /v1/responses alias for OpenAI SDK compatibility, bearer key authentication and official Python and TypeScript SDKs. A request names a model from OpenAI, Anthropic, Google or xAI, with fallback chains, or an effort preset, and can call web_search, fetch_url, finance_search and a sandbox tool, while custom functions reach the developer's own systems. Separate Search, Embeddings and Decisions APIs and an enterprise analytics API round out the platform. Sourcedocs.perplexity.ai/docs/agent-api/quickstart.mdread 2026-10-24 |
|
| Testing, Debugging & Optimization | Partial |
|
Check Sources reports how well a Search or Computer answer is supported by the sources it cites, and a Sources panel sets those citations beside the answer for side by side review. Automations keep a run history, and Computer Analytics tracks task duration, artifacts and credit use over time. There is no place to define test cases, score a Computer task or skill against expected results, or compare runs before an Automation or plugin goes out to the whole organization. Sourceperplexity.ai/changelog/role-based-access-controls-api-credentials-and-brain-for-maxread 2026-10-24 |
|
| Browser / Computer-use | Full |
|
Each Computer task runs with a real browser inside its sandbox, which the agent drives as part of its research and tasks, operating the same interfaces a person would. On a Mac, Personal Computer goes further and operates Mail, Finder, Slack, Messages, Notes and other native apps directly, with on device authorization protecting sensitive steps. Installed app control is not yet available on Windows, and admins of managed accounts decide whether local models may run on members' machines. Sourceperplexity.ai/help-center/en/articles/14659663-what-is-personal-computerread 2026-10-24 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Pricing
Free plan. Pro from $17 a month and Max from $167 a month, billed annually. Enterprise Pro is $40 a seat a month and Enterprise Max $325.
Flat subscriptions per person or per seat, with Computer work drawn from monthly credit allocations and an optional funded pool. Agent API usage bills per token and per tool call, and the Search API per request.
Included quota
Max includes 10,000 Computer credits a month, and paid plans carry a bonus grant of 4,000 credits on Pro and 35,000 on Max. Enterprise Pro includes 500 credits a seat a month and Enterprise Max 15,000. Standard searches never use credits, and seat allocations reset each billing cycle without rolling over.
What is public
Individual Pro and Max annual billed prices, Enterprise Pro and Enterprise Max seat prices monthly and annually, credit allocations per plan and per enterprise seat, and per call Agent API and Search API prices.
Billing mechanics
Individuals subscribe to Pro or Max. Organizations buy Enterprise Pro or Enterprise Max per active seat, monthly or annually, with credits per seat and an admin funded pool for overflow. The API platform bills usage separately through the API console.
Cost watchouts
Long or complex Computer tasks can drain a month of credits quickly. Audit logs need 50 seats or at least one Enterprise Max seat, custom roles need an annual contract, and Agent API usage is billed apart from any plan.
Variable cost rationale
Seats and subscriptions are flat, but Computer spends credits at a rate that rises with task complexity and length, so heavy agent use runs through the monthly allocation and into a funded pool or auto refill. API usage is metered per token and per tool call.
Additional watchouts
Credits do not roll over, and when they run out Computer pauses mid task until more are added.
Overage / add-ons
When credits run out, active Computer work pauses with progress saved and new credit tasks wait. Individuals can turn on auto refill below 500 credits with a spending limit. Enterprise admins fund a shared pool with auto refill up to a monthly cap and set per member limits by group.
Sales call required
No, self serve available
Free / trial
Free plan with search and citations, basic models and limited daily use, with no Computer credits
Lowest paid plan
Perplexity Pro at $17 a month billed annually
Commercial notes
Discounts are offered for teams of 250 or more, schools, nonprofits and government agencies. On premises and air gapped deployment through Valarian is arranged with Perplexity's sovereign AI team.
Key ambiguities
The price of extra Computer credits and the monthly billed prices for Pro and Max were not readable, and credit cost per task is described only as varying with complexity.
Missing data
Price per extra credit, monthly billed Pro and Max prices, and pricing for the Valarian deployment.
Related vendors
- 11th Estate — Agentic platform whose AI engine scans markets, matches an…
- Abacum — FP&A platform for growing companies whose AI Modeler builds, edits…
- Adopt AI — AI CPA firm whose agents run reconciliations, close, AP and AR, and…
- Aera Technology — Decision intelligence platform where an always on agent executes…
- Akro AI — On premise operational intelligence that automates document heavy…
- alfred_ — AI executive assistant that triages the inbox overnight and scores…