# Perplexity Computer

> Source: https://aiwiki.ai/wiki/perplexity_computer
> Updated: 2026-10-18
> Fact-checked: 2026-10-18
> Categories: AI Agents, AI Tools & Products, Enterprise AI
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "Perplexity Computer." aiwiki.ai, 18 Oct 2026. https://aiwiki.ai/wiki/perplexity_computer
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

**Perplexity Computer** (usually just "Computer") is a cloud-hosted [AI agent](https://aiwiki.ai/wiki/ai_agents) product from [Perplexity](https://aiwiki.ai/wiki/perplexity). It takes a goal written in plain language, breaks it into tasks and subtasks, hands those to subagents running different [large language models](https://aiwiki.ai/wiki/large_language_model), and works through them in an isolated sandbox with its own filesystem, browser and connections to outside services. Perplexity launched it on 25 February 2026 for subscribers to its $200-a-month [Perplexity Max](https://aiwiki.ai/wiki/perplexity_max) plan. At launch the company described it as "a general-purpose digital worker that operates the same interfaces you do" and said it could run workflows "for hours or even months".[1][2] Pro and Enterprise subscribers got access on 12 March 2026.[8] Desktop versions followed: Personal Computer for Mac and Windows, and the local-first [Perplexity Portable Computer](https://aiwiki.ai/wiki/perplexity_portable_computer).[11][23][26] Usage is metered in credits. Plans include an allowance (monthly credits on Max and Enterprise, weekly included usage on paid Pro), and heavier use requires buying more credits.[2][36] Since July 2026, every Computer session has run on [SPACE](https://aiwiki.ai/wiki/perplexity_space), Perplexity's in-house sandbox platform, which the company open-sourced under the Apache 2.0 license on 15 October 2026.[21][33][34]

Perplexity's main pitch for the product is that it is multi-model. The company argues that frontier models are "specializing, not commoditizing", so a single harness that sends each subtask to whichever model suits it beats any one model family.[1][5] The models behind the product have changed many times since launch. Perplexity treats the routing as a moving target, and from September 2026 it hid most of it behind an "effort" slider.[28][37]

## Overview

| Attribute | Detail |
|---|---|
| Developer | Perplexity AI |
| Type | Cloud-hosted, multi-model autonomous agent |
| Launched | 25 February 2026 (Max subscribers, web)[1][2] |
| Wider access | Pro and Enterprise, 12 March 2026[8][9] |
| Surfaces (October 2026) | Web, iOS, Android, Mac and Windows desktop apps, Slack, Microsoft Teams, Microsoft 365, email, Comet Assistant[8][13][22][23][25][35] |
| Billing | Credits, 100 credits = $1 (as of October 2026)[36] |
| Sandbox | Firecracker microVMs, on SPACE since mid-2026[18][21][33] |
| Local variants | Personal Computer (Mac, Windows), Portable Computer (DGX Spark, RTX PCs)[11][23][26][27] |

## History

### Origins

In a 3 March 2026 essay, Perplexity CEO [Aravind Srinivas](https://aiwiki.ai/wiki/aravind_srinivas) said Computer began as an internal experiment the company called "ASI". The name "didn't really stand for anything"; staff sometimes expanded it as "Artificial Super Intelligence" or, because it first lived in [Slack](https://aiwiki.ai/wiki/slack), "Another Slack Integration". Employees messaged the bot a task, and it delegated to subagents, made files, and checked in only when it needed to. Srinivas wrote that once the system had a filesystem, a shell for running code and a browser, it was "essentially a computer", so the product was renamed Perplexity Computer.[5] The essay contrasted Computer with the open-source agent [OpenClaw](https://aiwiki.ai/wiki/openclaw) (formerly Clawdbot), which Srinivas called "interesting, if you like malware reading your texts". He said Computer was safer because it ran in a cloud sandbox.[5] Perplexity's enterprise material also says the first prototype was built inside Slack.[9][13]

### Timeline

| Date (2026) | Event |
|---|---|
| 25 Feb | Computer announced and made available on web to Max subscribers; Perplexity says it orchestrates 19 models. Max plans get 10,000 credits a month plus a one-time 20,000-credit launch bonus[1][2] |
| 3 Mar | Srinivas publishes "The AI is the Computer", explaining the ASI origin and the multi-model thesis[5] |
| 5 Mar | Custom Skills, Model Council (GPT-5.4, Claude Opus 4.6 and Gemini 3.1 Pro run in parallel, for Max) and Voice Mode added; GPT-5.3-Codex becomes a dedicated coding subagent[6] |
| 11 Mar | Perplexity announces Personal Computer for a dedicated Mac mini, Computer for Enterprise, a four-API developer platform, and premium data from Statista, CB Insights and PitchBook[7] |
| 12 Mar | Computer opens to all Pro subscribers on web and iOS, and to Enterprise Pro and Enterprise Max; the Slack integration and custom MCP connectors launch[8][9] |
| 26 Mar | Inline editing of generated assets, a live credit counter, scheduled-task management, and enterprise controls over sandbox internet access[10] |
| 16 Apr | Personal Computer for Mac begins rolling out to Max subscribers; Claude Opus 4.7 becomes the default orchestrator[11][12] |
| 30 Apr | Computer arrives in Microsoft Teams; Perplexity announces Workflows with a library of more than 70 (Enterprise users had six since March); Computer in Excel enters beta[10][13] |
| 3 May | GPT-5.5 becomes the default orchestrator for Pro and Max; GPT Image 2 becomes the default image model; websites can be published to pplx.app addresses[14] |
| 5 May | Computer for Professional Finance launches[15] |
| 7 May | Personal Computer opens to all Mac users with a Pro or Max subscription[16][17] |
| 18 Jun | Deep Research runs inside Computer; a command panel, forking, an analytics API and per-member credit limits are added[19] |
| 13 Jul | Brain memory system introduced; Claude Fable 5 becomes available as an orchestrator, and orchestrators can now be switched mid-task[20] |
| 15 Jul | Perplexity says 100% of Computer sessions now run on its SPACE sandbox platform[21] |
| 27 Jul | Brain reaches Max subscribers globally; Claude Opus 5 goes live; an API credential vault and enterprise role-based access controls are added[22] |
| 28 Jul | Personal Computer for Windows launches for Pro, Max and Enterprise[23] |
| 4 Aug | Spaces become Projects (shared workspaces with persistent files and Brain); Model Council expanded to two to eight models; Kimi K3 added[24] |
| 24 Aug | Computer in Email (computer@perplexity.com); GPT-5.6 Terra becomes the default subagent model and GPT-5.6 Luna the main model for scheduled automations[25] |
| 25 Aug | Portable Computer launches on NVIDIA DGX Spark (Linux)[26] |
| 14 Sep | Portable Computer comes to Windows PCs with NVIDIA RTX GPUs[27] |
| 17 Sep | Effort controls (Light, Standard, High, Ultra) roll out on web[28] |
| 21 Sep | GPT-6 Astra available for Computer tasks for eligible Pro and Max subscribers; Hybrid compute on Mac; Skills Marketplace; Side Chat[29] |
| 29 Sep | Automations replace Scheduled Tasks, adding event triggers and memory of earlier runs[31] |
| 5 Oct | GPT-6.1 Sol powers Light effort and Claude Opus 5.5 powers Standard effort for eligible personal accounts; video generation with MiniMax H3 and Seedance 2.5[32] |
| 15 Oct | SPACE open-sourced under Apache 2.0[33][34] |

## How it works

### Task decomposition and subagents

A user describes an outcome. Computer splits it into tasks and subtasks and creates subagents to carry them out. According to the launch post, the subagents "might do web research, document generation, data processing, or API calls to your connected services", and several run at once: "A document is drafted by one agent while another gathers the data it needs." When it hits a problem, Computer can spawn more subagents to solve it, look up API keys, research missing information or write code, and it "check[s] in if it truly needs you".[1] The work is asynchronous. A user can leave a task running, or run "dozens of Perplexity Computers in parallel".[1]

Perplexity's help center lists the main capabilities as of October 2026. Computer can run tasks in the background with condition-based triggers. It can run wide parallel web research and hand work to domain-specific subagents (market research, financial analysis, news monitoring, coordination). It keeps persistent memory across sessions and works with authenticated integrations such as Gmail, Outlook, GitHub, Linear, Slack, Notion, Snowflake, Databricks and Salesforce.[35] The help center sums up the difference from Perplexity's answer engine this way: "Perplexity answers your questions. Computer does your work."[35]

Perplexity added more ways to steer a running task over 2026. Plan previews (May) draft a structured plan and wait for approval before a long or credit-heavy task starts.[14] Users can stop a running subagent or type a new instruction mid-task.[12] Side Chat (September) answers questions about a task in a separate read-only thread without interrupting the main work.[29]

### Model orchestration

The launch post named the model line-up "as of this writing": [Claude Opus 4.6](https://aiwiki.ai/wiki/claude_opus_4_6) as the core reasoning engine, [Gemini](https://aiwiki.ai/wiki/gemini) for deep research, [Nano Banana](https://aiwiki.ai/wiki/nano_banana) for images, [Veo 3.1](https://aiwiki.ai/wiki/veo_3_1) for video, [Grok](https://aiwiki.ai/wiki/grok) "for speed in lightweight tasks", and "ChatGPT 5.2 for long-context recall and wide search". Users could also pick specific models for specific subtasks.[1] Perplexity's launch changelog said Computer orchestrated "19 models in parallel".[2] By the 12 March Pro launch the company was citing "20+" models.[8]

The default orchestrator and the wider model line-up have changed several times, according to Perplexity's changelogs:

| Date (2026) | Change Perplexity announced |
|---|---|
| 25 Feb | Claude Opus 4.6 is the core reasoning engine[1] |
| 5 Mar | GPT-5.3-Codex added as a dedicated coding subagent[6] |
| 16 Apr | [Claude Opus 4.7](https://aiwiki.ai/wiki/claude_opus_4_7) becomes the default orchestrator; users can choose Opus 4.7, GPT-5.4 or Claude Sonnet 4.6[12] |
| 3 May | GPT-5.5 rolls out as the default orchestrator for Pro and Max[14] |
| 13 Jul | [Claude Fable 5](https://aiwiki.ai/wiki/claude_fable_5) offered as an orchestrator; Opus 4.8 fast mode; orchestrator switchable between turns[20] |
| 27 Jul | [Claude Opus 5](https://aiwiki.ai/wiki/claude_opus_5) live across Search and Computer[22] |
| 24 Aug | GPT-5.6 Terra is the default subagent model and an orchestrator option; GPT-5.6 Luna is the primary model for scheduled automations; [Grok 4.6](https://aiwiki.ai/wiki/grok_4_6) and US-hosted [DeepSeek V4 Pro](https://aiwiki.ai/wiki/deepseek_v4_pro) added[25] |
| 21 Sep | [GPT-6 Astra](https://aiwiki.ai/wiki/gpt_6_astra) available for tasks and supporting agents for eligible Pro and Max subscribers[29] |
| 5 Oct | [GPT-6.1 Sol](https://aiwiki.ai/wiki/gpt_6_1_sol) powers Light effort; [Claude Opus 5.5](https://aiwiki.ai/wiki/claude_opus_5_5) powers Standard effort[32] |

Perplexity backs many of these choices with its own WANDR benchmark ("Wide ANd Deep Research"), a set of 500 data-collection tasks it released on 14 July 2026.[44] It publishes WANDR scores alongside its model announcements, for example the claim that Claude Opus 5 "outperformed every model tested except Fable 5, while costing 57% less per task".[22] These are vendor-run evaluations.

### Effort mode

On 17 September 2026 Perplexity added a four-step effort slider (Light, Standard, High, Ultra). Each setting picks the coordinating model and how deeply it reasons, and Perplexity chooses the pairings "based on evaluation results". The help center says an effort setting "does not cap the number of steps Computer takes, the number of sources it reads, or the tools it uses, and it is not a spending limit." Lower settings use cheaper models, so the same task generally costs fewer credits on Light than on Ultra. A Custom mode still lets users choose a specific model and reasoning level.[28][37] Effort mode is available on web, iOS and Android, but not in the macOS and Windows desktop apps, which keep the regular model picker (as of October 2026). Enterprise organizations that disable a model get a fallback model for that setting.[37]

### Skills, connectors and plugins

**Skills** (added 5 March 2026) are reusable instructions, formats or workflows that Computer applies whenever they are relevant.[6] A public Skills Marketplace followed on 21 September.[29] **Connectors** link Computer to outside services. From 12 March, Pro, Max and Enterprise subscribers could add any remote [Model Context Protocol](https://aiwiki.ai/wiki/model_context_protocol) server as a custom connector, on top of what Perplexity then described as "400+ prebuilt curated connectors".[8] By mid-October 2026 Perplexity had begun merging the Connectors and Skills pages into a single **Plugins** page. A plugin bundles connectors and skills "packaged around a job". Plugins were rolling out first to individual plans and self-serve Enterprise organizations.[40]

For enterprises, Perplexity added **Workflows**: guided, repeatable tasks for jobs such as sales prep, market research and website audits. On 30 April the company said its library held more than 70.[13][14] Computer for Professional Finance (5 May 2026) added 35 finance workflows across ten segments. It also added bring-your-own-license MCP connectors for Morningstar, PitchBook, Daloopa and Carbon Arc, and source traceability that links "every numeric value" back to its origin.[15]

### Memory: Brain and Projects

**Brain** (13 July 2026) is what Perplexity calls a "self-improving memory system". It runs in the background, reviewing sessions, connector data, files and user corrections, and writes what it learns into Computer's memory, with each entry linked to its source. Users browse it as a wiki-style graph of "Concepts, Entities, and Workstreams". Brain skips incognito sessions and has its own on/off switch.[39] On launch, Perplexity said Brain had lifted answer correctness by 25% and recall by 16% in testing while cutting cost by 13% on tasks with prior context.[20] As of October 2026 the help center lists Brain as a Research Preview for Max and Enterprise Max subscribers.[39] On 4 August Perplexity turned Spaces into **Projects**: shared workspaces with a persistent filesystem that people and agents edit together, plus connectors scoped per project.[24]

### Automations

On 29 September 2026 Perplexity replaced Scheduled Tasks with **Automations**. An Automation runs on a schedule or fires on an event in Slack, Gmail, Outlook, Linear or GitHub, and it remembers earlier runs, so a weekly report can compare this week's results with last week's. Users decide which actions need human review. Perplexity says watching for a trigger costs no credits; credits are spent only when an Automation runs.[31]

### Where Computer can be used

Computer started on the web.[2] It now runs in the iOS and Android apps, in Slack (DMs and channel mentions), in Microsoft Teams (from 30 April 2026), in Excel, Word, PowerPoint and Outlook through Microsoft 365, and through email. A user sends, forwards or CCs a message to computer@perplexity.com, and Computer replies "only to the verified sender" in the same thread.[8][13][23][25][41] From 27 July it could also be started from inside the [Comet](https://aiwiki.ai/wiki/perplexity_comet) Assistant.[22] Computer can publish websites and full-stack apps to pplx.app addresses or, through Vercel, to a custom domain.[14][20]

## Desktop and local versions

### Personal Computer

Perplexity announced **Personal Computer** on 11 March 2026 as a version of Computer running on "a dedicated Mac mini that can run 24/7", connected to local apps and Perplexity's servers. The announcement promised approval prompts for sensitive actions, a full audit trail for every session, and a kill switch.[7] It started rolling out to Max subscribers on 16 April and opened to every Mac user with a Pro or Max subscription on 7 May in a rebuilt Mac app.[11][16][17] A Windows version launched on 28 July for Pro, Max and Enterprise subscribers on Windows 10 and 11.[23] The help center calls Personal Computer "the superset". The cloud product is a feature inside it, and Personal Computer adds access to local files, native apps (control of installed apps is Mac-only as of October 2026), voice control and remote control from an iPhone.[38] Hybrid compute on Mac (21 September) splits each task: cloud models do planning, search and reasoning, while a local model handles private files and sensitive steps.[29][38]

### Portable Computer

[Portable Computer](https://aiwiki.ai/wiki/perplexity_portable_computer), launched on 25 August 2026 with [NVIDIA](https://aiwiki.ai/wiki/nvidia), runs the orchestrator, planner, tool router, scheduler, task queue and a local search index entirely on the user's machine. At launch it ran on the [NVIDIA DGX Spark](https://aiwiki.ai/wiki/nvidia_dgx_spark) under Linux, using Qwen 3.8 27B or PPLX 27B (Perplexity's post-trained version of the Qwen model). Work done on the device uses no credits, and the agent asks permission before sending anything to the cloud.[26] On 14 September Perplexity extended it to Windows PCs with NVIDIA GeForce RTX or RTX PRO GPUs that have at least 24 GB of VRAM.[27] As of October 2026 the help center also lists AMD Ryzen AI Max systems on Windows.[38]

## Security and the SPACE sandbox

Perplexity says every Computer task runs inside a Firecracker microVM. Each session gets its own Linux kernel, an isolated filesystem that resets when the session ends, and a private network namespace with its own firewall rules. Sandboxes pause when idle and are destroyed after a period of inactivity. Only the credentials a task needs are injected, and subagents use short-lived proxy tokens rather than raw API keys.[18] Against [prompt injection](https://aiwiki.ai/wiki/prompt_injection), Computer reuses the defenses Perplexity built for Comet: machine-learning classifiers scan retrieved content in parallel with the agent's reasoning and trigger "a safe stop" when they find something suspicious. Perplexity says these defenses were audited by Trail of Bits.[18] Enterprise admins can stream audit logs to SIEM tools, switch Computer or individual connectors off, restrict which models it may use, and set per-seat credit caps. Perplexity says enterprise data is not used for model training.[18]

The sandbox layer underneath is **SPACE** (Sandboxed Platform for Agentic Code Execution), a platform Perplexity built in-house. It started rolling SPACE out in June 2026. By 15 July, Perplexity said, 100% of Computer sessions ran on it, and in a side-by-side test on the same production traffic, median sandbox creation time fell from 185 to 60 milliseconds compared with the previous provider.[21] SPACE keeps credentials and network egress control outside the VM and uses copy-on-write snapshots for pause, resume, recovery and branching.[21][33] In a red-team report published 23 September, Perplexity's Secure Intelligence Institute said nine models given root inside a guest VM achieved no VM-to-host escape in 108 runs. Four models (Claude Opus 5.0, GPT-5.6 Cyber, GPT-5.6 Sol and Kimi K3) did get around network restrictions when allowed partial network access. Perplexity says it fixed those weaknesses and that neither bypass worked on retest. The same report found similar network-policy bypasses in eight of ten third-party sandbox platforms, which Perplexity disclosed to the vendors.[30] On 15 October 2026 Perplexity released SPACE, written mainly in Rust, on GitHub under Apache 2.0. It said Akamai had launched SPACE on its cloud platform, Crusoe would add it to its managed AI services, and CoreWeave had published deployment guidance.[33][34] See also [AI agent sandbox escapes](https://aiwiki.ai/wiki/ai_agent_sandbox_escapes).

## Availability and pricing

Computer is billed in credits on top of the subscription, although each paid plan includes an allowance. Perplexity Max costs $200 a month or $2,000 a year.[42] Ordinary Perplexity "Ask" searches do not use credits; Computer tasks do, and so does the Comet Assistant in Work mode.[35][36] The following figures come from Perplexity's help center as of 19 October 2026, and Perplexity says they "may change":[36]

| Plan | Monthly Computer credits | One-time new-subscriber bonus |
|---|---|---|
| Pro (consumer) | None; paid plans get a weekly "Included in Pro" allowance, rolling out gradually and limited to the Light effort setting when no credits are held | 4,000 credits, expiring in about 30 days |
| Max (consumer) | 10,000 | 35,000 credits, expiring in about 30 days |
| Enterprise Pro | 500 | 8,000 credits |
| Enterprise Max | 15,000 | 40,000 credits |

Perplexity prices credits at 100 to the dollar. It gives example ranges of 15 to 70 credits for a quick definition, 350 to 950 for a multi-step research report, 2,400 to 9,800 for a data dashboard, and 15,000 or more for "an entire app or multi-day build". It says these are "examples, not guarantees".[36] When credits run out, active tasks pause rather than cancel. Users can turn on auto-refill, which is off by default. Max accounts have a monthly spending cap that defaults to $200 and can be raised to $5,000.[36] Credits used by a task that finished, or that the user stopped, are not refunded even if the result disappoints. Refunds apply to failures on Perplexity's side, such as sandbox errors, and where consumer law requires them.[36] Launch terms were less generous: the original Max bonus was 20,000 credits.[2] In its 21 September 2026 changelog Perplexity also said eligible non-paying US accounts could try Computer on the web with "included turns".[29] As of 19 October 2026, however, the credits help page states that "Computer isn't available on free accounts".[36]

## Reception

TechCrunch's launch coverage noted that Perplexity had invited reporters to a briefing meant to include a demo, but "canceled the demo because of flaws found in the product hours before the event". TechCrunch said running entirely in the cloud "might spare it some of the security concerns of other agentic tools like OpenClaw". It also questioned the economics of Model Council, a related feature that queries several models at once: the "unit economics of offering multiple queries at flat subscription rates aren't entirely clear".[3] The German site heise online reported that Perplexity would, for the first time, bill by usage. It wrote that, as with OpenClaw, whose autonomy "quickly leads to high bills", flat fees "are therefore no longer sufficient", and it noted that the launch version had no local system access.[4] When Personal Computer reached all Mac users, TechCrunch described Perplexity Computer as "dubbed, confusingly" and framed the desktop product as Perplexity's "answer to OpenClaw and other local AI agents". It added that the safety claim was the company's own: "(Or at least that's the claim.)"[17]

Perplexity has published its own measures of the product's value. In March 2026 it said a study of more than 16,000 internal queries found Computer saved its teams $1.6 million in labor costs and did "the equivalent of 3.25 years of work in four weeks".[7][9] In July it claimed Computer had performed "more than $9.4B in labor-equivalent work" for users since launch.[23] Neither figure has been independently verified. When SPACE was open-sourced, NVIDIA's AI account welcomed the release as "giving developers more control over where agents run code and what they can access".[43]

## See also

- [Perplexity](https://aiwiki.ai/wiki/perplexity)
- [Perplexity Portable Computer](https://aiwiki.ai/wiki/perplexity_portable_computer)
- [Perplexity Comet](https://aiwiki.ai/wiki/perplexity_comet)
- [Perplexity Max](https://aiwiki.ai/wiki/perplexity_max)
- [Perplexity Finance](https://aiwiki.ai/wiki/perplexity_finance)
- [Computer use](https://aiwiki.ai/wiki/computer_use)
- [OpenAI Operator](https://aiwiki.ai/wiki/openai_operator)
- [ChatGPT Agent](https://aiwiki.ai/wiki/chatgpt_agent)
- [Manus](https://aiwiki.ai/wiki/manus)
- [Claude Cowork](https://aiwiki.ai/wiki/claude_cowork)
- [Harness (AI)](https://aiwiki.ai/wiki/harness)

## References

1. Perplexity. "Introducing Perplexity Computer." *Perplexity Blog*, 25 February 2026. https://www.perplexity.ai/hub/blog/introducing-perplexity-computer
2. Perplexity. "Perplexity Computer, Perplexity on Samsung Galaxy S26, and upgraded Comet Voice Mode." *Perplexity Changelog*, 26 February 2026. https://www.perplexity.ai/changelog/what-we-shipped---february-27-2026
3. Fernholz, Tim. "Perplexity's new Computer is another bet that users need many AI models." *TechCrunch*, 27 February 2026. https://techcrunch.com/2026/02/27/perplexitys-new-computer-is-another-bet-that-users-need-many-ai-models/
4. "Perplexity AI: Agentic AI in a Sandbox with 19 Models." *heise online*, 26 February 2026. https://heise.de/-11191317
5. Srinivas, Aravind. "The AI is the Computer." *Perplexity Blog*, 3 March 2026. https://www.perplexity.ai/hub/blog/the-ai-is-the-computer
6. Perplexity. "Custom Skills in Computer, Model Council in Computer, and Voice Mode in Computer." *Perplexity Changelog*, 5 March 2026. https://www.perplexity.ai/changelog/what-we-shipped---march-6-2026
7. Perplexity. "Everything is Computer." *Perplexity Blog*, 11 March 2026. https://www.perplexity.ai/hub/blog/everything-is-computer
8. Perplexity. "Computer for Pro subscribers, Computer for Slack, and Personal Computer." *Perplexity Changelog*, 12 March 2026. https://www.perplexity.ai/changelog/what-we-shipped---march-13-2026
9. Perplexity. "Computer for enterprise." *Perplexity Blog*, 12 March 2026. https://www.perplexity.ai/hub/blog/computer-for-enterprise
10. Perplexity. "Comet iOS Launch and Computer Updates." *Perplexity Changelog*, 26 March 2026. https://www.perplexity.ai/changelog/what-we-shipped--march-27-2026
11. Perplexity. "Personal Computer is here." *Perplexity Blog*, 16 April 2026. https://www.perplexity.ai/hub/blog/personal-computer-is-here
12. Perplexity. "Personal Computer on Mac Launch and Computer Updates." *Perplexity Changelog*, 16 April 2026. https://www.perplexity.ai/changelog/personal-computer-on-mac-launch-and-computer-updates---april-17-2026
13. Perplexity. "Computer at work." *Perplexity Blog*, 30 April 2026. https://www.perplexity.ai/hub/blog/computer-at-work
14. Perplexity. "Improved Computer Models and Enterprise Updates." *Perplexity Changelog*, 3 May 2026. https://www.perplexity.ai/changelog/improved-computer-models-and-enterprise-updates---may-4-2026
15. Perplexity. "Computer for professional finance." *Perplexity Blog*, 5 May 2026. https://www.perplexity.ai/hub/blog/computer-for-professional-finance
16. Perplexity. "Personal Computer is available to all Mac users." *Perplexity Blog*, 7 May 2026. https://www.perplexity.ai/hub/blog/personal-computer-is-available-to-all-mac-users
17. Perez, Sarah. "Perplexity's Personal Computer is now available to everyone on Mac." *TechCrunch*, 7 May 2026. https://techcrunch.com/2026/05/07/perplexitys-personal-computer-is-now-available-everyone-on-mac/
18. Perplexity. "How we built security into Computer." *Perplexity Blog*, 13 May 2026. https://www.perplexity.ai/hub/blog/how-we-built-security-into-computer
19. Perplexity. "Deep Research, command panel, and enterprise controls." *Perplexity Changelog*, 18 June 2026. https://www.perplexity.ai/changelog/deep-research-command-panel-forking-inline-actions-and-enterprise-controls
20. Perplexity. "Brain, faster Computer models, and website publishing." *Perplexity Changelog*, 13 July 2026. https://www.perplexity.ai/changelog/brain-faster-computer-models-website-publishing
21. Perplexity Engineering. "Making SPACE: Secure and efficient runtimes for long-running agents." *Perplexity Research Blog*, 15 July 2026. https://www.perplexity.ai/hub/blog/making-space-secure-and-efficient-runtimes-for-long-running-agents
22. Perplexity. "Role-based access controls, API credentials, and Brain for Max." *Perplexity Changelog*, 27 July 2026. https://www.perplexity.ai/changelog/role-based-access-controls-api-credentials-and-brain-for-max
23. Perplexity. "Personal Computer is now available on Windows." *Perplexity Blog*, 28 July 2026. https://www.perplexity.ai/hub/blog/personal-computer-on-windows
24. Perplexity. "Shared workspaces, Personal Computer for Windows, and Model Council." *Perplexity Changelog*, 4 August 2026. https://www.perplexity.ai/changelog/shared-workspaces-personal-computer-for-windows-and-model-council
25. Perplexity. "Computer in Email, GPT-5.6 Terra and Luna, and Grok 4.6." *Perplexity Changelog*, 24 August 2026. https://www.perplexity.ai/changelog/computer-in-email-gpt-5-6-terra-luna-and-grok-4-6
26. Perplexity. "Introducing Portable Computer for local-first AI." *Perplexity Blog*, 25 August 2026. https://www.perplexity.ai/hub/blog/introducing-portable-computer-for-local-first-ai
27. Perplexity. "Portable Computer for Windows is here." *Perplexity Blog*, 14 September 2026. https://www.perplexity.ai/hub/blog/portable-computer-for-windows-is-here
28. Perplexity. "Computer adds effort controls for model selection." *Perplexity Blog*, 17 September 2026. https://www.perplexity.ai/hub/blog/computer-adds-effort-mode-for-model-selection
29. Perplexity. "Effort Mode, GPT-6 Astra, and Skills Marketplace." *Perplexity Changelog*, 21 September 2026. https://www.perplexity.ai/changelog/effort-mode-gpt-6-astra-and-skills-marketplace
30. Perplexity Secure Intelligence Institute. "Escaping SPACE: Part I." *Perplexity Research Blog*, 23 September 2026. https://www.perplexity.ai/hub/blog/escaping-space-part-i
31. Perplexity. "Computer adds Automations for ongoing work." *Perplexity Blog*, 29 September 2026. https://www.perplexity.ai/hub/blog/computer-adds-automations-for-ongoing-work
32. Perplexity. "Automations, inline visualizations, and GPT-6.1 Sol." *Perplexity Changelog*, 5 October 2026. https://www.perplexity.ai/changelog/automations-inline-visualizations-and-gpt-6-1-sol
33. Perplexity. "Open-sourcing SPACE, the sandbox platform behind Computer." *Perplexity Blog*, 15 October 2026. https://www.perplexity.ai/hub/blog/open-sourcing-space
34. perplexityai. "SPACE" (repository). *GitHub*, accessed 19 October 2026. https://github.com/perplexityai/SPACE
35. Perplexity. "What is Computer?" *Perplexity Help Center*, last modified 14 October 2026. https://www.perplexity.ai/help-center/en/articles/13837784-what-is-computer
36. Perplexity. "How credits work on Perplexity." *Perplexity Help Center*, accessed 19 October 2026. https://www.perplexity.ai/help-center/en/articles/13838041-how-credits-work-on-perplexity
37. Perplexity. "Effort mode in Perplexity Computer." *Perplexity Help Center*, accessed 19 October 2026. https://www.perplexity.ai/help-center/en/articles/20260917-effort-mode-in-perplexity-computer
38. Perplexity. "What is Personal Computer?" *Perplexity Help Center*, accessed 19 October 2026. https://www.perplexity.ai/help-center/en/articles/14659663-what-is-personal-computer
39. Perplexity. "What is Brain?" *Perplexity Help Center*, accessed 19 October 2026. https://www.perplexity.ai/help-center/en/articles/19700001-what-is-brain
40. Perplexity. "What are plugins?" and "Where did Connectors and Skills go?" *Perplexity Help Center*, accessed 19 October 2026. https://www.perplexity.ai/help-center/en/articles/2026100701-what-are-plugins ; https://www.perplexity.ai/help-center/en/articles/2026100703-where-did-connectors-and-skills-go
41. Perplexity. "Computer in Email." *Perplexity Help Center*, accessed 19 October 2026. https://www.perplexity.ai/help-center/en/articles/20260715-computer-in-email
42. Perplexity. "Perplexity Max." *Perplexity Help Center*, accessed 19 October 2026. https://www.perplexity.ai/help-center/en/articles/11680686-perplexity-max
43. NVIDIA AI (@NVIDIAAI). Post on X, 17 October 2026. https://x.com/NVIDIAAI/status/2111488462264410553
44. Perplexity Research. "WANDR benchmark: Evaluating research agents that must search wide and deep." *Perplexity Research Blog*, 14 July 2026. https://www.perplexity.ai/hub/blog/wandr-benchmark-evaluating-research-agents-that-must-search-wide-and-deep