April 2026 Summaries

Perplexity is expanding Computer for Enterprise with integrations designed to embed AI assistance into common workplace tools and recurring business processes. Computer can now be used directly in Microsoft Teams and is available as a beta side panel in Excel, allowing users to seek help within conversations and financial models. The company has also introduced more than 70 customizable workflows that package prompts, context, and output formats for repeatable tasks such as deal sourcing, sales preparation, and competitive analysis, with options to share, schedule, and run them asynchronously. New Snowflake and Databricks connectors enable teams to query internal data without moving it between systems, while a partnership with 1Password provides credential-free, governed, and auditable access to authorized actions. Computer for Professional Finance adds licensed data sources including Morningstar, PitchBook, Daloopa, and Carbon Arc, supporting sourced outputs such as tearsheets, annotated stock charts, and equity research comparisons. Perplexity also previewed Personal Computer, a continuously running, multi-model system for local files, applications, and the web that can operate on a Mac mini, with the new enterprise features rolling out now and broader Excel access planned later.

Perplexity describes its approach to improving the accuracy of frontier AI models through post-training techniques designed for its Search, Comet, and Computer products. Its process first trains models to follow instructions, remain consistent, and use tools appropriately, then develops their ability to handle complex search tasks requiring evidence gathering, verification, and synthesis across multiple sources. For open-ended activities such as editing, summarizing, and planning, the company uses structured rubrics to assess whether responses meet user needs and preserve intended meaning. Perplexity prioritizes factual correctness and task completion over stylistic preference, while also training models to avoid unnecessary searches and overly long answers. The company acknowledges that its systems are not fully accurate and emphasizes ongoing refinement through research and user feedback.

Perplexity describes a two-stage post-training pipeline for open-source Qwen3-based web-search agents that separates deployment requirements from search optimization: supervised fine-tuning establishes instruction following, safety guardrails, language consistency, and production-style tool use, while on-policy reinforcement learning improves factual accuracy and tool efficiency. RL training combines synthetic multi-hop, web-verified question answering with rubric-based general-chat tasks that preserve performance on non-verifiable requests such as rewriting and planning, using a 90/10 mixture to prevent easier rubric tasks from dominating optimization. Its composite reward gates preference-based quality scores on verified correctness or complete rubric compliance and applies group-relative penalties for excessive tool calls and response length, aiming to limit reward hacking without discouraging needed exploration. Reported evaluations show improved search accuracy, abstention, language consistency, and user-preference metrics, with the post-trained Qwen model outperforming or matching cited competitors on several factual benchmarks while requiring fewer searches and lower estimated query costs. The authors note remaining challenges involving training–inference mismatch, multi-objective optimization at scale, and long-horizon workflows involving multiple tools.

Perplexity has begun rolling out Personal Computer for Mac, an extension of its Computer product that combines multi-model AI orchestration with access to local files, native applications, connected services, and the web. Designed to handle complex, repetitive, and ongoing workflows, it can interpret objectives, break them into steps, and act across a user’s existing tools, such as organizing files, processing to-do lists, comparing local documents with web information, and communicating through connected apps. The company highlights the Mac mini as a persistent, always-available setup that can be managed remotely from a phone, while users can also activate the system directly in Notes or through voice commands. Perplexity says sensitive actions remain user-controlled, with visible operations, secure sandboxed file creation, and auditable, reversible actions. Personal Computer is initially available to Perplexity Max subscribers, with waitlisted users receiving priority as access expands.

Perplexity has expanded its personal finance features through Plaid, allowing signed-in users in the United States and Canada to securely link bank accounts, credit cards, loans, and previously supported brokerage accounts from more than 12,000 financial institutions. The service provides read-only connections, with Perplexity stating that user data does not reach its servers, and consolidates financial information into a single interface for spending analysis, debt monitoring, net-worth calculation, and investment tracking. Its Computer feature can create customized tools such as budgets, cash-flow forecasts, retirement dashboards, debt-payoff plans, charts, spreadsheets, and interactive apps based on users’ financial data and freeform questions. Perplexity also combines permissioned account information with real-time data and research from sources including FactSet, Coinbase, Nasdaq, S&P Global, and SEC filings, with citations linking to original sources. Basic portfolio functions are available to standard users, while advanced Computer capabilities require Pro or Max subscriptions, and the company plans to add crypto wallets, real estate, and other asset types in the future.

Perplexity has introduced U.S. federal tax modules for its Computer product, aiming to assist with tax tasks that require current knowledge, accuracy, and judgment, including drafting returns on official IRS forms, reviewing professionally prepared filings, and building customized planning tools. Based on the Agent Skills protocol and grounded in IRS materials and regulations, the modules are intended to keep pace with changing laws and forms that older AI models may miss. The company says Computer can analyze tax documents, ask users relevant financial questions, identify possible errors or missed deductions in existing returns, and create dashboards or software for matters such as depreciation, startup equity exercises, and rental-property deductions. Perplexity cites an example in which its system found an understated deduction under 2025 overtime-tax provisions in an attorney-prepared return, while emphasizing that its tax information is for reference only and that users should seek advice from qualified tax professionals for specific situations.

Perplexity has launched Computer in Slack, an AI agent integration that enables teams to initiate and manage tasks directly within their existing collaboration workflows. Users can tag Computer in channels or message it privately to conduct web research, access connected tools and business data, generate documents and presentations, write code, create apps or websites, and run multi-step workflows, while preserving requests, context, and results in shared Slack threads. Originally developed internally as a Slack tool called ASI, Computer reportedly completed $1.6 million worth of work for 300 Perplexity employees in its first four weeks, with public channel use helping colleagues discover and refine shared applications. The integration supports collaborative review, handoffs, and follow-up questions, can continue working in the background, and mirrors Slack tasks in Perplexity’s web platform. A new MCP connector allows it to search Slack workspace knowledge, read threads, and send messages, and Computer is now available through the Slack Marketplace.