Perplexity 推出法律智能体系统 Computer for Counsel

Perplexity launched Computer for Counsel. It is an agentic AI system built for legal teams. The product extends Perplexity Computer, the company’s LLM-agnostic agentic system. It is available now to Perplexity Enterprise and Max subscribers. Lawyers lose hours to administrative work. Computer for Counsel targets that work directly. Nearly 75% of lawyers call administrative tasks a major time challenge, a Thomson Reuters survey found. The story is mostly architectural. It is an orchestration layer wired into the tools lawyers already use. TL;DR Perplexity launched Computer for Counsel on June 24, 2026, for Enterprise and Max subscribers. It routes 20+ frontier AI models per subtask, with no single-vendor lock-in. Premium sources include Midpage (case law + citator), Deel, and LegalZoom; 400+ tools connect via MCP. Every output links back to its source, so lawyers verify each citation before use. It is a workflow layer, not a Westlaw replacement; good-law checks still depend on Midpage. What is Computer for Counsel? It is not a new legal research database. Perplexity is explicitly not trying to replace Westlaw, LexisNexis, or Bloomberg Law. Instead, it sits as a research, drafting, and workflow layer. That layer reasons over the open web, firm systems, and specialized legal sources. The mechanics are agentic. The system decomposes a legal task into subtasks. It routes each subtask to a model and a data source. It then assembles the results into a brief, memo, or deal summary. Every output links back to its source. Attorneys verify a citation in seconds before it enters client work. Judgment and strategy stay with the lawyer. The Multi-Model Orchestration Layer Computer is powered by 20+ frontier AI models. It selects the best model for each subtask automatically. Research, reasoning, and contract work can each use a different model. Perplexity keeps the model pool current through ongoing evaluation. For legal teams, this removes the pressure to bet on one AI vendor. Connectors run on the Model Context Protocol (MCP). MCP is an open standard for linking AI systems to external tools and data. Administrators can also install custom MCP connectors for internal systems. The Data Layer: Sources and Connectors Premium legal sources ground the answers. The connector list spans research, contracts, and document management. Source / ConnectorTypeWhat it providesAccess at launchMidpageLegal researchUS case law (federal + state appellate), statutes, regulations, a citator to check if a case is still good lawUncapped for all Computer users; activate with @midpageDeelCompliance dataWorker classification, EOR rules, immigration, cross-border payroll across 150+ countriesFree, limitedLegalZoomContract templatesCustomer agreements, employment contracts, NDAs via a template flowLimited, coming soon, exclusive to PerplexityDocusignContracts / e-signatureAgreement history and automated contract workflowsAvailableNetDocuments / BoxDocument managementSecure file systems and a legal context graphAvailableDeepJudgeInstitutional intelligenceGrounds outputs in a firm’s prior work and accepted positionsAvailableClio (Vincent)Legal researchCited answers across 1B+ legal sources in 100+ jurisdictionsComing soonCarta / IroncladEquity / contractingCap tables, 409A data; AI contract repository searchCarta available; Ironclad coming soon App Connectors also reach Microsoft 365, Google Workspace, and 400+ other tools. Inside Microsoft 365, Computer drafts in Word and retrieves files from SharePoint. It references context from Outlook or Teams conversations. How Legal Teams are Using It Three current workflows show the agentic pattern in practice: Third-party NDA intake: Computer reviews third-party NDAs for red flags. It fills in entity and signatory information. It prepares clean copies. It routes them for approval and signature via Docusign. Regulatory monitoring: Computer builds a shareable dashboard for US state privacy and adtech laws. It shows which states have laws in effect. It cites Midpage for relevant cases. Case research with citation review: Computer researches non-compete enforceability after the FTC’s 2024 ban. It summarizes key cases and flags unsettled ones. It exports a PDF with citations. Interactive Explainer Illustrative simulation for engineers. Not connected to live legal data or real client matters.</p> </div> <div class="section-label">1 · Choose a legal workflow</div> <div class="tabs" id="tabs"> <div class="tab active" data-k="nda"> <div class="t">Third-party NDA intake</div> <div class="d">Review red flags, fill entities, route for signature.</div> </div> <div class="tab" data-k="reg"> <div class="t">Regulatory monitoring</div> <div class="d">Build a dashboard of US state privacy & adtech laws.</div> </div> <div class="tab" data-k="case"> <div class="t">Case research + citations</div> <div class="d">Research non-compete enforceability, export cited PDF.</div> </div> </div> <div class="controls"> <button class="act run" id="run"><img src="https://s.w.org/images/core/emoji/17.0.2/72x72/25b6.png" alt="▶" class="wp-smiley" style="height: 1em; max-height: 1em;" /> Run agentic workflow</button> <button class="act reset" id="reset">Reset</button> </div> <div class="progress"><div class="bar" id="bar"></div></div> <div class="section-label">2 · Pipeline execution</div> <div class="pipe" id="pipe"></div> <div class="out" id="out"> <h3 id="outTitle">Deliverable</h3> <div class="deliver" id="outBody"></div> </div> <div class="foot"> <span class="src">Facts sourced from Perplexity's "Introducing Computer for Counsel" (Jun 24, 2026).</span> <span class="brand">Marktechpost</span> </div> </div> <script> // ---- Illustrative workflow definitions (facts drawn from Perplexity's launch blog) ---- var FLOWS = { nda: { title: "Clean NDA package, routed for signature", steps: [ {n:"Plan task", model:null, meta:"planner", body:"Decompose into: extract parties → flag risky clauses → normalize template → route for e-signature."}, {n:"Route models", model:["Reasoning model","Extraction model"], meta:"router · 20+ models", body:"Router assigns a reasoning model to clause analysis and an extraction model to entity capture."}, {n:"Call MCP connectors", model:null, conn:["Docusign","NetDocuments"], meta:"MCP", body:"Pulls the counterparty draft from NetDocuments; prepares the Docusign envelope and signatory fields."}, {n:"Synthesize", model:["Reasoning model"], meta:"synthesis", body:"Flags an uncapped indemnity and a missing residuals clause. Fills entity and signatory information. Prepares a clean copy."}, {n:"Verify & route", model:null, meta:"human-in-loop", body:"Each flag links to the source clause. Lawyer approves; Computer routes for signature via Docusign."} ], deliver: 'Reviewed NDA with 2 red flags surfaced <a class="cite" title="Illustrative source link">[clause 4.2]</a> <a class="cite" title="Illustrative source link">[clause 7.1]</a>, entities auto-filled, and a clean copy queued in Docusign for approval. The attorney makes the final call before

原文超出正文长度上限，此处截断——上游还有内容，完整版见上方「原文 ↗」。