S

Solve Intelligence

Every Solve capability is model output: drafting applications, claims, figures, chemical structures and sequence listings, drafting office action responses, reviewing drafts, and building infringement, invalidity and freedom to operate charts. Agents take on whole multi step workflows, and retrieval decides which sources to query and how to iterate. Solve says its models are trained and tuned for patent law and run on providers including OpenAI, Anthropic, Google, Microsoft and AWS. Without the models, Patent Copilot is an empty browser editor.

Solve's responses and charts carry citations to multiple sources, and charts show their logic. Retrieval preserves citations, follows citation trails and ties each reference back to the legal or technical question, and case law and MPEP provisions are reachable inside the platform. No accuracy rate, error rate, test set or evaluator is published.

Solve works as a copilot inside a document editor: attorneys draft and respond interactively, with output matched to their own style and templates. Agents can take on entire prosecution and litigation workflows, and agentic retrieval chooses its own sources and filters, with each step building on the user's input. Solve's terms make the customer responsible for every decision and action based on output, and its trust center lists oversight of human decisions on AI recommendations as a control. Nothing states what an agent may complete without approval, where it stops, or what review comes before filing.

Solve names its customers widely. DLA Piper partner Larissa Bifano, Intel chief patent counsel Jeff Draeger, Banner Witcoff's Laura Brutman, Ashurst Perkins Coie's Thomas Yee and BCLP's George Chen are quoted by name. Siemens, Finnegan, Quinn Emanuel, Fish, Amgen and Foley Hoag sit on its customer advisory board. Solve reports more than 433,000 applications, 103,000 office action responses and 6,500 claim charts created, and an average workload saving above 50 percent. Those figures carry no method, baseline or date.

Each user's data stays sandboxed, and Solve says neither it nor any third party monitors it. It is not used to train Solve's or any foundation model, and model providers keep nothing under zero data retention agreements. Solve's terms bind those providers to confidentiality and to using inputs and outputs only to provide their service. The terms treat all customer data as confidential information. Customers choose the jurisdiction where data is stored and processed, and data can be exported and deleted at the end of a contract.

Solve gives no legal advice, its terms state, and customers should consult their own legal advisors. Customers and their users are responsible for every decision, piece of advice and action based on the service, and must use it in line with applicable rules and professional standards. The terms also bar customers from presenting any output as human generated. Competence and supervision duties, USPTO guidance and limits by jurisdiction are not addressed.

AI governance controls on Solve's trust center cover the source of model data, exclusion of customer data from training, continuous model performance monitoring, oversight of human decisions on AI recommendations, and fairness and bias testing. ISO 42001 certification is in progress, and Solve says its controls meet the standard's requirements. No test results, model card or named accountable owner are published, and nothing is disclosed about uneven output across technical fields or offices.

Customer data is not used to train or fine tune Solve's models or third party foundation models. Model providers process it under zero data retention agreements, with nothing stored for abuse monitoring or human review. Data is encrypted with TLS 1.3 in transit and AES 256 at rest, internal access is role based and least privilege, and access audits, vulnerability management and penetration tests run continuously. Thirteen subprocessors are named for each of the United States and EU deployments, and incident response protocols appear among Solve's listed controls.

Customer data can be exported, and is deleted at the end of a contract in line with the customer's agreement. No retention period is stated for data held in the platform during a subscription.

Under terms last updated 6 June 2026, Solve gives an indemnity against third party claims that the services infringe intellectual property. Outputs are carved out, since Solve gives no warranties, indemnities or liability for them, and the customer answers for every decision and action based on them. Solve warrants professional and workmanlike service only. Liability is capped at fees paid in the prior 12 months, with indemnity obligations outside the cap, consequential damages excluded, and a $1,000 cap during a free trial.

Solve's home page shows Microsoft Word, Microsoft SharePoint, NetDocuments, Google Drive and Anthropic as integrations, and refers to patent office connections. Its terms mention connectors to third party services, and enterprise customers can arrange custom integrations. What any integration moves, in which direction, or how it is set up is not documented, and no docketing or IP management system is named.

Customers choose a United States or EU deployment, and the data never leaves the chosen jurisdiction. Each deployment has its own 13 subprocessors, and EU customer data is processed entirely on EU infrastructure and never shared with the United States set. Processing follows the same choice, so model calls stay in the chosen region. The service runs on AWS, and payments for every customer go through Stripe to the United States parent company. No single tenant or on premises option is described.

Solve holds SOC 2 Type II and ISO 27001 certification, and ISO 42001 certification is in progress. Its trust center offers the SOC 2 Type II report, ISO 27001 certificate, a penetration testing report and its OpenAI zero data retention agreement through an access request. It also lists 14 controls on AI data governance, model testing and AI security, and 13 subprocessors per deployment. Auditor names and report periods sit inside the documents, which open after the request.

Solve's trust center names OpenAI, Anthropic, Google, Microsoft and AWS as language model providers in both its United States and EU deployments, with NLPatent providing patent and literature retrieval. Its terms commit Solve to use only model providers that agree not to train on inputs or outputs and to keep them confidential, and an OpenAI zero data retention agreement is available on request. No model or version is named for any feature, and no notice of provider changes is committed.

Solve publishes no price, tier or unit of charge. Its terms refer to subscriptions bought through a pricing page by card through Stripe, with renewal fee changes on notice and no refunds, but no such page is published. Enterprise pricing is set by arrangement.

Solve serves law firms, in house IP teams and companies, and reports more than 700 IP teams on six continents. Drafting handles figures from any file type, chemical structures and Markush claims, and biological sequence listings. Solution pages cover life sciences, telecommunications, standard essential patents, software and AI, medical devices, mechanical, electrical, consumer goods and design patents. Workflows run from invention disclosures and drafting through office actions, review, translation, charts, freedom to operate, invalidity, portfolio analysis and litigation.

Solve reports 20 percent of Am Law 100 and Chambers ranked IP firms and 30 percent of the Vault 30 IP firms as customers. No list of patent offices or date range is published for its patent collections.