Solve Intelligence

AI Centrality

How much of the product is actually AI. Whether the machine learning is the mechanism the buyer is paying for or a feature layered onto conventional software, and whether the vendor is specific about which is which.

Founded in 2023 to build this and nothing else, with domain expertise assembled deliberately: the team is described as patent attorneys, AI PhDs and software developers, and the product is an AI editor rather than an editor with AI in it. Every capability is model output: application drafting, claim drafting and amendments, office action responses, continuations and divisionals, invention harvesting from disclosure, claim charting, freedom to operate analysis, infringement and validity work. Style matching is model behaviour rather than templating, configuring output to an individual attorney's drafting voice and adapting for field, client and region. Remove the models and the product is a blank browser editor. Second consecutive A on this axis in ip-and-patents, and the category is tracking the plaintiff pattern: these vendors were built model first because the work was never automatable any other way.

Citation Accuracy and Hallucination Disclosure

Whether the vendor publishes measured accuracy on citations and assertions, grounds output to primary sources, and says plainly what its system does when it does not know. Legal has a documented public record of fabricated citations reaching filed briefs, so an untested claim of accuracy is not evidence.

A PUBLISHED EVALUATION METHODOLOGY, which is rare enough on this index to name, and no published result. The vendor states that large language model response correctness is measured through continuous human evaluation against real patent drafting tasks, assessing factual grounding. Three elements of that matter: the evaluation is continuous rather than a one time benchmark, it is conducted by humans against real drafting work rather than synthetic prompts, and factual grounding is named as the criterion. Almost every record on this index either publishes a number with no methodology or publishes neither; this vendor publishes the methodology and not the number. Grounding is also structural: output carries citations to source documents and integrated case law, so an attorney can trace an assertion. Held at B because nothing measured is published. No accuracy rate, no error rate on claim drafting or office action responses, no sample size, no evaluator identity, and no result of any kind from the continuous evaluation it describes. A vendor running continuous human evaluation has the numbers and has chosen not to publish them, which is a different and more interesting gap than not measuring at all.

Autonomy and Oversight Model

What the system decides on its own, what a lawyer must approve, and whether the vendor documents where the review point sits. A tool that drafts under review and a tool that files without one are different products and different risks.

Oversight is built into the product model and into the company structure. The product is framed throughout as a Copilot operating inside a document editor where the attorney drafts, which places the human at the point of authorship rather than at the end of a pipeline, and style matching exists to make output conform to the attorney's own drafting rather than replace their judgement. Continuous human evaluation of model correctness is an oversight mechanism operating on the system itself. A Customer Advisory Board of senior patent professionals is stated to bring real prosecution and litigation experience into the product roadmap, which is governance of product direction by practitioners. Held at B because no boundary is published: nothing states what the system does unattended, whether any drafting or response can be filed without review, what confidence signal accompanies generated text, or what the continuous evaluation triggers when correctness falls. A copilot framing implies a pilot and never specifies what the pilot must do.

Operational and Outcome Evidence

Named, dated evidence that the product works in production at real firms or legal departments. Case studies with figures and identified customers count. Unattributed testimonials and launch announcements do not.

The strongest named customer roster in the pull, published by the vendor rather than relayed by a directory. Named IP practices and corporate teams: DLA Piper, Siemens, Finnegan, BCLP, Troutman Pepper Locke, Haynes Boone, HGF, Bookoff McAndrews, Khurana and Khurana, Altacit Global, MKS and HG Law, against a stated base of more than 700 IP teams across six continents. Attributed comment with name and title: Richard Hodkinson, Chief Technology Officer at HGF, describing a competitive evaluation before committing. A Customer Advisory Board of senior patent professionals is named as a structure rather than a testimonial. Funding is dated and specific: a $12m Series A in April 2025 with Microsoft participating, a $40m Series B in December 2025, backers including Y Combinator and Thomson Reuters, and the acquisition of Palito.ai in March 2026. An outsider can identify the firms, the executive and the dates without contacting the vendor. Outcome claims are the weak part and the note records it: 60 to 90 percent efficiency and quality improvements, more than 50 percent productivity, and one firm exceeding billing targets by $1 million after reallocating AI saved time, all without methodology, baseline or sample. The grade rests on the named and checkable elements.

Privilege and Confidentiality Posture

How client confidences are handled: attorney client privilege and work product treatment, segregation of one client matter from another, whether client data trains any model, and what the vendor commits to in writing rather than in marketing.

The most specific confidentiality architecture on the index, published as a structural claim rather than an assurance. Stated: data is sandboxed to individual users, the vendor has no access to or control over customer data, a zero trust and least privilege access model operates with regular vulnerability scans, penetration tests and access audits, and the vendor processes customer data solely on the customer's instructions without determining the purposes or means of processing, which is processor language drawn from data protection law rather than marketing. Encryption is named to the algorithm and protocol version, AES-256 and TLS 1.3. Customer selectable jurisdiction means confidential material can be kept where the client requires. The vendor publishes practitioner facing material specifically on how patent practitioners should evaluate AI tools for data security and confidentiality, which engages the professional question rather than only the technical one. Graded A because sandboxing to the individual user with a stated absence of vendor access is a stronger structural position than the tenant level segregation the B and A records elsewhere describe. Held short of perfection because the architecture is asserted rather than documented and no privilege specific attestation exists.

UPL and Professional Responsibility Posture

Whether the vendor is clear that it supplies a tool rather than legal advice, who its audience is, and how it addresses unauthorized practice of law, competence and supervision duties, and jurisdiction limits. ABA Formal Opinion 512 is the reference point.

Not located, and the absence is conspicuous against what this vendor does publish. The company is built by patent attorneys, maintains a Customer Advisory Board of senior patent professionals, sponsored and presented at the FICPI ABC Meeting 2026 on what is required for AI to be ready for patent practice, and publishes detailed guidance on evaluating AI tools for confidentiality. None of that engages professional responsibility: no statement on the practitioner's duty of competence over machine drafted claims, no positioning on who is responsible for a generated office action response filed under a registered practitioner's signature, no reference to USPTO Rules of Professional Conduct or 37 CFR, and no bar guidance. Compare Patlytics at B in this same category, which names Model Rule 1.6, 37 CFR, USPTO Rules and 35 U.S.C. section 102 explicitly. Checked the home page, the trust centre summary, the security and confidentiality guidance, the blog and the funding announcements on 29 Aug 2026.

AI Governance and Bias Disclosure

Published governance over model behaviour: who owns it inside the vendor, what is tested before release, and what is disclosed about disparate output across matter types, parties, or populations.

The most operationally specific AI governance disclosure in the pull, without the certification Patlytics holds. Published: policies aligned with ISO 42001 and ISO 27001 and reviewed at least twice yearly, a stated compliance roadmap toward ISO 42001, an undertaking to comply with the European Union Artificial Intelligence Act as both AI Provider and Deployer, and continuous human evaluation of model correctness against real drafting tasks assessing factual grounding. Naming its own role under the EU AI Act as both provider and deployer is a legally consequential self classification that determines which obligations attach, and no other record on this index makes it. Continuous evaluation is a live governance process rather than a policy document. Held at B rather than A because none of it is certified or evidenced: ISO 42001 is a roadmap rather than a certification, no evaluation result is published, no bias or fairness testing appears, no model card exists, and no named governance body was located. Compare Patlytics at B with the ISO 42001 certificate and no published output. Two routes to the same grade: one holds the certificate and publishes nothing from it, the other publishes the process and holds no certificate.

AI Safety and Data Stewardship

Retention, deletion, access control, and what happens to prompts and documents after they are processed. Whether the vendor states its subprocessors and its incident practice, or leaves the buyer to assume.

The most complete stewardship position on the index, on all three limbs at once. Training: no data uploaded to or output from the product is ever used to train any AI model of any kind, which is the broadest formulation encountered, covering output as well as input and admitting no carve out for third party, public, shared or internal models. Retention: customers control zero data retention settings for the large language model providers and models used across the platform, which is customer exercisable rather than a policy to trust and reaches the downstream model layer. Access: data is sandboxed to individual users, neither the vendor nor any third party monitors it, and the vendor states it has no access to or control over customer data. Supporting controls are specific to the algorithm and protocol version. Graded A because it answers what the vendor does with data, what the downstream providers do with it, and who can see it, with a customer control on the retention limb, and no other record on this index answers all three. Held short of perfection because it is policy and product configuration rather than contractual terms located in this pass, and no retention default is stated for customers who do not configure it.

AI Liability and Recourse

What the vendor stands behind contractually when its output is wrong. Indemnities, caps, carve outs, insurance, and whether any of it is published or only reachable through a negotiated agreement.

No published position located. Nothing was found on liability for AI output, warranty, service levels or remedy. The exposure in this category is severe and specific: a generated claim set that is too narrow loses scope permanently once granted, an office action response that mischaracterises prior art can create prosecution history estoppel, and a missed reference in a freedom to operate analysis can expose a client to infringement liability. This vendor publishes more about how it protects data than any peer and nothing about what happens when its output is wrong. Checked the trust centre summary, the security and confidentiality guidance, the home page, the blog and the site navigation on 29 Aug 2026. Enterprise agreements govern this and are not public.

Practice Systems Integration Depth

How deeply the product reaches into the systems legal work already lives in: document management such as iManage and NetDocuments, Word and Outlook, contract lifecycle management, matter management, e-billing, and court filing systems.

Nothing located, and the product architecture explains why without excusing it. Solve is an in browser document editor, and independent comparison material draws the contrast explicitly, describing a competitor as appearing in a Microsoft Word sidebar so attorneys need not leave their existing workflow, and positioning Solve as the browser based alternative. No Word or Office integration, no IP docketing system connector, no patent management platform integration, no document management system and no API were located. Docketing integration is the material gap for a prosecution product, since deadlines, annuities and family relationships live in the docketing system. One March 2026 event is recorded and not credited as an integration: the acquisition of Palito.ai, described as unifying patent litigation and prosecution in one platform, which is consolidation of the vendor's own products rather than connection to a customer's estate. Checked the home page, the product pages, the acquisition announcement and independent comparison material on 29 Aug 2026.

Deployment Model and Data Residency

Where the software runs and where the data sits. Multi tenant cloud, single tenant, private deployment, on premises, and whether region of residence is a published option or an enterprise conversation.

THE BEST RESIDENCY DISCLOSURE ON THE INDEX, and the only one that is a customer control rather than a vendor statement of fact. Published: customers choose where their data is stored and processed, with the United States and Europe named as examples, and the vendor states the data never leaves the chosen jurisdiction. That is selectable residency with an exclusivity commitment attached. It extends down the chain: based on customer selection, only United States or European Union based subprocessors are engaged for data processing, which closes the gap that undermines most residency claims, where storage is regional and processing or inference quietly is not. The vendor states the purpose plainly, being to help enterprises meet internal and regulatory requirements. For patent work this is not a preference: unpublished applications engage foreign filing licence requirements and export controls, and a practitioner may be unable to let an application leave a jurisdiction at all. Graded A because it is checkable, actionable and complete across storage, processing and subprocessors. Held short of perfection because no hosting provider is named, no full jurisdiction list is published beyond examples, and no single tenant option is described.

Security Certifications and Trust Center

Independent attestation a buyer can pull without a sales call: SOC 2, ISO 27001, penetration test summaries, a trust center with current reports and named scope rather than a badge image.

CHECK THE NOUN. A live trust centre operates at trust.solveintelligence.com and SOC 2 Type II certification is stated consistently, which alone would sit high on this axis. But the vendor's certification claims are not consistent between its own pages, and the difference is material. The home page states adherence to standards including SOC 2 Type II and ISO 27001 certification, as well as ISO 42001, GDPR and CCPA requirements, which a reader would take as three certifications. The vendor's own detailed security guidance says something weaker and more precise: SOC 2 Type II certified, with controls aligned with ISO 27001, the NIST Cybersecurity Framework and CSA CCM, and policies aligned with ISO 27001 and ISO 42001 standards, and it separately describes a compliance roadmap that builds on preparing for ISO 42001. Aligned with is not certified to, and preparing for is not holding. On the detailed page's own account this vendor holds one certification, not three. Only SOC 2 Type II is credited. Held at B rather than A on that basis and on the absence of certificate dates, examination periods or a named auditor. Contrast Patlytics in this same category, which states ISO 42001 certification without the qualifying language and is graded on that claim.

Model Supply Chain Disclosure

Which models sit underneath, whose they are, where they run, and whether the vendor commits to telling customers when that changes. A legal buyer inherits every dependency it cannot see.

The chain is disclosed structurally and no party in it is named, which is the inverse of the usual gap and is more useful than it sounds. Published: the platform uses large language model providers, plural, confirmed by the existence of customer controlled zero data retention settings for those providers and models; only United States or European Union based subprocessors are engaged for data processing, based on customer selection; and no uploaded data or output is used to train any model of any kind. A buyer therefore knows that third party models are involved, where they may be located, that the customer chooses, and that a zero retention posture can be enforced against them. That is more actionable for a procurement review than a provider name with no accompanying controls. Held at B rather than A because no provider, model family or version is named and no subprocessor list was located outside the trust centre, which was not entered in this pass. Compare Onspring at B for naming Anthropic and saying nothing about handling: two records at the same grade from opposite disclosures.

Commercial Transparency

Whether a buyer can learn what this costs without entering a sales process: published rates, the unit being charged, what sits behind an enterprise tier, and what implementation adds.

No pricing located in vendor material at any level. No price, no range, no tier structure and no unit of charge, and no indication of whether the platform prices per seat, per application drafted, per office action or per chart. An independent legal software directory reports a starting price of $199 per month, which is a specific figure suggesting a published pricing page exists that was not reached in this pass, and it is recorded as third party context rather than credited as disclosure. Flagged as a correction candidate in the upward direction on that basis. The gap is notable against the rest of this record, which is among the most disclosure forward in the pull on every other dimension. Checked the home page, the product pages, the pricing navigation, the trust centre summary and independent directory material on 29 Aug 2026.

Firm and Practice Coverage

Who the product is actually built for. AmLaw, midlaw, small firm and solo, in house departments, government and courts, and which practice areas are supported rather than merely claimed.

The most completely characterised coverage statement in the pull, across three dimensions rather than one. Technical domains are enumerated with the specific artifacts each requires: life sciences with biological sequences and chemical structures, telecoms and standard essential patents at volume, software and electronics from machine learning architectures to chip designs, and mechanical work with native CAD file support, auto generated figures and labelling. That is a claim a practitioner can test against their own docket, because a platform that cannot handle a sequence listing or a CAD figure is disqualified for their work regardless of drafting quality. Workflow coverage runs invention harvesting, drafting, continuations and divisionals, office action responses, claim charting, freedom to operate and clearance, infringement, validity, standard essential patent mapping and litigation. Geographic coverage is stated as more than 700 IP teams across six continents with US, Europe and Asia named and offices in New York and Munich, and drafting is configurable by region. Graded A because coverage is stated at the level a buyer checks rather than as a count. Held short of perfection because no patent office list, corpus date range or update lag is published for the search and charting side.