Enterprise Trust
Controls shown reflect the current product architecture during private beta. Independent security assurance reports have not been issued yet.
Built for regulated patent teams handling confidential invention data, audit-sensitive workflows, and multi-jurisdiction portfolios.
Design Your Invention Team
IP Management Platform
An attorney-in-the-loop drafting assistant that turns invention details and saved prior art into a structured first draft for attorney review: claims, abstract, description, and novelty analysis, prepared for the drafting conventions of the target jurisdiction. AI drafts are dramatically better when they know the art on record.
Without prior-art context, a model writes generic claims. With the relevant references in front of it, including pasted claim text, it drafts claims against the art on record, which the attorney then tests for novelty. Drafting lives inside the patent family, because that is how patent work flows.
Worldwide patent search via EPO Open Patent Services, covering 100+ jurisdictions through INPADOC.
Save, annotate, paste claim text, and track the relevance of references per patent family.
Feed invention details and saved prior art to an enterprise Claude model to generate jurisdiction-specific claims, abstract, description, and novelty analysis, asynchronously.
The drafting engine applies jurisdiction-specific claim format, statutory rules, and mandatory section structure. US, EP, IN, and PCT are the production set.
35 USC 101 / 112 and obviousness-type double patenting addressed, with the full US section structure (Field, Background, Summary, Brief Description of Drawings, Detailed Description).
EPC Article 84 clarity, conciseness, and support, plus the EPC 2000 "compound for use" claim format.
Section 3(d) enhanced-efficacy considerations flagged for attorney review on pharma claims, the single most important rule for Indian pharma patents.
Jurisdiction-neutral claims compatible with USPTO and EPO conventions, with national-phase amendments flagged for entry.
Search runs through EPO OPS, returning bibliographic data and family links worldwide. Full-text claims coverage is exactly as honest as it should be for a legal audience.
| Data | US | IN | EP | PCT |
|---|---|---|---|---|
| Bibliographic data (title, abstract, applicants, IPC/CPC, dates, family) | ||||
| Full-text claims | Paste | Paste | via EPO | via EPO |
Provenance you can defend. Every AI revision is generated against a frozen, immutable snapshot of the exact prior-art set used at that moment. What art the AI saw when it wrote a given claim is always answerable, even years later and even if the live reference was later deleted.
Every claim is its own record with a real dependency tree. Abstract, novelty analysis, and each description section are their own units. Every change to every unit is its own append-only revision, capturing who, when, why, AI versus manual, and what art the AI saw.
Refine claim 3 without burning tokens regenerating claims 1, 2, and 4 through N. Each claim and section is its own record.
Natural-language instructions ("broaden claim 2 to cover oral and IV administration") produce a new revision of one unit, not a full rewrite.
Attorney edits are revisioned identically to AI edits, distinguished only by a source field.
Restoring an earlier revision creates a new revision, never a deletion. History only grows.
Editing claim 2 marks dependent claims, novelty analysis, and description sections stale. You decide what to regenerate, the system never does it for you.
Side-by-side comparison between any two revisions of any unit, plus reorder, reparent, split, and merge of claims.
Approval lock and branching. An approved draft is hard-locked. To change it, you branch a new version that carries forward all content with full lineage. "Approved" stays a strong signal, and a draft cannot be approved while any unit is stale.
Mandatory reason-for-change. Every revision carries a reason, system-filled for AI revisions, attorney-supplied for manual edits and restores. The field is never null.
After an enterprise Claude model generates the draft, a background review runs a separate, independent reviewer model as a judge over the claims, abstract, description, and novelty analysis, banding how well each unit is supported by the prior art on record so you can spend review time where it matters. The bands guide attorney attention; they are not a legal determination.
High, medium, or low per unit, describing how well the generated text is supported by the prior art on record, not a probability and not a legal opinion. The reviewer defaults to medium when uncertain; high must be affirmatively justified.
Every rating cites a specific claim number, quoted phrase, or publication number from the prior-art snapshot. No vague hand-waving.
The draft is readable the instant generation completes; scores backfill seconds later while you read. No single long spinner.
Bands rank where attorney attention is most useful, and do not assess patentability, novelty, or filing readiness. They do not block export or approval, and are not shown to the end client.
AI drafting is governed by the same audit-first controls that protect the wider platform: tenant boundaries, retained evidence, controlled model use, and review history that stands up to enterprise procurement and regulated IP workflows.
Enterprise review
Controls designed for procurement, security questionnaires, and regulated IP teams.
Evidence retained
Every generated draft remains tied to the prior-art snapshot and review trail behind it.
Part 11-aligned patterns
Audit and change-control patterns follow 21 CFR Part 11 expectations. Validation stays with the customer.
Each customer workspace is separated at the application and data layers. Tenant identity is verified before any drafting record, prior-art source, or portfolio item can be accessed.
Draft revisions, source references, and prior-art snapshots are preserved as append-only evidence, creating a review history that can be inspected later without being rewritten.
Model access, prompt handling, and spending limits are governed per tenant. Generation and saving are controlled separately, with retention-limited provider settings and no personal data by default. Current provider retention terms are shared as part of security review.
Frontend gating is defense-in-depth only. The token-costing draft:generate permission is deliberately separate from draft:write, so admins control who can trigger paid AI calls.
| Action | Admin | Attorney | Paralegal | Read-only |
|---|---|---|---|---|
| Search patents (EPO) | ||||
| Save / paste prior art | ||||
| Fill invention details | ||||
| Generate / refine AI draft | ||||
| Manual edit / restore / tree edit | ||||
| Approve / reject draft | ||||
| View drafts + history + scores | ||||
| Export to DOCX |
Every draft starts in DRAFT status and requires review before filing. The novelty analysis flags potential overlaps with the prior art, and per-claim refinement lets attorneys tighten one claim without re-running the whole draft. The UI and DOCX export carry the disclaimer "AI-generated content. Not legal advice."
What the assistant does, and where the attorney stays in control.
No. This is a professional tool for attorneys: the attorney is the author and the AI is the assistant. Every generated draft starts in DRAFT status and requires attorney review before any filing. The UI and the DOCX export carry the disclaimer "AI-generated content. Not legal advice."
US, EP, IN, and PCT (WO) are the production set, each with jurisdiction-specific claim format, drafting rules, and mandatory section structure applied. A missing required section is flagged rather than silently omitted. JP and CN are planned and are not available yet.
Bibliographic data is retrieved for all jurisdictions via EPO OPS. EP and PCT patents also return full-text claims directly from EPO. US and Indian patents have no reliable public claims API today, so a claims-retrieval helper guides the attorney to paste claim text from the source (USPTO ODP or InPASS). Pasted claims are stored on the reference and fed to the AI for novelty analysis.
No. Tenant data is sent to the model provider only during generation, under an enterprise API agreement that does not permit training on customer content. Prompts contain only invention details and public prior art, and no personal data unless the attorney puts it there. The retention and processing terms in force, along with the current provider and processing regions, are shared as part of security review.
Every AI revision is generated against a frozen snapshot of the exact prior-art set used at that moment, and revision history is append-only at the database layer. So "what art did the AI see when it wrote claim 3?" is always answerable, even years later and even if the live reference was later deleted.
Generation runs on enterprise Claude models from Anthropic, configured for confidential business workflows, and a separate background review runs a second, independent model as a judge to band the claims, abstract, description, and novelty analysis as high, medium, or low support. Model selection is governed by task, availability, and customer requirements, so exact versions change over time and are confirmed during security review rather than pinned in marketing copy.
No. The draft:generate permission (which costs tokens) is deliberately separated from draft:write. Paralegals can search, save prior art, fill invention details, view drafts, and export, but only attorneys and tenant admins can generate, refine, or approve.
Yes. The append-only audit trail, electronic records, and access controls follow 21 CFR Part 11 patterns, and the India Section 3(d) enhanced-efficacy requirement is flagged for attorney review on pharma claims. The structured, versioned record is built to stand up in enterprise procurement and regulatory review. Compliance and validation still depend on your intended use, configuration, procedures, and quality system.
A general chatbot writes generic claims with no access to the prior art on record, no jurisdiction-specific statutory rules, no version history, and no audit trail. This assistant grounds every draft in the actual prior art saved to the family (including pasted claim text), applies US, EP, IN, and PCT rules and section structure, runs a second-model confidence review, and keeps an append-only audit trail with mandatory attorney review before filing. The attorney is the author; the AI is the assistant.
Yes. Generation produces a novelty analysis that flags potential overlaps with the prior art on record, and a separate, independent reviewer model reviews each claim, the abstract, the description, and the novelty analysis, banding each as high, medium, or low support with a cited rationale. You see the least-supported sections first. It assists the review; the attorney makes the call.
Yes. India (IN) is in the production jurisdiction set. Drafts apply Indian Patent Office rules, including the Section 3(d) enhanced-efficacy requirement that is decisive for pharma claims, plus the mandatory section structure, with US, EP, and PCT supported alongside it. A missing required section is flagged rather than silently omitted.
Book a walkthrough, or start upstream with free patent search and bring the references into your draft.