The best legal AI for in-house counsel.

You are the entire legal function at a startup. There is no legal ops, no paralegal, and no budget line that survives a $500-per-seat quote. Sales wants the MSA back today, the CTO wants to know whether a new subprocessor is a problem, and a board meeting needs minutes.

There is a GC, a commercial counsel, maybe an employment or privacy specialist, and someone doing legal ops part-time. The bottleneck is no longer knowing the answer. It is contract throughput and the fact that everyone answers the same five questions in five slightly different ways.

You have Harvey, Legora, CoCounsel, or Lexis+ AI. It is good, and the three or four people who have seats use it. The problem is the twelve people who do not: the commercial counsel in another region, the two legal ops analysts, the privacy lead, the paralegals, and the business partners who keep sending work to legal because they have no other option.

A citation you can open

Where an answer rests on a document you supplied, it carries a verbatim quote along with the document and page it came from. Before the citation reaches you, the system searches the source text for that exact quote and anchors it to the character range it occupies. Clicking the citation opens the document at that passage with the text highlighted, so checking a claim is one click rather than a search. This is how document review works in Repository and Tabular Review, which are Enterprise-tier features.

A quote that fails the check does not become a link

Re-locating a quote in the source can fail — that is the case where a model has paraphrased something into quotation marks. When it fails, the citation is marked unverified, renders inert instead of clickable, and says so: “Quote could not be verified in the document.” The source viewer will not jump to a near match. Unverified is a state you can see, not a silent downgrade to a link that goes somewhere approximate.

Case-law citations are checked against the public record first

When an answer contains case citations, they are checked against CourtListener's citation-lookup API before the answer is delivered. Existence alone is not treated as sufficient: a made-up case name pointed at a real reporter volume and page resolves to a real but unrelated case, so the resolved case name must also match the one claimed. Citations that do not resolve, or that resolve to a different case, send the answer back for correction. Where a direct quote is attributed to a case, the quote is checked against the text of the opinion.

When the documents do not answer the question, that is the answer

Extraction is instructed to return no answer — not a plausible one — where the documents do not contain it or confidence is low, and the field is marked as having no answer rather than quietly filled. The assistant is instructed never to fabricate document ids, quotes, or citations, and to use only ids and text its tools actually returned. Where a verification step could not run at all, the system asserts nothing rather than reporting a clean check.

What is the best legal AI for in-house counsel at a startup?

For a startup legal department the practical answer is set by price and coverage, because the department is one or two generalists covering everything. White Shoe AI Associate is $49/month for 1 user and 40 hours of included capacity; Partner is $99/month for 3 users and 130 hours. Both are self-serve with published pricing and no seat minimum. If the department's dominant need is deep case-law research rather than commercial volume, CoCounsel or Lexis+ AI is the right tool instead.

Is White Shoe AI built for law firms or for in-house legal departments?

Both, and the in-house case is the one this page is about. A mid-market or startup legal department is the primary buyer: vendor agreements, NDAs at volume, board minutes, privacy questions from product, employment questions, and commercial contracts. The plan names are Associate and Partner, but the plans are sized by users and included hours, not by practice setting — Associate is 1 user, Partner is 3 users, Enterprise starts at 5.

Does White Shoe AI have a SOC 2 report?

No. White Shoe does not currently have a SOC 2 report. There is an active readiness program, but readiness work is not certification and does not establish operating effectiveness. If a completed SOC 2 report is a hard requirement in your security review, that is a real constraint and you should weigh it. The security page states the full position, including what is implemented today and what is not.

How do I know White Shoe AI will not make something up?

You check it, and the product is built to make checking fast. Where an answer rests on a document you supplied, it carries a verbatim quote plus the document and page; the system re-finds that exact quote in the source and anchors it, so the citation opens the document at the passage with the text highlighted. If the quote cannot be re-located in the source, the citation is marked unverified and rendered inert rather than linking to an approximate location. Case-law citations are checked against CourtListener's citation-lookup API before the answer is delivered, and because a fabricated case name pointed at a real reporter slot resolves to a real but unrelated case, the resolved case name must match the one claimed — citations that do not resolve, or resolve to a different case, send the answer back for correction. Where the documents do not contain the answer, extraction returns no answer rather than a plausible one. White Shoe has not published a scored accuracy benchmark and does not quote one. Outputs are informational and should be reviewed by a qualified legal professional before use.

How much does legal AI cost for a small in-house team?

Published prices vary by roughly an order of magnitude. White Shoe AI publishes $49/month for Associate and $99/month for Partner, which is $33.00/user/month at 3 users. Vaquill AI publishes $120/seat/month on Plus and $500/seat/month on Power (app.vaquill.ai/pricing, read Aug 1 2026). Irys publishes $299/month all-in and states no seat minimum (irys.ai, verified Jun 2026). Paxton AI publishes $499/month self-serve or $2,999/year (vaquill.ai/blog/legal-ai-pricing-benchmark, Jul 2026) and GC AI $500/seat/month (gc.ai, updated Jul 29 2026). Harvey, Legora, Spellbook, CoCounsel and Lexis+ AI do not publish pricing; those are quote-only. Third-party pages print estimated ranges for the quote-only vendors, but an estimate is not a published price and this page does not reprint them as one.

When is a different legal AI tool the right choice?

Three clear cases. If you need deep case-law research with a primary-law corpus and a citator, CoCounsel or Lexis+ AI is the right tool. If your problem is contract lifecycle management — repository, obligations tracking, renewals, approval routing — buy a dedicated CLM; an assistant sits on top of one rather than replacing it. And if you are a large legal department with a procurement function and a security-review budget, Harvey and Legora are built for that buyer and are strong at it.

Where does White Shoe AI actually run?

The Associates run in Web, Email, Slack, Microsoft Word, Chrome, and iOS. That matters more for in-house work than it does for firm work, because in-house questions arrive in Slack and email rather than in a matter file, and contract redlines happen in Word. Separately, an OAuth-protected MCP server hands approved workspace context to supported third-party AI clients — it is not a surface the Associates run in, and it is documented at whiteshoe.ai/platform/mcp. There are 31 AI Associates covering commercial, corporate, privacy, employment, and litigation work, plus Firm IQ as the organization-context layer.

Do I have to talk to sales to buy it?

No. Associate and Partner are self-serve at published prices with no seat minimum. Enterprise is quoted, but the quote formula is published too: it starts at $10,000/year for 5 users and 200 shared hours per month, $15,000/year at 10 users and 400 shared hours, $600 per additional user per year above 10, and additional usage in 200-hour monthly blocks at $1/hour.

In-house teams do not bill hours. What do the included hours mean?

The hours are the plan's included capacity, not a client-billing obligation. Associate includes 40 hours of work per month, Partner includes 130 across 3 users, and Enterprise includes 200 shared hours at 5 users. For a legal department the useful way to read that number is throughput, not billing.

Is customer content used to train AI models?

White Shoe does not train its own foundation models on customer content. Third-party model-provider handling depends on the applicable provider contract and configuration, which is why subprocessors are disclosed in the DPA rather than covered by a single blanket statement. Data is encrypted in transit over TLS and encrypted at rest via managed-provider protections, and designated third-party integration credentials receive an additional Google Cloud KMS layer.

Should an in-house team measure ROI on legal AI?

Yes, and most do not. Only 12% of in-house teams track technology return on investment, per the ACC / Everlaw CLO survey of 284 respondents published Nov 17 2025. The practical version is not a spreadsheet exercise: score a fixed set of real documents against an answer key you wrote yourself, and track revision time alongside issues found. A tool that surfaces every issue but requires a full rewrite has not saved the department anything.