San Francisco Legal AI Startups: The 2026 Map (Harvey, Eve, GC AI, Ivo)

A no-hype map of San Francisco's legal-AI startups in 2026: who they are, what they do, who they're for, and how much they raised, from Harvey to Ivo.

San Francisco is where legal AI raises its money. In 2025 the category pulled in a record $2.4 billion, and the biggest pure-play, Harvey, is now valued at around $11 billion after raising more than $1.2 billion in under four years. If you want to see where legal AI is going, you read the Bay Area.

Most "top legal AI" lists blur three different things: software for lawyers, marketing agencies, and law firms that happen to use AI. This map is only the first kind, and only the companies actually clustered around San Francisco and the Peninsula. It also absorbs three pages we used to publish separately — on GC AI, Eve and Ivo — because splitting one buying decision across four thin pages helped nobody. For tool-by-tool scoring, see our benchmark and compare page.

How we built this map, and what to distrust in it

Three rules. First, software only: no marketing agencies, no law firms that practise AI law. Second, geography is literal — the five companies below are Bay Area, and the three reference points at the bottom of the table are labelled with where they actually are. Third, and most important: every description of a vendor here records what that vendor markets publicly, not the result of us testing their product. "English-first" means English is what the marketing and the documentation are in. It does not mean the underlying model cannot produce another language.

The same discipline applies to money. Where a pricing cell reads "Not published", it means we could not confirm a current figure on the vendor's own material as of 22 July 2026. Read that as unconfirmed, not as non-existent. Earlier versions of these pages carried per-seat prices we could no longer trace to a first-party source, so we removed them rather than repeat a number from a software directory. If you are building a budget, get the quote in writing from the vendor.

The 2026 San Francisco legal AI map

• Harvey — General legal copilot for BigLaw and enterprise in-house. ~$1.2B raised, ~$11B valuation. San Francisco.
• Eve — Operating system for US plaintiff personal-injury firms. ~$164M raised. San Mateo.
• EvenUp — AI demand letters for plaintiff injury firms. ~$135M Series D, unicorn. San Francisco.
• GC AI — In-house counsel copilot. ~$73M raised. SF / San Mateo.
• Ivo — Contract review and redlining inside Word, enterprise in-house. ~$77M raised. San Francisco.

Funding and valuation figures are as reported in 2026 and they decay fast. Re-check them before you put one in a procurement memo — a stale valuation is the easiest thing in this market to get wrong.

Two names often grouped into San Francisco legal AI lists are not actually here: Sandstone (Brooklyn) and Ruli (Austin). And a search trap worth knowing: "San Francisco AI lawyer" mostly returns law firms that practice AI law, or job postings, not software.

Note the pricing column. We could not confirm a first-party, self-serve price for any of the five San Francisco companies as of 22 July 2026 — each of them sells through a conversation. That is not a criticism; enterprise software is sold that way for good reasons. It is simply the fact that decides whether a solo practitioner can buy the thing this afternoon.

What every Bay Area legal AI has in common

Run down the list and the same four assumptions repeat. English is the working language of the product and the documentation. The law is US common law, which shapes the citation formats, the research databases and the way the model is evaluated. The workflow is desktop, usually with Microsoft Word somewhere in it. And the buyer is an organisation — a firm, a legal department — with a software line in its budget and someone whose job includes running a procurement process.

None of that is a mistake. It is the rational choice when your investors are on Sand Hill Road and your comparables are US enterprise SaaS. A company that raises at a venture-scale valuation has to sell seats at a price and a volume that only large US organisations can absorb. Every product decision downstream — the language, the jurisdiction, the surface, the sales motion — follows from the funding decision upstream.

The consequence is a market where the most capable legal AI in the world is systematically pointed away from most of the lawyers in the world. A civil-law practitioner in Cairo, a two-person firm in Amman, an advocate who does most of their work from a phone between courthouses — none of them are in any of these companies' ICP documents, and it would be strange if they were.

Two honest caveats before the comparisons. First, "not marketed for X" is not the same as "cannot do X" — a lawyer in Beirut can open Harvey and get useful English output. The gap is in the corpora, the citation behaviour, the jurisdictional guardrails and the support you get when it is wrong. Second, the US-built cohort is not the only alternative to itself. Lexzur, for one, ships native Arabic, right-to-left interfaces and a Hijri calendar, and has done for years. The category is not "American tools versus us".

GC AI alternative: who GC AI fits, and who it leaves out

"GC AI alternative" is the most common of the three searches, and it is almost always asked by one of two people: an in-house lawyer whose budget request was refused, or a private-practice lawyer who saw a demo and wondered whether the same thing exists for them.

What GC AI is

GC AI is a copilot aimed squarely at in-house counsel — the corporate legal teams at US technology and consumer companies. It drafts, reviews and researches with citations, and it works alongside Microsoft Word, which is where the documents in a corporate legal department actually live. Within that lane it is a genuinely strong product, and the team behind it has been unusually direct about who it is for.

Who it is built for

One to twenty lawyers sitting inside a US company, handling commercial contracts, vendor paper, employment questions and the constant stream of "can we do this?" from the business. English, US law, Word, Slack, a budget owner. If that description fits you, stop reading comparison pages and book the demo.

Pricing and access

Not published, as of 22 July 2026. A per-seat monthly figure circulated widely for GC AI — including on an earlier version of this page — and we could not re-confirm it against the vendor's own material, so we pulled it. Ask them directly. What we can say without a number: the sales motion is a conversation, not a signup form, which tells you the target buyer has a procurement process.

Where it leaves you

The product's centre of gravity is the US corporate legal department. If your work is Arabic-language contracts under a civil-law code, or family and labour matters in a Gulf jurisdiction, or anything where the authority you need to cite is a code article rather than a case, you are outside the design target. Not blocked — outside. The difference matters when you are the one checking the citation.

When HAQQ fits better

HAQQ is built for the lawyer GC AI was not aimed at: the solo or small-firm practitioner, often outside the US, working in Arabic or French under a civil-law system, frequently from a phone. There is a free tier, the plans are published, and the mobile surface is HAQQ Legal AI Chat rather than an add-in for a desktop word processor. If you also need the firm side — matters, billing, trust accounting — that is eFirm, and it is a different product from the chat app.

When to pick GC AI anyway

If you are in-house at a US company, work mostly in English and US law, and live inside Microsoft Word, GC AI is likely the better fit and we would rather you use the right tool than the wrong one. The same is true if your legal department has already standardised on a Word-native review workflow — fighting that is a change-management project, not a software purchase.

Eve legal AI alternative: the specialist problem

The "Eve legal AI alternative" search has a different shape. People rarely leave Eve because it is bad at its job. They search because they discovered it is built for one specific job and theirs is a different one.

What Eve is

Eve is an operating system for US plaintiff personal-injury firms: intake, medical-record chronologies, demand letters, case valuation and caseload audits. It is not a general legal assistant with a PI module bolted on; the whole product is shaped around the economics of contingency-fee injury work in the United States.

Who it is built for

High-volume US plaintiff injury practices, where the bottleneck is turning a pile of medical records into a defensible demand package fast enough to keep the case pipeline moving. For that firm, Eve and EvenUp are the two serious names and the comparison is between them, not with anything on this page.

Pricing and access

Not published, as of 22 July 2026. The entry point is a demo request.

Where it leaves you

Everywhere that is not US personal injury. A general practice doing contracts, advisory work, family law and corporate matters will find that most of Eve's depth simply does not apply — medical chronology tooling is worth nothing to a lawyer drafting a shareholders' agreement. Specialisation is why it is good, and it is also why it does not travel.

When HAQQ fits better

HAQQ is for lawyers whose work is not US injury litigation: general practice, contracts and advisory, across civil-law and MENA jurisdictions, in Arabic, French or English, on a phone, priced for an individual. The trade is real and worth naming — you get breadth across practice areas and languages instead of a deep vertical workflow for one US case type. Compare the broader field in the best legal AI tools of 2026.

When to pick Eve anyway

If you run a US personal-injury firm, Eve (or EvenUp) is built for exactly that workflow and HAQQ is not. Use the specialist for the specialty. This is not diplomacy; a general assistant that has never seen an ICD code is the wrong tool for a medical chronology.

Ivo alternative: contract review outside the enterprise

"Ivo alternative" is usually a price-and-shape question. Ivo does one job extremely well, and the people searching for an alternative are typically people for whom the job is the same but the shape of the firm is not.

What Ivo is

Ivo reviews and redlines commercial contracts inside Microsoft Word and Google Docs, benchmarked against a company's own playbooks, for enterprise in-house legal teams up to the Fortune 500. The playbook layer is the point: it encodes the positions your organisation has already agreed to take, so review becomes a check against a standard rather than a fresh reading every time.

Who it is built for

A corporate contracts team with volume — enough NDAs, MSAs and DPAs each month that the marginal cost of a human first pass has become the constraint, and enough institutional memory to have written playbooks in the first place. A three-lawyer firm has neither.

Pricing and access

Not published, as of 22 July 2026. An annual per-seat figure appeared on an earlier version of this page and we could not re-confirm it first-party, so it is gone. Assume enterprise annual contracting until the vendor tells you otherwise.

Where it leaves you

Two gaps. The obvious one is the buyer profile: a per-seat enterprise contract is not a purchase a solo practitioner makes. The subtler one is the corpus — the product is marketed around US and UK commercial contracting, and a playbook-driven review engine is only as good as the body of contract norms it was tuned against. Arabic-language contracts governed by a civil code are a different body.

When HAQQ fits better

HAQQ is a broad, multilingual assistant rather than a Word plugin sold by the seat. If your contracts are in Arabic or French, governed by civil-law codes, and you work from a phone, the enterprise-English-Word model is the wrong shape for you. See also AI contract review for lawyers and Spellbook alternatives for the Word-native end of the market.

When to pick Ivo anyway

If your job is high-volume commercial contract review in Word at an enterprise, against established playbooks, Ivo is built for that and HAQQ is not trying to be. The same applies if your review workflow has to stay inside Word for audit or version-control reasons that predate the AI question entirely.

Where this page does not help you

Ranking pages that never admit a limit are marketing. Here are ours. This page cannot help you choose an eDiscovery platform for US litigation — that is a different category with different vendors. It will not tell you which tool is best for US case-law research, where the incumbents with the case databases have a structural advantage nobody on this list can argue away. It does not cover patent prosecution, and it does not price anything, because we could not confirm first-party prices for the vendors on it.

It also cannot tell you whether a tool is accurate enough for your matter. That is a testing question, not a comparison question, and the only honest answer is to run your own documents through a trial and check the citations yourself. Our benchmark work is a starting point, not a substitute.

How to choose

Start with three questions, in this order. What legal system does your work sit in — common law or civil law? What language do your documents and your authorities come in? And who signs the invoice, you or an employer? Those three answers eliminate most of this market before you look at a single feature list.

If you are a US, English-speaking lawyer at a firm or in-house team, this map is your shortlist and the rest of this page is context. If you are a solo, small-firm, civil-law or non-English lawyer, most of these tools were not built for you — which does not mean they are useless, only that you are not who the roadmap is written for. That gap is the one HAQQ is built for: multilingual, civil-law-aware, mobile-first, and free to start. See also legal AI for small law firms.