Sodura Note #002
Legora: vertical integration for legal work
From Leya in Stockholm to a layered “agentic operating system” aimed at large firms and in-house teams—and the clearest head-to-head with Harvey.
This note maps Legora’s “agentic OS” stack—models through Word/Outlook interfaces—and how that vertical bet compares to Harvey, with press-reported funding figures called out as disclosures, not audits.
Legora sells Legora aOS—branded as an “agentic operating system for legal work.” The pitch is not a single chat box. It is a vertically integrated stack where information, communication, and execution stay in one loop so lawyers can move faster without splintering work across disconnected tools.
On its homepage, Legora diagrams that stack as a sequence:
1. Large language models — the general reasoning layer.
2. Agentic harness — planning, tool use, and multi-step execution (its “Agent” product sits here).
3. Data and integrations — firm systems, DMS, and external sources wired in.
4. Context and knowledge — retrieval, matter memory, and curated legal corpora.
5. Legal capabilities — research with structured citations, document parsing, bulk review, workflows tuned to practice areas.
6. Products and interfaces — Word and Outlook add-ins, editor, mobile app, client Portal, tabular review, monitors, lists, and more.
7. Security and governance — SOC 2, ISO 27001/42001, GDPR, HIPAA positioning, and “zero training on your data” messaging.
That ordering is the strategic claim: own the harness, the legal-specific layer, and the places lawyers already work, instead of stopping at a model API or a thin wrapper.
From Leya to “OS”
The company started in Stockholm in 2023 as Leya, joined Y Combinator’s Winter 2024 batch, and rebranded to Legora in early 2025 as ambition grew from “AI assistant for lawyers” toward the aOS framing. Wikipedia and YC’s company profile document the Leya → Legora name change and the W24 batch; treat older “Leya” press as the same entity.
Who they sell to: large law firms and in-house legal teams—buyers who need security reviews, matter-level isolation, and workflow fit across practice groups.
Scale (company marketing): Legora’s newsroom boilerplate cites more than 130,000 legal professionals at more than 2,000 firms and in-house teams in 80+ markets. Its customers page sometimes shows slightly different totals as pages update; use the newsroom figures as the current public line, not as verified census data. The company also states 875+ employees on its site as of our check.
Valuation (completed round, press-reported): TechCrunch and Reuters covered a ~$5.6 billion post-money valuation after Legora’s $550 million Series D (March 2026) and a $50 million extension (April 2026) that added investors including NVentures and Atlassian. That is the latest closed round journalists reported—not a live trading price.
Why vertical integration matters here
Legal work fails in the gaps: a strong answer in a browser that never reaches the redline in Word, or a review batch that cannot cite the clause it relied on. Legora’s story is that the aOS “facilitates the flow of information, communication, and execution” so outputs stay tied to delivery surfaces (Word, Outlook, portal) under shared governance.
Contrast that with a horizontal “legal GPT” that stops at the chat window. Integration depth is the moat they are betting on—especially for M&A, litigation, banking, tax, and other document-heavy practices they highlight on the site.
The rival in the mirror: Harvey
The most cited peer is Harvey—another well-funded legal AI platform aimed at elite firms and in-house teams. Press in spring 2026 framed the race as Legora at ~$5.6B versus Harvey at ~$11B after Harvey’s own large round. The products differ in geography, GTM, and feature emphasis, but the buyer and the budget line are often the same: firm-wide AI for research, drafting, and review with enterprise security.
For builders outside the US–EU bubble, the lesson is structural: winning categories may converge on full-stack legal OS plays, not single-feature bots.
Product lesson: if your domain has expensive document loops and strict provenance, ask whether you are shipping a layer or the loop—models, agents, knowledge, domain tools, interfaces, and governance together.
Sources checked 2026-09-28. Customer, headcount, and valuation figures are company or press disclosures unless noted; they are not independent audits.