DeepJudge
by DeepJudge
Institutional-knowledge search and AI workflows for law firms, with no data migration
DeepJudge is an AI knowledge platform for law firms and in-house legal teams, built by three ETH Zurich PhDs who previously worked on Google search. It indexes a firm's document management system, email, SharePoint and intranet in place — no migration, existing permissions and ethical walls respected — then runs search and multi-step LLM agent workflows over that institutional knowledge.
DeepJudge is a Zurich-based institutional-intelligence platform for legal AI, founded in 2021 by ETH Zurich PhD graduates Paulina Grnarova, Kevin Roth and Yannic Kilcher, several of them former Google search engineers. The product has two halves. DeepJudge Knowledge Search indexes a firm's existing data where it already lives — iManage and other document management systems, SharePoint, OneDrive, HighQ, email and intranet — automatically classifying documents and detecting duplicates across billions of data points, and searching across clients, matters, documents and people while continuously synchronising the firm's existing access permissions and ethical walls. DeepJudge AI Workflows sits on top and lets firms build, manage, orchestrate and govern LLM agents that execute multi-step tasks, with named applications for negotiation intelligence over past agreements, multi-document chat and matter and client overviews with generated timelines. The deliberate architectural choice is that nothing moves: the company describes it as a data-first approach with no complex implementation and no data migration, and it is LLM-agnostic, letting firms choose the model and the data residency. DeepJudge raised a $10.7 million seed in June 2024 and a $41.2 million Series A in November 2025 led by Felicis with continued participation from Coatue, reported at a $300 million valuation, for $51.9 million total. Customers named publicly include Freshfields, Holland & Knight, Greenberg Traurig, Gunderson Dettmer, Cozen O'Connor, CMS and Homburger. On 13 August 2026 DeepJudge published the Agent Handoff Protocol, an open Apache-2.0 specification for passing a user and their working context between independently operated AI applications, which Harvey and Thomson Reuters both committed to adopt.
A knowledge-management or innovation lead at a large law firm sitting on decades of work product in a DMS that nobody can find, who cannot move that data for confidentiality reasons.
Firm-wide precedent and negotiation history becomes searchable and agent-usable in place, with existing permissions and ethical walls enforced rather than reimplemented.
At a Glance
- Category
- Enterprise Search & Knowledge
- Pricing
- Contact for pricing, Subscription
- Target Market
- Law Firm CIOs, Knowledge Management Directors, General Counsel, Legal Innovation Leads, Managing Partners
- Deployment
- Cloud-first, Self-hosted, Hybrid
- Founded
- 2021
- Headquarters
- Zurich, Switzerland
- Customers
- 15+ named law firm customers including Freshfields, Holland & Knight, Greenberg Traurig, Gunderson Dettmer, Cozen O'Connor, CMS and Homburger
Key Features
- ✓In-place indexing with no data migration
Connects to iManage-class DMS, SharePoint, OneDrive, HighQ, email and intranet and indexes where the data lives, removing the migration project that normally blocks legal AI deployments.
- ✓Permission and ethical-wall synchronisation
Access rights are continuously synchronised from source systems and least-privilege is enforced, so search results never expose a matter a user could not already open.
- ✓Semantic knowledge search across the firm
Understands content, context and relevance to search across clients, matters, documents and personnel, with automatic classification and duplicate detection across billions of data points.
- ✓AI Workflows for governed legal agents
Lets a firm build, manage, orchestrate and govern multi-step LLM agents on its own knowledge, rather than running ungoverned prompts against a general-purpose chatbot.
- ✓Negotiation intelligence over past agreements
Searches historic executed agreements to surface the positions the firm has previously taken and conceded, turning precedent into leverage during live negotiation.
- ✓Agent Handoff Protocol
Open Apache-2.0 specification, published August 2026, that transfers objective, conversation history, resources and thread identity between AI products so context survives a switch.
- ✓Choice of deployment and model
Runs in DeepJudge's cloud, the client's cloud or on-premises, is LLM-agnostic, and lets the firm select data residency — the flexibility large firms' security reviews demand.
Capabilities
Use Cases
- •Finding the firm's own precedent fast
An associate drafting a novel clause locates the closest prior work product across the firm's whole DMS instead of asking partners by email.
- •Negotiation preparation from executed agreements
A deal team reviews positions the firm has historically taken and conceded on a clause type before entering a live negotiation.
- •Matter and client onboarding briefings
A lawyer new to a matter gets a generated timeline and summary of every related document rather than reading a decade of correspondence.
- •Handing context to a specialist legal AI
Via the Agent Handoff Protocol a lawyer moves from DeepJudge research into Harvey or CoCounsel with objective, materials and prior analysis already loaded.
- •Governed multi-document review
A knowledge team builds an auditable agent workflow that queries a defined document set, with audit logging and access controls applied throughout.
Ideal For
Best For
- ✓Law firms with 20+ lawyers whose institutional knowledge is trapped in a document management system, email and intranet
- ✓Finding precedent, past negotiating positions and prior work product across clients, matters and people
- ✓Deployments where confidentiality or regulation forbids migrating client data into a vendor's own store
- ✓Firms that need existing access controls and ethical walls enforced automatically rather than rebuilt inside a new tool
- ✓Building governed, multi-step legal agent workflows on top of firm-specific knowledge rather than public law
- ✓In-house legal departments consolidating contract history for negotiation intelligence
Not Ideal For
- ✗Solo practitioners and small firms — the platform targets firms of 20+ lawyers, pricing is enterprise and quote-only, and there is no self-serve tier
- ✗Buyers outside legal; unlike horizontal enterprise search such as Glean, DeepJudge is built around matters, clients and ethical walls and does not generalise to other departments
- ✗Teams that want published pricing before engaging sales — DeepJudge lists none, and enterprise legal AI seats commonly start around $500 per month
- ✗Firms wanting a single end-to-end legal AI suite from one vendor; DeepJudge is explicitly the knowledge layer and expects to hand work off to Harvey, CoCounsel and others
Deployment
Market & Ratings
15+ named law firm customers including Freshfields, Holland & Knight, Greenberg Traurig, Gunderson Dettmer, Cozen O'Connor, CMS and Homburger
Market Analysis
Pros
- ✓No-migration architecture removes the single biggest blocker to legal AI deployment — moving privileged client data
- ✓Permission and ethical-wall synchronisation is enforced from source systems, which is a hard requirement in law firms and a common failure point for horizontal search tools
- ✓SOC 2 Type II and ISO 27001 certified with on-premises, client-cloud or vendor-cloud deployment and selectable data residency
- ✓Credible reference customers among global elite firms, including Freshfields, Holland & Knight and Gunderson Dettmer
- ✓Publishing the Agent Handoff Protocol as an open Apache-2.0 spec — implementation code withheld, so no vendor lock-in — established DeepJudge as a standards-setter before competitors proposed alternatives
Cons
- ✗No published pricing, no free trial and no self-serve tier, so evaluation requires a sales cycle and the platform is out of reach below roughly 20 lawyers
- ✗The Agent Handoff Protocol is still Draft v1 with 65 GitHub stars and only DeepJudge shipping an implementation; Artificial Lawyer noted it risks becoming niche infrastructure without broad ecosystem adoption
- ✗That same coverage flagged unaddressed questions the spec does not answer — the security implications of passing detailed work context between systems, data custody during a handoff, and whether firms will even permit lawyers to work across multiple AI platforms
- ✗The protocol currently specifies one-way handoffs with optional return; multi-application chains are unspecified
- ✗No verifiable user reviews exist on G2, Capterra, TrustRadius or Product Hunt, and Hacker News and Reddit carry no practitioner discussion, so there is no independent read on day-to-day quality
- ✗DeepJudge's $51.9M total is small next to Harvey and Thomson Reuters, who are simultaneously its protocol partners and the platforms most able to absorb the knowledge layer themselves
- ✗The security page does not state GDPR or SSO specifics, which a Swiss and EU customer base will ask about directly
Pricing
DeepJudge Knowledge Search
Contact for pricing
- ✓In-place indexing across DMS, SharePoint, OneDrive, HighQ, email and intranet
- ✓Permission and ethical-wall synchronisation
- ✓Automatic classification and duplicate detection
- ✓Choice of cloud, client cloud or on-premises deployment
DeepJudge AI Workflows
Contact for pricing
- ✓Build, manage, orchestrate and govern LLM agents
- ✓Negotiation intelligence and multi-document chat
- ✓Matter and client overviews with generated timelines
- ✓LLM-agnostic model selection
Agent Handoff Protocol
$0
- ✓Open Apache-2.0 specification on GitHub
- ✓Draft AHP v1 wire contract
- ✓Free for any vendor to implement
DeepJudge publishes no pricing at all — every commercial route is a sales conversation, which is normal for enterprise legal AI but means buyers cannot budget without engaging. The platform targets firms of 20 or more lawyers, so there is no self-serve or small-firm entry point, and no free trial is advertised. For calibration, third-party 2026 legal-AI pricing surveys put enterprise platforms such as Harvey and CoCounsel at roughly $500 per seat per month with annual commitments, while the broader legal AI market spans free tiers to over $1,200 per seat; DeepJudge's own position within that range is not public and should not be inferred. The Agent Handoff Protocol specification is separately free and open under Apache 2.0.
Security & Compliance
Connect
Sources
This page was written from 9 sources, 6 on domains other than deepjudge.ai.
- 1.deepjudge.ai — deepjudge.aivendor
- 2.deepjudge.ai — securityvendor
- 3.deepjudge.ai — deepjudge introduces agent handoff protocol with harvey and vendor
- 4.lawnext.com — deepjudge releases an open protocol for passing users and th
- 5.artificiallawyer.com — deepjudge launches agent handoff protocol harvey tr adopt
- 6.github.com — agenthandoffprotocol
- 7.legaltechnologyhub.com — deepjudge
- 8.venturelab.swiss — USD 42M Series A for DeepJudges AI knowledge platform for la
- 9.forbes.com — meet the ex google researchers building ai search for law fi
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
Subscribe