Swiss AI knowledge management start-up DeepJudge has introduced a new open protocol designed to enable users to transfer ongoing work and its context between AI platforms.
The Agent Handoff Protocol (AHP), which is backed by Harvey and Thomson Reuters, aims to allow users to move between different AI products without having to rebuild elements such as source materials, conversation history and previous work.
Existing protocols such as the Model Context Protocol (MCP) allow AI-powered systems to exchange information and perform specific actions. However, DeepJudge says the wider context needed to continue tasks can be lost.
AHP is designed to preserve that context, allowing users to move an ongoing task from one AI platform to another without having to re-upload materials, recreate prompts or rebuild previous analysis.
In the open-source technical documentation, published on GitHub, DeepJudge explains that it created the protocol "to address a missing layer in the agent ecosystem: portable, user-controlled continuity between complete agent experiences".
The protocol promises to provide agent builders with "one predictable handoff contract instead of inventing a bespoke integration for every partner".
Yannic Kilcher, CTO and co-founder of DeepJudge, said: “No single AI experience will be the right one for every task. People should be able to use the AI tool that’s best for what they need to do and take their context with them. That is the experience we built Agent Handoff Protocol to enable."
Both Harvey and Thomson Reuters are supporting the protocol, with a Harvey integration entering beta stage this month.
Siva Gurumurthy, CTO at Harvey, said: “The most capable legal workflows will draw on more than one specialised system. Agent Handoff Protocol lets a user move from Harvey into a partner's product and back without losing the thread of their work, while leveraging both products to their full potential.”
Joel Hron, chief technology officer at Thomson Reuters, said: “Professionals will increasingly work across specialised AI experiences, and those systems need to work together without sacrificing the context, trust, and expertise required for high-stakes work. Open standards like Agent Handoff Protocol can help enable that interoperability, which is why Thomson Reuters is excited to support this effort and help shape what comes next.”
Thomson Reuters also struck a partnership agreement with DeepJudge in October last year, combining DeepJudge's AI-powered enterprise search for law firms and legal departments with Thomson Reuters’ legal platforms.
In May, Harvey and DeepJudge announced that they had entered into a partnership, with the DeepJudge platform allowing legal teams to integrate their previous work and decision-making into Harvey’s workflows. The aim is to help lawyers research, draft and analyse using AI.
Last November, DeepJudge raised $41.2m in an oversubscribed Series A funding round led by venture capital firm Felicis, which was supported by technology investment firm Coatue, valuing the company at $300m.
Zurich-based DeepJudge is an AI-powered legal research and knowledge management platform that performs searches of a law firm’s previous work, turning its internal knowledge into a searchable resource.
It was founded in 2021 by Kilcher, CEO Paulina Grnarova and COO Kevin Roth.
Prior to co-establishing the firm, all three had spells working for Google. Kilcher worked at Google AI Language, while Grnarova was at Google Brain for just over two-and-a-half years, latterly as a researcher. Roth also had a short three-month role at Google Brain as a research scientist.
The launch comes as legaltech providers increasingly look to connect their products with other AI and legal technology platforms. In March, Intapp struck a deal to integrate its compliance software with Harvey, while separately partnering with Anthropic to use its Claude models to build AI agents.
And in May, Anthropic said it had forged agreements to connect Claude with the software of more than 20 leading legaltech suppliers
Email your news and story ideas to: [email protected]
