Lagrange Labs has launched Halo, a coordination engine designed to keep swarms of drones, robots and uncrewed vessels operating in agreement when communications degrade or networks fragment. The system runs beneath existing autonomy stacks without requiring teams to rewrite deployed mission software. Halo operates across three layers: full-group consensus when time permits, independent operation when connectivity is severed, and sub-200-millisecond threat responses using locally held operational pictures. The shared-state protocol is authenticated, ordered and tolerant of patchy connectivity. At launch, Halo includes four reference coordination modes and developer interfaces compatible with Modular Open Systems Approach standards. A browser-based playground allows operators to test system behaviour under failure conditions. Over the next six months, Lagrange Labs will conduct field trials under electronic-warfare pressure and GPS denial, expanding to ground and maritime domains.

Talus partners with Lagrange to solve the trust problem in AI agents. Talus Labs, Inc. is excited to partner with Lagrange, the team behind DeepProve, the world's leading zero-knowledge machine learning library (zkML). This collaboration brings cryptographic verification to Talus' AI agents, solving one of the most critical challenges in decentralized AI: trust in agent actions. The trust problem in decentralized AI. As AI agents become more autonomous and handle increasingly critical tasks, from managing digital assets to executing complex on-chain workflows, the question of trust becomes essential. How can users verify that an AI agent executed correctly? How can Talus Labs, Inc. prove an agent used the intended model and produced legitimate results? Today's AI systems operate largely as black boxes. Users must simply trust that agents are functioning as intended, with no way to independently verify their operations. This gap between capability and accountability creates fundamental barriers to adoption, especially in high-stakes environments like finance, healthcare, and autonomous transactions. Lagrange's DeepProve: proof you can trust. Lagrange addresses this challenge with DeepProve, a zero-knowledge machine learning library that generates cryptographic proofs for AI inferences. DeepProve proves that a model ran correctly and produced accurate results without revealing sensitive data or proprietary logic. The technology works by splitting AI inference into multiple parts, proving each independently using parallel machines, and aggregating them into a single compact proof. This architecture makes it possible to verify complex AI operations without sacrificing performance or exposing intellectual property. DeepProve has already demonstrated real-world viability. DeepProve is currently production-ready for major LLM models, such as GPT-2 and Gemma3, as well as MLPs and CNNs, with integration pilots underway with enterprise, DeFi, and defense partners. Building verifiable agent workflows on Talus. This partnership enables Talus to integrate cryptographic verification directly into its agent execution layer. By integrating DeepProve as a tool into Nexus, its decentralized automation framework for AI agents, Talus Labs, Inc. is creating a new standard for verifiable autonomous systems. The integration unfolds in two phases: Phase 1: Technical Integration PoC (Proof of Concept) Talus Labs, Inc. is co-developing a proof of concept that integrates DeepProve as a verification step within Talus agent workflows. This foundational work establishes the infrastructure for agents to generate cryptographic proofs of their on-chain decisions, which dramatically increases technical trustworthiness. For Lagrange, this provides a concrete, working use case for their zkML technology integrated into Talus's agent workflows. Phase 2: Product and Ecosystem Expansion Verifiable AI becomes an integrated core value proposition across its products and ecosystem. For idol.fun, its AvA Markets platform, this means Talus Labs, Inc. can upgrade the fairness of agent competitions with ZK proofs to maximize trust in its Agent vs Agent (AvA) markets. For Talus Vision, its workflow builder, Talus Labs, Inc.'ll differentiate by adding a verifiable workflow template. This phase secures a real-world application for Lagrange's technology within the growing agent ecosystem of idol.fun, and provides potential customers through the Talus Vision builder community. Why this matters. The convergence of AI agents and blockchain technology creates immense possibilities, but only if Talus Labs, Inc. can solve the trust equation. Users need confidence that agents are executing as intended. Developers need tools to build verifiable systems without sacrificing performance. Enterprises need compliance guarantees before deploying autonomous systems at scale. Lagrange has the potential to be a core strategic partner that solves the fundamental problem of trust in Crypto AI. In turn, Talus provides Lagrange with real-world use cases for zkML technology. For Talus, this partnership solves the fundamental challenge of trust in decentralized AI. Its agents can now provide cryptographically verifiable outputs beyond the workflow execution being onchain, dramatically increasing their reliability and opening new possibilities for autonomous on-chain execution. For Lagrange, Talus provides practical use cases where zkML technology addresses actual market needs. From AvA gaming to enterprise workflows, its ecosystem offers diverse environments to demonstrate the value of verifiable AI. Looking ahead. Lagrange is expanding DeepProve to support broader LLM families (GGUF, ONNX), enhanced SDK tooling, and additional cross-chain integrations in the coming months. As these capabilities mature, Talus agents will inherit increasingly sophisticated verification mechanisms. The future of AI isn't just about making agents smarter. It's about making them trustworthy. By combining Talus' agent orchestration with Lagrange's cryptographic verification, Talus Labs, Inc. is building the foundation for AI systems that don't just work well, but prove they work correctly. Participate in its testnet. On September 18th, with the launch of its testnet, Talus Labs, Inc. introduced its vision of what AI agents can become when they have autonomy and economic purpose. There's $125,000 in cash waiting to be claimed, plus physical merchandise, surprise rewards, and leaderboard points that unlock exclusive benefits.Learn more about testnet participation and rewards here Talus is the premier infra for decentralized AI agents, an AI network that enables developers to launch revenue-generating agents. Talus Labs, Inc. is pioneering agents vs agents (AvA) Markets, with idol.fun as the killer app platform. Talus sits at the unique intersection of AI agents, decentralized infrastructure, and consumer AI building the foundational layer that enables AI agents to compete & earn in verifiable, transparent information markets. To learn more about Talus, follow Talus Labs, Inc. on twitter, check out its github, visit its discord, and read its blog. About lagrange. Lagrange is pioneering safe AI with DeepProve, the world's leading zkML library. DeepProve enables developers to cryptographically prove that AI models ran correctly and produced accurate results without revealing sensitive data or proprietary logic. This breakthrough makes it possible to build safe, private, and verifiable AI for high-stakes industries and Web3 systems alike. The team has raised $17.2M from top-tier investors, including Founders Fund, Archetype, and 1kx. DeepProve has already partnered with key enterprise players like NVIDIA, Intel, and more. Learn more at lagrange.dev.