Band raises $17M to build AI agent interaction mesh

Band, a startup based in Tel Aviv and San Francisco led by CEO Arick Goomanovsky and CTO Vlad Luzin, emerged from stealth with a $17 million seed round to develop an interaction mesh for autonomous AI agents used in enterprises.

Enterprises now run independent AI agents across engineering, customer support and security functions. When those agents need to share context, transfer tasks or operate across different clouds and frameworks, integrations can fail and engineering teams often must connect systems manually.

Band’s platform is designed to sit between autonomous actors rather than replace them. The mesh enforces routing rules, error recovery and authority boundaries at runtime. It applies fine-grained access controls to interactions and logs each exchange with cryptographic records so organizations can trace automated decisions back to their source.

The company frames demand around three conditions: agents have moved into regular operations and take real actions; enterprise environments are heterogeneous, with teams using different frameworks, clouds and isolated vector databases; and standards such as the Model Context Protocol and early agent-to-agent communication efforts define handshakes but do not manage runtime routing, limits or governance.

Band highlights a financial risk from unmanaged multi-agent workflows. Continuous API calls between agents can trigger extended inference cycles. A routing error or loop between agents can consume substantial compute budgets within hours. To address this, the mesh includes financial circuit breakers and token-budget enforcement that halt interactions when they exceed preset cost or compute thresholds.

The product also targets integration with legacy enterprise systems, including on-premises data warehouses, mainframes and custom ERP installations. The mesh enforces capability limits to prevent simultaneous writes or conflicting changes that could corrupt databases. For vector databases used for contextual memory, the platform aims to preserve provenance by allowing agents to exchange cryptographically verified context rather than relying on lossy summaries.

Governance features are built into the operational layer. The mesh records delegation chains, enforces authority limits and maintains audit trails of runtime actions. Human approval gates and collaboration controls are integrated so automated workflows remain subject to organizational policies and compliance requirements.

Band says the seed funding will support product development and go-to-market work. The company positions the platform as framework-agnostic and cloud-agnostic and is targeting enterprises moving from single-model pilots to networked deployments of specialized agents while maintaining security, cost controls and regulatory traceability.