Airbyte is expanding its AI agent platform with semantic search and granular access controls designed to address two of the biggest hurdles to putting enterprise AI agents into production: finding relevant business context and controlling what agents can see and change. The updates bring meaning-based retrieval and entity-level governance directly into Airbyte’s Context Store, giving organizations a way to connect agents to distributed business knowledge without relying on separate search and permission layers.

The changes come as enterprises move beyond AI pilots and begin deploying agents across sales, engineering, operations, and other business functions. Those deployments require access to information scattered across collaboration platforms while maintaining the security policies that govern human users.

Semantic Search Targets Unstructured Enterprise Knowledge

Airbyte Agents now supports semantic search across data from Google Drive, Gong, Granola, and Linear. These systems contain large amounts of organizational knowledge that is difficult to access through traditional keyword-based search, including meeting notes, call transcripts, documents, and issue discussions.

Instead of requiring an agent to find an exact keyword or phrase, semantic search retrieves content based on meaning and context. An agent could, for example, identify conversations about pricing objections or engineering problems even when the original discussion used different terminology.

That distinction becomes important as AI agents increasingly operate as research and decision-making assistants. The quality of their output depends not only on the model itself but also on whether they can retrieve the right business context.

Airbyte performs this retrieval through its pre-indexed Context Store, a replicated, search-optimized index that sits within the Agents platform. The company says this approach can reduce the amount of information that needs to be sent through model inference compared with repeatedly querying source APIs. Internal benchmarks showed up to 80% fewer tokens when querying Gong data and up to 75% fewer for Linear.

The company plans to extend semantic search to additional connectors in future releases.

Granular Policies Put Boundaries Around Agents

Finding information is only half the challenge. Enterprises also need to ensure that AI agents have access only to the systems and information appropriate for their roles.

Airbyte’s new entity policies extend the workspace capabilities introduced previously by allowing organizations to control access at the level of individual data connectors and sources. Read and write policies can be assigned to users and agents across workspaces, giving administrators more precise control over what can be discovered, accessed, or modified.

This can help organizations separate development, staging, and production environments, restrict access to sensitive business systems, and align agent permissions with existing security and compliance requirements.

The approach is notable because it puts governance inside the same platform responsible for retrieving enterprise context. Instead of creating a separate permission framework for AI agents, organizations can apply explicit policies to the agents and users accessing connected data.

Michel Tricot, Founder and CEO at Airbyte, said, “AI agents are only as valuable as the context they can safely access. Organizations don’t need another disconnected vector database or another permission system, they need agents that understand the information that already exists across their business while respecting the same governance policies employees rely on every day. These new capabilities move us another step closer to making enterprise AI both more useful and more trustworthy.”

The broader implication is that enterprise AI infrastructure is moving beyond model access toward a combination of connectivity, context, retrieval, governance, and execution. As organizations deploy more autonomous systems, those layers will increasingly need to work together rather than operate as isolated tools.

Airbyte’s latest additions position its Context Store as part of that emerging infrastructure, combining indexed enterprise knowledge with semantic retrieval and granular governance. For developers and platform teams, the challenge ahead will be determining how to expose enough organizational context for agents to be effective without expanding their access beyond what business and security policies allow.