Airbyte adds semantic search & tighter agent controls

Airbyte has added semantic search and fine-grained governance controls to its Airbyte Agents platform, targeting companies that want AI agents to access internal information under tighter access rules.

The additions expand the platform's Context Store, a replicated search index for business data used by AI agents. They are meant to address two common challenges as companies move AI systems into day-to-day use: finding relevant information across unstructured content and applying security policies to that access.

Semantic search is now available across content stored in Google Drive, Gong call transcripts, Granola meeting notes, and Linear issues and comments. The system retrieves information based on meaning rather than exact keyword matches, allowing agents to identify relevant discussions even when the wording in source material differs from a user's query.

That means an AI agent could surface references to pricing concerns, contract objections, rival products, login problems, or engineering issues even when those exact terms do not appear in the original content. It also lets staff ask questions in natural language across documents, meeting notes, and collaborative tools instead of relying on specific file names or exact phrases.

The search function runs on Airbyte's pre-indexed Context Store rather than making repeated calls to source application programming interfaces. In the company's internal benchmarks, that approach reduced token use by up to 80% when querying Gong and by up to 75% for Linear compared with native API methods.

Access controls

The second part of the update introduces entity policies for workspaces. These policies let organisations define which data connectors, data sources, and other resources can be discovered or accessed by users and AI agents within each workspace.

Airbyte had already introduced workspaces as separate environments for different users or teams, each with access to selected data connectors. The new entity-level rules add a more detailed layer of control, allowing businesses to set visibility and permissions by connector or data source rather than only at the broader workspace level.

Read and write policies can now be assigned to every user and agent for each data connector in every workspace. This is intended to let organisations share connectors between teams while limiting access to sensitive systems or information.

Such controls could restrict agent access to particular data sources, separate development, staging, and production environments, or align AI access with existing internal security requirements. The structure is also designed to avoid the need for separate permission systems for agents and employees.

Michel Tricot, Chief Executive Officer and Co-Founder of Airbyte, outlined the company's position on the update.

"AI agents are only as valuable as the context they can safely access," said Michel Tricot, Chief Executive Officer and Co-Founder of Airbyte. "Organisations 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."

Broader push

The release reflects a wider shift among software providers trying to turn experimental AI tools into systems that can be used in regular business processes. For many companies, one obstacle has been that useful information often sits across meeting notes, call transcripts, internal documents, and issue trackers rather than in neatly structured databases.

Another obstacle has been governance. Businesses adopting AI agents face pressure to ensure those systems do not gain broad access to internal material without controls that match existing rules for employees, teams, and business units.

Airbyte is positioning its platform at the intersection of data access and control. It describes Airbyte Agents as part of a wider data platform for AI agents, combining data replication with indexed search, software development tools, and integration options for clients that support the Model Context Protocol.

Additional data connectors will receive semantic search support in later releases, according to Airbyte. The company also said its wider platform is used by 7,000 enterprises to move structured and unstructured data across multi-cloud and hybrid environments.

The latest changes show how vendors in the AI infrastructure market are focusing less on model novelty and more on the practical issues of retrieval, permissions, and operational boundaries inside large organisations.