DigitalOcean opens private preview of Agent Data Services for AI agents

One managed place for agent context

DigitalOcean argues that agents are only as capable as the context they can reach. In its announcement, the company says that context is usually scattered across documents, video and audio libraries, operational databases and execution logs, each with its own format, access pattern and freshness guarantees. Making it usable, according to DigitalOcean, means building and maintaining transcription jobs, indexing pipelines and sync scripts.

Agent Data Services is designed to remove that work. Data Flows process, enrich and index new data as it arrives, so multimodal content, structured records and agent sessions land in one governed catalog that can be searched by meaning in a single call. DigitalOcean says the service runs on the same cloud as a customer's compute and models, so teams can build search, analytics and agents on their own data without running their own pipeline infrastructure.

What the service includes

DigitalOcean lists four foundational capabilities:

• Managed Data Flows: new data is processed, vector-indexed and enriched by the flows a customer sets up, with no pipelines to manage.
• Data Catalog: structured records and vectors sit together in a single governed catalog, with lineage back to the source.
• Query through SQL and API: data can be queried in SQL or called from agents through an API.
• Built on DigitalOcean: it runs on the platform where compute, inference models and managed databases already live.

The company says the platform handles the lifecycle from raw ingestion and automated multimodal processing to indexing and real-time execution tracing. Everything is stored in one logically isolated catalog that can be queried through standard SQL, vector search or developer APIs. It can integrate with DigitalOcean Managed Agents or run as a standalone workspace, and it can connect internal and external sources, from DigitalOcean Spaces to third-party cloud storage providers.

On governance, DigitalOcean lists scoped access per workspace, roles from a customer's DigitalOcean team, API tokens, and audit logging for workspace and access events. It also says customers get no separate vector database to provision and maintain.

Use cases and the agent feedback loop

DigitalOcean points to three example uses. The first is capturing multi-step agent execution loops and feeding session state back to RAG applications in real time. The second is making video libraries searchable by meaning, subtitle and category without manually tagging pipelines. The third is unifying images, audio recordings, call transcripts and product documentation into a single queryable endpoint.

The service also captures runtime execution traces and behavioral telemetry. DigitalOcean says it converts agent turns, tool calls and session history into structured, queryable context, which feeds its observability and evaluation tools, including Insights and Agent Learning Services (Signals, Evaluations and Simulations). It is also described as the data layer for DigitalOcean Learning Services, so teams can use recorded sessions to test changes and improve their agents over time. Models running on DigitalOcean can pull permitted, up-to-date context from the service without custom pipeline glue code, the company says.

Availability and what it means

Agent Data Services is available to participants accepted into the Private Preview. DigitalOcean states that preview services are provided as-is under its preview terms, without any service level agreement or support commitment, and are not covered by its standard compliance certifications. It adds that features, availability, pricing and timing are in development and may change, and that the post is not a commitment to deliver any functionality.

For teams already building on DigitalOcean, the pitch is fewer moving parts between their data and their agents, along with a way to store agent traces next to the knowledge agents draw on. Because access is limited to accepted participants and no pricing or general availability date is given, the practical impact will depend on how the service develops beyond the preview.