Datris Releases Open-Source Data Control Plane for AI Agents

New release adds credential brokering, isolated execution, per-action policy, audit logging, row-level provenance, and automated recovery.

NEW YORK, NY, UNITED STATES, September 22, 2026 /EINPresswire.com/ — Datris today announced a major update to its open-source platform, which the company now describes as a data control plane for AI agents. The release adds credential brokering, isolated script execution, per-action policy, audit logging, row-level provenance, and an automated recovery agent. The features allow agents using the Model Context Protocol (MCP) to acquire, validate, and land production data without holding credentials or running unsupervised code. The self-hosted software is available immediately under the AGPL-3.0 license.

The update comes as organizations move AI agents from pilot projects into production data work, raising questions about agent identity, permissions, credential handling, and auditability. Datris first released the platform in April 2026 with support for agents building and operating data pipelines over MCP. According to the company, the features in this release respond to requests from teams evaluating that model for production use.

“In April we showed that an agent could run the whole data loop. The harder problem is production, where an agent is an identity that has to be scoped, code that has to be contained, and an actor that has to be accountable,” said Todd Fearn, founder of Datris. “The agent gets the data. The team keeps control.”

Access and policy
All capabilities are exposed through a single MCP server. Each agent connects with its own key and is shown only the tools that key permits. A new Agent Policy setting determines, for each type of action, whether an agent may run it unattended, must wait for human approval, or is refused. The policy is enforced by the platform rather than by prompt instructions, and it applies to both the built-in Assistant and external MCP clients.

Credentials and execution
Agents do not handle credentials. Secrets are referenced by name and resolved by HashiCorp Vault at run time. Agent-written scripts run in a separate sidecar container with no secrets inside and no network route to the platform’s database, object store, or secrets store. Teams can alternatively host data feeds as their own services, so that no agent-written code runs on the platform.

Audit and provenance
An audit log records every actor, including on denied requests: humans by login, agents by key identity, and the Assistant by the person it acted for. Generated scripts are versioned in the team’s GitHub repository, runs record the exact approved version, and the log can be mirrored to an existing SIEM.

When enabled for a pipeline, provenance stamps each landed row, document, and vector chunk with the run, script commit, pipeline version, and source that produced it. A single call resolves a row to its origin. Vector search results carry the same stamp, so retrieval-augmented answers can cite their sources. A lineage view traces source to data feed to pipeline to dataset, and agents can search for datasets by meaning.

Automated recovery
When a data feed fails repeatedly or stops producing data, a load errors, or volume changes unexpectedly, the platform opens an incident, diagnoses it, and either proposes or applies a narrow repair. A fix that does not produce a clean run is reverted. The recovery agent is disabled by default.

Destinations and models
Snowflake and Databricks join PostgreSQL, MongoDB, S3 Parquet/ORC, message queues, and five vector databases as supported destinations. Model providers include Anthropic, OpenAI, Azure OpenAI, Amazon Bedrock, xAI, and local models through Ollama, with bundled embeddings for air-gapped environments.

Availability
The release is available now at https://github.com/datris/datris-platform-oss under AGPL-3.0. It runs on-premises, in any cloud, or on a laptop using Docker. Datris does not operate a managed or hosted service, and the platform collects no telemetry. Commercial licenses with support terms are also available; the commercial and open-source builds are identical. Documentation: https://docs.datris.ai. Licensing: https://datris.ai/licensing.

About Datris
Datris develops an open-source data control plane for AI agents. The self-hosted platform acquires, validates, and lands data in the stores a team already runs and serves it to agents over MCP with row-level provenance. More information: https://datris.ai.

Todd Fearn
Datris.ai
+1 973-714-6347
email us here

Datris Platform Overview – The data control plan for AI agents

Media gallery