Dotmatics introduces Luma Agent: the AI co-scientist built on structured scientific data - AiThority

Agentic AI that can plan and execute complex scientific work using natural language, configure the platform itself, and deliver traceable and reproducible answers that will pass the governance test

Dotmatics announced Luma Agent, a new agentic AI capability embedded in the Luma Scientific Intelligence Platform. Luma Agent is an AI co-scientist that plans and executes complex scientific work, going beyond answering questions to completing tasks end-to-end, including analyzing data, generating reports, managing workflows, and configuring the platform. Unlike general-purpose AI tools that operate on fragmented, unstructured data, Luma Agent is built on a foundation of structured, ontology-backed scientific data captured at the point of work. That foundation makes every answer verifiable, every action traceable, and every result trustworthy enough to act on.

Luma Agent shifts what AI does inside the lab—from analyzing to acting. Scientists describe a goal in natural language. They can explore data, manage workflows, execute calculations, or hand off work to the wet lab. They can also configure the platform themselves without requiring a service engagement, including setting up data models and creating workflows through conversation. Luma Agent builds a plan, executes it across multiple systems and steps, and returns a complete result. Work that previously took days of manual querying and coordination can now be completed in minutes. To help ensure safety and accountability, every action is logged with full audit trails, requires human approval before any data changes, and can be verified and reproduced.

“Scientists want more than data—they want insights, answers they can act on, trust, and trace back to the work that produced them. Luma Agent is built on that principle,” said Kalim Saliba, chief product officer at Dotmatics. “Because data is structured during the initial scientific work, every answer can be traced back to the queries and source data that produced it. That traceability is what gives scientists confidence in the result. We’ve designed Luma Agent to be able to function as a node in any AI workflow, so any external agent can call it with the full scientific context that no general-purpose tool can replicate.”

Also Read: AiThority Interview with Glenn Jocher, Founder & CEO, Ultralytics

Luma Agent is a part of Luma, the AI-native Scientific Intelligence Platform that lets scientists query their data in plain language—no SQL or tool-switching. Luma and its scientific applications support multimodal drug development, team collaboration, and connected data across research and production. It’s built to be open, integrating with third-party systems so data moves wherever teams need it.

“Databricks powers Luma’s ability to move from raw data to actionable intelligence at scale,” said Michael Sanky, VP, Healthcare & Life Sciences GTM at Databricks. “With Luma Agent, life sciences teams can now leverage that structured foundation to not just analyze, but to act, turning insights into experiments in minutes rather than days. Luma Agent changes the equation by making complex data pipelines conversational and self-configuring. It’s an important step toward AI that understands domain context, not just data volume.”

Designed to Pass the Governance Gate That Will Stop 80 Percent of AI Agents
Giving AI the ability to act raises the stakes. When AI is wrong about an answer, you ask again. When AI takes the wrong action on production scientific data, that is a different problem entirely. Luma Agent is engineered with this distinction at its core. Every step the agent takes is logged, with full tool execution traces capturing exactly what the agent did, with what inputs, and what it returned. The scientist remains in control of every decision while the agent handles the work in between.

Gartner predicts that 80 percent of agentic AI initiatives in healthcare and life sciences will not progress beyond initial governance checkpoints in 2026¹, not because the underlying models are insufficient, but because most platforms cannot demonstrate the level of traceability and explainability that regulated environments demand. Luma Agent’s architecture is designed specifically to pass that gate.

The foundation that makes this possible is Luma’s structured, ontology-backed data, captured at the point of scientific work rather than reconstructed after the fact. This is the representation layer most vendors overlook: AI can only be as traceable and verifiable as the data it operates on.

The Agent That Builds the Lab
Most AI tools in life sciences are analytical and answer questions about data that already exists. Luma Agent goes further. Scientists and administrators can describe a schema or a data flow in plain language and the agent configures it, writing the underlying SQL, setting up task types, and adding metadata automatically. What previously required a specialist services engagement now happens in a single conversation.

For digital lab and informatics teams, this changes the economics of deployment. Onboarding is faster, configuration expertise is no longer a bottleneck, and teams can iterate on the platform without opening a support ticket. The same agent that helps a scientist analyze a phage display campaign can help an administrator set up the next study’s data model before the day is out.

Open by Design: A Co-Scientist Any Agent Can Call On
Luma is built to work with the AI tools each team already uses. Developers can connect external language models (e.g. Claude, ChatGPT, or proprietary systems) directly to Luma Agent via Model Context Protocol (MCP) to query experimental results, retrieve protocols, or run calculations without rebuilding scientific context. But the connection works both ways: external AI tools don’t just read from Luma; they can also configure it. Scientists and admins can build schemas, set up data flows, and customize workflows by simply asking any connected AI assistant, which then executes those changes directly in Luma.

“While the industry races to connect more data sources, we’ve focused on making the data at the center worth connecting to,” Saliba said. “We’re building the open ecosystem life sciences teams choose to build on, where their agents, their data, and our scientific capabilities work as one. We’re enabling external agents to call on Luma as a co-scientist within their own workflows, publishing developer resources, and deepening integrations with the broader enterprise AI landscape so Luma works wherever teams need it.”