Solidatus 2026.3 puts data lineage where AI agents can reach it

MCP support, Bring Your Own LLM, and an assistant that keeps its context

By Danny Waddington, Chief Technology Officer, Solidatus

At the Gartner Data & Analytics Summit in London this May, Director Analyst Anurag Raj told the audience that “data governance will be the single point of failure for organizations’ AI ambitions.” He set out the work ahead in three parts, namely governance of AI, governance by AI, and governance for AI.¹

Those three parts describe where we have been building. Solidatus 2026.3 puts something into each of them, and one requirement runs through all three. AI-assisted lineage work happens across days and across people, so a platform that forgets what happened yesterday makes the person do the remembering.

Governance by AI: sessions that hold their context

The AI Assistant now keeps its conversations. History persists across browser sessions, organized by recency, and you can search, rename, or delete past chats from the assistant panel. Administrators can switch history off where policy requires it.

Long investigations used to end at the context limit. Chat Rollup now carries your active state and filters into a fresh conversation when a session fills up, along with the traces and key entities you were working on, so a half-finished impact analysis survives into the next chat rather than being rebuilt from memory.

The Prompt Library stores reusable prompts, tagged and scoped either to a single data lineage model or to every session, which turns one steward’s phrasing for a recurring regulatory question into something the whole team can run. Quick Actions propose the next step from the current session, and you can queue a follow-up request while the assistant is still answering.

Governance of AI: your model, your keys, your record

Bring Your Own LLM (BYOLLM) lets you run the AI Assistant against your own model provider and keys. For a team with data sovereignty requirements, an approved-model list, or a security review that has already cleared one provider, that removes a question which used to stall AI adoption before it started. Misconfigured LLM settings now return a specific error rather than a generic failure.

Underneath, the governance model is unchanged. Every suggestion the assistant makes is staged for a person to approve, and every accepted change becomes an attributed, timestamped revision in the data lineage model’s history. When someone asks which AI-proposed mapping reached production and who approved it, the version history holds the answer. That record is what separates AI governance questions a data catalog cannot answer from ones your lineage can.

Governance for AI: grounded context for other tools

Solidatus now exposes a Model Context Protocol (MCP) server. External AI tools and agents can read and write data lineage model content through it, and both basic and extended MCP capabilities are supported.

Most AI assistants deployed inside a bank answer from whatever context they happen to reach, which is how AI failures start in financial services. Connecting one to a governed data lineage model means its answer traces back to something a person modeled and approved, with the evidence attached. Access is enabled per customer, so speak to your Account Manager.

The map still has to be readable

Faster answers raise the speed at which people expect to navigate the enterprise data they are asking about. Several changes in 2026.3 go to that.

Data Domains now include a Domain Spine that stays visible as you move through a domain, keeping Home, the AI Assistant, the Reference Hub, Lineage Models, and saved Data Maps one click away. Lineage models have a proper home inside the domain, with a dedicated page listing everything connected. Entity and sub-entity pages carry a searchable, paginated hierarchy table, so nested structure is navigable without leaving the domain.

Data Maps now save without reloading the page, and display rules persist on saved views. Analytics reports can be edited and deleted straight from the domain home page, and you can export the records behind any metric to CSV from the drill-down panel, which is the difference between reporting a governance number and handing someone the evidence that produced it. Continued modernization work also improves how large models load, navigate, and render.

Where to start

When you upgrade, take the regulatory question your team currently answers by hand and make it a Prompt Library entry scoped to the model it applies to, so the next person who needs that trace runs it in one step instead of reconstructing it. Point BYOLLM at whichever provider your security team has already cleared, ahead of your next model risk review rather than during it. Then pull the CSV behind one governance metric you report upward, so the number arrives with the records that produced it.

The full 2026.3 release notes cover everything in this release, including the usability work and platform fixes not described here. For on-premises upgrades, check the Ops Release Notes before planning your rollout, and for questions about your environment, BYOLLM, or MCP access, contact your Account Manager or Customer Success Manager.

¹Gartner. “Gartner Data & Analytics Summit 2026 London: Day 2 Highlights.” Gartner Newsroom, May 12, 2026.
https://www.gartner.com/en/newsroom/press-releases/2026-05-12-gartner-data-and-analytics-summit-london-2026-day-2-highlights