Release notes: June 10 to 26, 2026

Back to the changelog

A two-week round of improvements: broader and more accurate mainframe coverage, a smoother and more resilient chat experience, faster searches, and a clearer view of long-running analyses. The headline change is a rename that better reflects what the product is about, plus two capabilities that make the agent noticeably more useful day to day: a memory of its own, and the ability to read your documents alongside your code.

Highlights

• "Knowledge Graph" is now "Systems." We renamed the central concept to reflect what it really represents: a structured model of one of your systems, drawn from its code, JCL, copybooks, and documentation.
• Persistent agent memory. The chat agent now remembers what you have told it (your role, your preferences, project context) from one conversation to the next.
• Multi-repository systems. A single system can now be built from several source repositories, with per-repository scoping rules.
• Documentation in the conversation. The agent can now find and quote from your ingested documents, and surfaces a single, deduplicated glossary alongside its results.
• Streaming that survives a hiccup. Chat answers no longer get lost if you switch tabs, move between the web app and the VS Code extension mid-stream, briefly lose the network, or we deploy a new version.
• Self-serve single sign-on. Tenant administrators can set up SSO themselves from the Account screen.

Nomain app and agent

"Knowledge Graph" is now "Systems"

Throughout the app, what used to be called a "Knowledge Graph" is now a System. The name reflects what the object actually is: a structured, queryable model of one of your software systems, assembled from its code, job control, copybooks, and documentation. Everything you built before is unchanged; only the label is different.

Persistent agent memory

The chat agent can now keep durable notes about you and your work, and use them in later conversations. See Agent memory.

• Tell it your role, your preferences, or context about a project, and it carries that forward, so you do not have to repeat yourself in every new chat.
• Memory is written by the agent, not by you. If you want it to remember something specific, say so in chat with a cue like "My preference is...", "Remember that...", or "Memorize this...".
• Everything it remembers is visible and editable. The full Memory screen for reviewing and deleting notes arrives in the following release.

Multi-repository systems

A single System can now draw from more than one source repository. See Repositories and sources.

• Add several repositories to the same system, for cases where one application spans multiple code repositories.
• Set per-repository scoping rules, so you can include only the relevant folders or file types from each one.

Documentation in the conversation

The agent now treats your ingested documents as first-class sources, right alongside the code.

• Ask a question and the agent can find and quote from your Markdown, PDF, and Word documents, not just the source code.
• Results come with a single, deduplicated glossary of the terms involved, rather than the same definition repeated several times.

Streaming that survives a hiccup

Chat answers used to be fragile while they were still being generated. That is fixed. See Chat.

• Switch to another browser tab and back, and the answer is still streaming where you left it.
• Start a question in the web app and open the same chat in the VS Code extension (or the other way around), and it reconnects to the live stream.
• A brief network drop, or a new version of Nomain being deployed mid-answer, no longer loses your answer. The stream picks back up.

Business process, as a diagram

Business processes can now be viewed as a diagram, not just a list, making the flow between steps easier to follow at a glance. See Domains, areas and processes.

Usage insights

The Usage insights page gives administrators a clear view of how the platform is being used across the tenant: chats, analyses, and activity over time, with day, week, and month ranges.

Faster search

Searches across your systems return results noticeably faster, particularly on large systems with a lot of history.

Administration

Self-serve single sign-on

Tenant administrators can now connect their own identity provider without going through Nomain support. See Single sign-on.

• Start the setup from the Account screen, follow the guided steps, and connect your IdP (Okta, Entra ID, Google Workspace, or any SAML or OIDC provider) yourself.
• Once connected, your users sign in through your own identity provider, with your policies (MFA, conditional access) applied.

Analysis and system management

Broader, more accurate mainframe coverage

This release widens and sharpens the reader for mainframe languages.

• Better PL/I coverage, including constructs that were previously skipped.
• More accurate COBOL and JCL parsing, so the resulting system is a closer match to what the code actually does.

A clearer view of long-running analyses

Job history for an analysis is easier to follow, and easier to recover when a single step goes wrong.

• Grouped job history. Steps are grouped so you can see the shape of a long analysis at a glance, rather than scrolling a flat list.
• Per-step retries. When one step of an analysis fails, you can retry just that step, instead of re-running the whole analysis from the start.