Private beta — onboarding data teams now

Releases

What we have shipped.

The releases that changed what you can do with NerveStax, newest first. Everything here is live for beta workspaces unless it is marked early access.

  1. Operations

    LatestOperations

    Pipelines: Airflow 3, and a view of your cluster you can rely on

    The Pipelines screens now work with Apache Airflow 3 as well as Airflow 2, and say plainly what they know. Freshness is judged by one rule everywhere, and a scheduled run that never fired is reported.

    • Airflow 3.0 and 3.1 are supported alongside Airflow 2, for reading runs and for the actions you approve.
    • Task logs open in the product, showing the end of the log, where the failure is.
    • One freshness rule across data assets, data products, ingestions and the agent. A model with no schedule says so.
    • Missed scheduled runs are shown on the Schedules and Runs pages and raise one alert per outage.
    • The stored cluster credential is sent only to the address it was saved for.

    How the platform operates your pipelines →

  2. PR tests

    PR tests

    PR tests that cannot touch production

    Pull request tests can now run under a separate warehouse login that cannot write production. Risky changes are refused or held for a second reviewer, and test tables are cleaned up when the pull request closes.

    • A separate test login per project, checked every day to confirm it cannot write production.
    • Risky changes are refused, or held until someone who did not write the code approves: files outside the project, symlinks, credentials typed into code and snapshots. Pull requests from forks are off by default.
    • Production is compared before and after every test run.
    • Clean-up is automatic: test tables are removed within minutes of a pull request closing, and a daily check reports anything left behind.
    • A commit status on your git server shows the result on the pull request. A failed test says what failed: models, tests or the run.
    • This release needs the current version of the NerveStax component in your Airflow repository.

    Tested before merge, with a film →

  3. Data apps

    Data apps

    Data apps, and conversational analytics beside every page

    A data app is a page built for one decision, running on your certified data products. Conversational analytics now opens beside dashboards, data apps and data assets, already knowing what you are looking at.

    • Upload an app that names the data products it uses. It gets numbers only through NerveStax, as the viewer, within a budget, and every query is audited.
    • Access per app, versions you can roll back, full-screen mode and share links.
    • Five sample apps to start from, including a weekly one-page story and a slice-and-dice explorer.
    • Ask about this page: a side panel that knows the dashboard’s filters, the selected block or the app’s own facts.
    • Permanent addresses for every workspace page, with breadcrumbs.

    See the product →

  4. Git

    Git

    Git access: your own account, on GitHub or GitLab

    Each person connects their own git account, and the agent’s commits and pushes are made as that person. Organisation-level access is read-only and used only for background syncing.

    • Personal credentials per git server, organisation credentials for syncing, and GitHub App installs.
    • GitHub and GitLab, cloud and self-hosted, including GitHub Enterprise Server and nested GitLab groups.
    • One health status per credential, with when it was last checked.
    • Any other HTTPS git server can clone and push a branch; pull requests need GitHub or GitLab.

    How credentials are handled →

  5. Dashboards

    Dashboards

    Dashboards on the semantic layer

    Dashboards are built over your data products, so they show the same numbers as every other route, and every query runs as the person viewing. They sit beside the BI tools you already run.

    • An editor with drag and resize, undo and redo, and a preview before you publish.
    • Blocks: KPI with comparison and target, chart, table, text, editable inputs, and a written summary that uses only numbers on the page.
    • Filters and interaction: date ranges, dependent filters, click to filter, drill-down, saved and shared views.
    • Export to CSV or Excel from any block.
    • Embedded dashboards for your own customers, with row rules applied per end user.
    • Drafted by the agent on request; publishing waits for your approval.

    The semantic layer behind it →

  6. Security

    Security

    Roles, tokens and an agent that acts as you

    Access is now a set of roles a person holds, checked on every screen, every API call and every agent tool. The agent offers only what the person’s roles allow, and approving an action needs the same permission as doing it.

    • Roles: owner, admin, engineer, analyst and consumer. A person can hold several.
    • Analysts browse and query data products, use conversational analytics and build dashboards, without the engineering pages.
    • Tokens and integrations on one page: personal tokens, service tokens, embed keys and AI-assistant clients, each with a purpose, a scope and an expiry.
    • NerveStax staff are read-only in your organisation by default. Any change needs a written reason for one hour of access, recorded in your audit log.

    Security in detail →

  7. Analytics

    Analytics

    Conversational analytics, with ChatGPT and Gemini connectors

    Business users ask questions in plain language and get answers built only from certified data products. Every number comes from a query and is shown with its source and freshness. The AI model never writes a number.

    • One cited card per query, as a chart or a table, with the filters in words.
    • Charts chosen by rules from the shape of the data, including bar, line, funnel, waterfall and heatmap. Switch to a table or download CSV.
    • Pins and a board for the answers you want to keep.
    • ChatGPT and Gemini join Claude as assistants that can query your data products, each acting as the member who consented.
    • Test questions on each data product run after every change and every night; a failing one marks the product stale.

    How answers stay trustworthy →

  8. Semantic layer

    Semantic layerEarly access

    Data Products: one governed definition of every metric

    Define a metric once, in your dbt repository, and serve the same answer to dashboards, assistants and applications. A data product is a small set of files next to your models. NerveStax checks it against your models and tests before it is published, and applies your access rules to every query that reaches it.

    • BI tools connect over a standard PostgreSQL connection, encrypted in transit.
    • Claude can answer questions from your governed metrics through a connector your organisation approves, and you can switch it off at any time.
    • Access policies, query budgets and the audit log apply to every consumer, whether that is a dashboard, an assistant or the API.
    • Authoring in chat: the agent drafts a product from a model, checks it and previews the result before you publish. Runs on DuckDB and MotherDuck, ClickHouse and PostgreSQL.

    Ask for early access →

  9. Ingestion

    Ingestion

    SAP joins MySQL and PostgreSQL as an ingestion source

    Agents can now load SAP data over OData, the interface SAP systems publish for reporting and extraction. Behind it is a new way of adding sources: each one is a single, described module, so the next source a customer needs is a short piece of work rather than a project.

    • Four SAP authentication modes, including SAP Business One.
    • Every request stays on the SAP host you configured.
    • The same review card and pull request as every other load, so nothing installs without a person.

    The SAP integration →

  10. Security

    Security

    Single sign-on and an organisation audit log

    Connect your own identity provider and decide who signs in how. Any provider that supports OpenID Connect works, including Okta, Microsoft Entra ID, Google Workspace and Auth0. Alongside it, a new audit log records the actions people take in your organisation, with a retention policy your owners control.

    • Email-first sign-in: the email domain decides between password and single sign-on.
    • New members are created on first sign-in from your allowed domains, with the default role you choose.
    • Since 9 October the audit log covers 93 actions, from roles and secrets to schedules, backfills and ingestions.
    • Retention you set, a purge hold, a preview of what a shorter retention would delete, and exports.

    How we handle security →

  11. Ingestion

    Ingestion

    Set up an ingestion by asking for it

    Ask the agent to bring a MySQL or PostgreSQL database into the warehouse and it walks the setup with you in the conversation. It explores the source, proposes tables, keys and sync columns, and delivers the load as a pull request.

    • Credentials go through a secure form straight to encrypted storage. The model never sees them.
    • A review card lists every table with its load mode and any warnings before anything is installed.
    • A new Ingestions page shows the state, runs and history of every load, with Run now.
    • Source schema changes are detected and classified, so a new column reaches you as a decision, not a broken load.

    The MySQL integration →

  12. Scheduling

    Scheduling

    Schedules your agents understand

    Put a model on a cadence with one tag in its config. Your workspace defines the cadences it allows, each one runs in your own Airflow, and every change to a schedule arrives as a pull request.

    • A Schedules page in workspace settings to manage the cadences.
    • Each model’s cadence is shown on Data Assets, with a filter for models that are not scheduled.
    • The agent can read and check schedules and explain why a run fired. It never creates or retimes one on its own.
    • A pull request with a schedule tag that does not exist is flagged before merge.

    The Airflow integration →

  13. On-call

    On-call

    An on-call engineer for your data pipelines

    Point Airflow, dbt or any webhook at NerveStax and every alert is investigated before anyone is paged. The agent reads the logs and recent changes with read-only access, closes the noise with a written reason, and escalates real incidents with the likely cause attached.

    • Duplicate alerts are grouped, so a storm becomes one investigation.
    • Escalations are saved first, then delivered to PagerDuty, Opsgenie or Splunk On-Call.
    • If triage is unsure or fails, the alert is escalated anyway.

    Meet the on-call agent →

  14. Git

    Git

    GitLab, including self-managed

    NerveStax now works with GitLab as well as GitHub. Merge requests, checks and the PR Tests page behave the same way on both, and self-managed GitLab and GitHub Enterprise Server are supported.

    • Choose the git provider per project.
    • Agent changes arrive as merge requests your team approves, exactly as they do on GitHub.
    • Commits are authored by the person who asked for the change.

    The GitLab integration →

  15. Operations

    Operations

    Production backfills, planned with you

    Ask the agent to rebuild a model for a date range and it plans the backfill across everything downstream, asks the questions that matter, and waits for approval. Your Airflow runs the work, and a new Backfills page shows each run as it progresses.

    • The full downstream set of models is included, so nothing is left half-rebuilt.
    • Date ranges are split into chunks you can follow one by one.
    • Nothing runs without a person approving it.
  16. Deployment

    Deployment

    Self-hosted on Kubernetes, with sandboxes in your cluster

    NerveStax can now run inside your own infrastructure. A Helm chart installs the application on Kubernetes, and agent sandboxes can run as pods in your own cluster as well as on E2B’s hosted sandboxes, so code and data can stay in your network.

    • A Helm chart with production images. You bring your own PostgreSQL.
    • Choose the sandbox provider per organisation: E2B or Kubernetes.
    • Pair it with a self-hosted model for a deployment where nothing leaves your network.

    Security and deployment →

  17. Testing

    Testing

    Every pull request tested in your own Airflow

    When an agent opens a dbt pull request, your Airflow now builds and tests it before anyone reviews it, in an isolated test schema that never touches production tables. The result comes back to the pull request and to the conversation.

    • The test harness is installed into your Airflow repository through a pull request you approve.
    • Every test run is isolated from production.
    • A PR Tests page lists open pull requests and their results.

    The Airflow integration →

Private beta

Need something that is not on this list yet?

Beta teams help decide what ships next. Tell us what you run and where the time goes, and we will tell you plainly whether it is close.

→ hello@nervestax.ai

Get early access →