AI agents for data teams — on the stack you already run
AI agents that build,
test and operate
your data pipelines.
You approve
every change.
NerveStax connects to your warehouse, transformation project, orchestrator and git. Its agents load data, change models, look after schedules and investigate pipeline alerts. Each change is tested in a sandbox and reaches you as a pull request.
Private beta · hosted or self-hosted · Meet the on-call agent ›
Works with the tools you already run
Outcomes
What changes
for your team
You can see how it fits together
Models, tests, schedules and runs in one record, with lineage, freshness and drift.
Changes ship with less guesswork
Each change shows what it affects downstream and arrives already tested.
Fewer pages at night
Every pipeline alert is investigated first. Noise is closed with a reason; real incidents page you in PagerDuty, Opsgenie or Splunk On-Call with the likely cause.
How it works
One set of agents
across the lifecycle
The agents share one record of your platform, so each job starts from what is already known about it.
-
01 · IngestBeta
Load source databases
Point an agent at a MySQL or PostgreSQL database. It picks tables and sync columns, then proposes the load as a pull request. Schema changes are detected.
-
02 · Model & testLive
Build and change models
Agents write new models or edit existing ones, run them against a development target, and list the downstream models and schedules each change touches.
-
03 · Schedule & runLive
Keep production runs healthy
Agents find why a run failed, check models against your schedules, and can test a change through a verification DAG in your own Airflow before you review it.
-
04 · On-callLive
Investigate every alert
Alerts from Airflow, dbt or any webhook are checked with read-only access. Real incidents go to PagerDuty, Opsgenie or Splunk On-Call. If triage is unsure or fails, the alert is escalated anyway. More on on-call ›
How changes ship
Every change is a
reviewed pull request
Agents do not write to production. Your CI and review rules still apply.
-
Request
A question from your team, or a failed run.
-
Sandbox
Built in an isolated copy, against a development target.
-
Tests
Your tests run before anyone reviews it.
-
Pull request
Opened in your repository, with results and reasoning.
-
You approve
A person on your team merges it, or sends it back.
Why NerveStax
Built around
your stack
Your stack stays
It works with the warehouse, orchestrator and GitHub or GitLab you already run. Nothing to migrate.
No lock-in
Everything it produces is ordinary code in your repository. Your pipelines run without it.
One shared record
What was built, tested and run, shared across agents and sessions, so nobody explains the platform twice.
Security
Safe by design
Isolated sandboxes
Each conversation runs in its own sandbox, cleaned up when idle.
Encrypted credentials
AES-256-GCM at rest, and redacted from logs and from everything the model reads.
Your git is the record
Work lands as branches and pull requests in your repository.
Approval before changes land
A person approves before a pull request opens or anything runs in your orchestrator.
Your model keys
Bring your own provider key; NerveStax has no fallback key.
Hosted or self-hosted
Use our cloud, or run it on Kubernetes in yours.
Questions
Common questions
Q01What is NerveStax?−
Agentic data engineering: a set of AI agents for data teams. They work on your existing warehouse, transformation project, orchestrator and git, and every change they make reaches you as a tested pull request.
Q02Which tools do you support?+
dbt for transformation and Airflow for orchestration. Snowflake, Trino, DuckDB/MotherDuck, Postgres and ClickHouse as warehouses. GitHub and GitLab, including self-hosted. On-call escalation to PagerDuty, Opsgenie, Splunk On-Call (VictorOps) or any paging API. Ingestion sources: MySQL and PostgreSQL (beta). Running something else? Tell us below.
Q03Does it write to production?+
Agents work in a sandbox and deliver changes as pull requests, which they can’t merge. A person approves before an agent opens a pull request, triggers Airflow runs, runs a backfill or runs an ingestion. dbt runs use your project’s configured connection; mark a project as protected to require approval for those too.
Q04Does my code or data leave my infrastructure?+
Hosted NerveStax runs on Google Cloud, with code sandboxes on E2B. Model calls go to the provider you choose, under your key, and include schemas, code, logs and, when an agent queries or samples a table, the rows it retrieves. To keep everything in your network, self-host NerveStax and point it at a self-hosted OpenAI-compatible model.
Q05Do I have to restructure my project?+
No. NerveStax reads your existing project as it is.
Q06What is live, and what is in beta?+
Modeling and testing, Airflow operations and schedules, and on-call triage are live. Ingestion from MySQL and PostgreSQL is in beta. Column-level lineage is not built yet. Beta teams help decide what comes next.
Q07What does it cost?+
The private beta is free. Beta teams get founding-customer pricing at launch.
Early access
Tell us about
your data platform.
Beta workspaces open in small batches. Tell us what you run and where the time goes. We read every message and reply within a working day.