Introducing Pulse-Agent: A Data Agent That Plugs Into the Tools You Already Use

Today we're shipping Pulse-Agent v0.1.0 — a general-purpose data context agent that runs on your machine and plugs into the tools you already use.

If you spend your days bouncing between Airflow, Snowflake, dbt, Tableau, and a dozen Jupyter notebooks, Pulse-Agent is built for you. It is not a new orchestrator, not a new warehouse, not a new BI tool. It's an agent that understands the tools you already have and helps you operate them.

And it's free.

The problem we kept hitting

Modern data teams don't suffer from a shortage of tools. They suffer from the opposite — every workflow crosses five of them. A pipeline broke? That's Airflow logs, then Snowflake query history, then a dbt manifest, then a Tableau dashboard that someone said looked off. A new metric request? That's a catalog lookup, three SQL drafts, a dbt model change, and a downstream impact check.

The work isn't hard. The work is fragmented. And every existing "AI for data" tool we tried wanted to own one of those tools end-to-end — replace your warehouse, replace your orchestrator, replace your notebook. None of them met us where we actually work.

So we built the opposite. An agent with no opinions about your stack, that learns your stack, and acts on it.

What Pulse-Agent does

Pulse-Agent runs locally as a desktop app (Tauri) or CLI. You point it at the tools in your environment, and it builds a working context: what pipelines you have, what tables exist, what models depend on what, what dashboards consume which sources.

Then you talk to it like you'd talk to a teammate.

• "Why did the orders pipeline fail at 3am?" — it pulls the Airflow logs, finds the failed task, checks the upstream table, and tells you the schema drifted on a source column.
• "What breaks if I drop customers.legacy_id?" — it walks the lineage through dbt, into the warehouse, into Tableau, and gives you the blast radius before you touch anything.
• "Draft a dbt model for monthly active users from the events table." — it reads the events schema, follows your existing project conventions, and writes the model.

It's designed around three roles — Data Engineer, Data Analyst, Data Scientist — with 34 built-in skills, 10 role-based and 24 industry-specific.

Works with what you have

This is the part that matters. Pulse-Agent's first-class integrations include:

• Orchestrators: Airflow, Prefect, Dagster, Kestra, F-Pulse
• Warehouses & engines: Snowflake, BigQuery, Redshift, Databricks, Postgres, DuckDB
• Transforms: dbt, SQLMesh
• Catalogs & lineage: DataHub, Marquez, OpenLineage, D-Pulse
• BI: Tableau, Looker, Superset, Metabase, Power BI
• Notebooks & IDEs: Jupyter, VS Code, Cursor

F-Pulse and D-Pulse are in that list as first-party integrations, not as a requirement. If you never touch a Hybridyn product, Pulse-Agent still works. That's the whole point.

Plan → Act → Verify

Agents that just do things are dangerous near production data. Pulse-Agent uses a Plan → Act → Verify loop for every non-trivial task:

1. Plan. It writes out what it intends to do — which tools it will call, which queries it will run, which files it will touch. You see the plan before anything happens.
2. Act. It executes the plan step by step. Read-only steps run automatically. Write steps stop and ask, unless you've explicitly authorized them.
3. Verify. After acting, it checks the result against the plan. Did the query return rows? Did the dbt run pass tests? Did the file actually get written? It tells you what changed and what didn't.

This is the difference between an agent you can leave running and an agent that needs a babysitter.

Ten hard guardrails

On top of Plan → Act → Verify, Pulse-Agent ships with ten guardrails that are not optional:

• No destructive SQL (DROP, TRUNCATE, unscoped DELETE) without explicit per-call approval.
• No writes to production warehouses unless the connection is tagged as writable.
• No credential exfiltration — secrets never enter the LLM context.
• No silent file edits — every file write is logged and diffable.
• No long-running queries without a cost estimate first.
• No schema changes without an impact preview.
• No cross-environment moves (dev → prod) without confirmation.
• No external API calls outside the configured allowlist.
• No retention of query results beyond the session unless you ask.
• No background actions in ambient mode without your explicit opt-in.

These are enforced at the agent runtime, not by the prompt. They survive jailbreaks because they live below the model.

Local-first, by default

Pulse-Agent runs eight LLM providers out of the box: Claude, OpenAI, Gemini, Azure OpenAI, Ollama, DeepSeek, Mistral, and Groq.

The default is Ollama. No API key. No data leaving your laptop. No vendor lock-in. If your security team won't let you send schema or query history to a third-party API, you can run the entire agent against a local model and it works the same way.

If you do want frontier model quality, swap the provider in one line of config and bring your own key. Pulse-Agent doesn't have a hosted backend, doesn't proxy your requests, and doesn't see your data. Ever.

Ambient mode

Ambient mode is opt-in. When it's on, Pulse-Agent runs as a background watcher — it sees pipeline failures as they happen, notices when a table's row count drops by 90%, flags when a dbt run starts taking 4× longer than yesterday. It surfaces these as quiet notifications, not alarms.

You can ignore them. You can also click them and have the agent already halfway through a root-cause analysis by the time you read the message.

How to get it

Pulse-Agent v0.1.0 is available now. The desktop app is built with Tauri and ships for macOS, Windows, and Linux. The CLI is a single binary.

Both are free. There is no paid tier of Pulse-Agent. There is no telemetry-on-by-default. There is no "pro features" upsell hiding behind a sign-up wall.

If you want to see how it fits into your stack, head to the Pulse-Agent product page for installation instructions and the integration list.

Why we built this

Hybridyn already has F-Pulse (the free pipeline engine) and D-Pulse (the enterprise data operating system). We could have built an agent that only worked with our own products. That's the obvious play and it would have been easier.

We didn't, because the bet we're making is that the next decade of data tooling isn't about replacing your stack. It's about giving your stack a brain that works across all of it. Pulse-Agent is that brain. It happens to know F-Pulse and D-Pulse really well — but it knows Airflow and Snowflake just as well, and that's how it should be.

If you try it, tell us what breaks. We'll fix it.