Your data context, in every tool you use.
Pulse-Agent is a general-purpose autonomous AI agent for Data Engineers, Data Analysts, and Data Scientists. Plug it into the tools you already use — it carries your data context across them, plans, acts, and verifies. Local-first.
One agent. Every tool you already use.
Other data agents are tied to a single product. Pulse-Agent is a portable context layer — it plugs into the orchestrator, warehouse, transform engine, BI tool, and notebook you already run, and carries your data context across them.
First-party Hybridyn integrations · everything else via open connectors
Real personality differentiation.
Not three configs of the same bot — three agents with different verbs, different fears, different toolboxes.
Data Engineer
verb: buildDiagnoses pipeline failures across any orchestrator, patches broken transforms, profiles new sources, and proposes layered models.
Data Analyst
verb: answerBuilds dashboards from a sentence, writes SQL against any warehouse, and validates results before they reach a stakeholder.
Data Scientist
verb: investigateProfiles features, runs experiments in a sandboxed Docker, compares model variants, and flags leakage before training.
Plan → Act → Verify.
Every task runs the same loop. You see the plan before any action. You approve every WRITE. The agent verifies its own output before reporting done.
Plan
Decompose into steps. Classify each as READ / WRITE / DESTRUCTIVE. Show the user.
Act
Execute one step at a time. Halt at any WRITE for approval. Halt at DESTRUCTIVE for double-confirm.
Verify
Check output against expected shape. Re-run sanity checks. Surface anything unexpected.
Autonomous, not unsupervised.
Every guardrail is enforced in code, not in a system prompt. The agent literally cannot bypass them.
Four flows. Real tasks.
Fix the broken pipeline
DEOrchestrator failure → diagnose → patch → rerun
Build a churn dashboard
DAFrom one sentence to a working chart
Why is this query slow?
DAProfile → rewrite → benchmark
Onboard this CSV
DEProfile → infer schema → propose layered model
Role + Industry packs.
Skills are Markdown files with YAML frontmatter. Add your own — Pulse-Agent picks them up automatically.
Runs fully local. Zero config.
Pulse-Agent auto-detects a local Ollama on localhost:11434. No API key, no cloud, no data leaving the machine. Bring an API key for any of 8 providers when you want a stronger model.
It watches so you don't have to.
A background watcher polls your connected tools — orchestrator failures, dbt test failures, dashboard alerts, data quality drops. When something needs attention, Pulse-Agent pings you with a one-line summary and a proposed plan. Rate-limited to 3 pings/hour by default.
A day in the life with Pulse-Agent
Real workflows where the agent saves the most time, across all three data roles.
3am pipeline failure
Airflow DAG fails. Agent pulls the task log, traces the upstream Snowflake table, finds the schema drift, and posts a root cause to Slack with a one-line fix. You wake up to a draft PR.
Impact analysis before a drop
You want to drop a column. Agent walks the lineage through dbt, the warehouse, and Tableau, then lists every model, dashboard, and downstream consumer that breaks. Decide before you act.
Draft a dbt model
Ask for a monthly active users model. Agent reads your events schema, follows your existing project conventions, drafts the SQL, runs it in dev, shows you the row count, and stops for review.
Profile a new source
Point the agent at a new Postgres database. Within a minute it inventories tables, profiles row counts and null rates, flags PII candidates, and proposes a Bronze landing schema for ingest.
Explain a slow query
Paste a Snowflake query that takes 90 seconds. Agent reads the query plan, identifies the broadcast join causing the spill, suggests the rewrite, and tells you what the new estimated cost is.
Compliance answer in two minutes
Auditor asks 'where does customer_email live and who can see it.' Agent queries the catalog, walks lineage, lists every table, every consumer, and every role with access. One screen, one answer.
Common questions
Is Pulse-Agent really free?
Yes, fully free. There is no paid tier of Pulse-Agent. No telemetry-on-by-default, no usage caps, no feature gating, no signup wall. The desktop app and CLI are both free to download and run forever.
Do I need an API key?
No. Pulse-Agent ships with Ollama as the default provider — runs entirely on your machine, no API key, no data leaving your laptop. If you want to use Claude, OpenAI, or another hosted provider for higher quality, you bring your own key. Pulse-Agent has no hosted backend and never proxies your requests.
Which tools does it integrate with?
Orchestrators: Airflow, Prefect, Dagster, Kestra, F-Pulse. Warehouses: Snowflake, BigQuery, Redshift, Databricks, Postgres, DuckDB. Transforms: dbt, SQLMesh. Catalogs: DataHub, Marquez, OpenLineage, D-Pulse. BI: Tableau, Looker, Superset, Metabase, Power BI. Notebooks/IDEs: Jupyter, VS Code, Cursor.
How is it different from a chat-with-your-data tool?
Most chat-with-your-data tools are read-only. They answer questions but can't act. Pulse-Agent has a Plan → Act → Verify loop with ten guardrails — it can actually execute work (run queries, edit files, trigger pipelines) under explicit, auditable controls. It's an agent, not a chatbot.
Will it touch my production data without asking?
No. Read-only steps run automatically. Any write step (DDL, DML, file edit, pipeline trigger) stops and asks unless you've explicitly authorized that exact operation in advance. Connections are tagged read-only or writable, and writes against unmarked connections are blocked at the runtime.
What are the ten guardrails?
No destructive SQL without per-call approval; no writes to production warehouses unless tagged writable; no credentials in LLM context; no silent file edits; no expensive queries without cost preview; no schema changes without impact preview; no cross-environment moves without confirmation; no calls outside the configured allowlist; no result retention beyond the session unless requested; no background actions in ambient mode without opt-in.
Does it work without F-Pulse or D-Pulse?
Yes. F-Pulse and D-Pulse are listed as first-party integrations, not requirements. If you never use a Hybridyn product, Pulse-Agent still works against Airflow, Snowflake, dbt, Tableau, Jupyter, and the rest. The breadth is the entire point.
What is ambient mode?
An opt-in background mode where Pulse-Agent watches your data systems for failures, anomalies, and slow queries, and surfaces them as quiet notifications. You can ignore them or click in to find the agent already halfway through a root-cause analysis.
One agent. Every tool you already use.
Free. Local-first. Plugs into the data tools you already run.