Skip to main content

Pulse — Data Intelligence

An agent that does real work generates and consumes a lot of data. It reads history to make a decision, reacts to things happening right now, and produces a running record of what it did that someone — a person watching a dashboard, another agent, an auditor months later — needs to see. Building that yourself usually means assembling four or five separate services: a warehouse for the history, a message bus for the events, a websocket or pub/sub layer for the live updates, an object store holding it all together. Each is a separate vendor, a separate bill, a separate security model, and a separate place for data to cross a boundary you didn't mean to open.

Pulse is that whole data plane as one governed thing. Your analytical history and your live event traffic sit under the same roof, addressed the same way, and everything inside it inherits the platform's guarantees: your data stays in the region you pinned, under the custody model you chose, whether it's a petabyte of history in cold storage or a single event on its way to a browser tab.

Pulse is built as four modules, in two pairs — the data you analyze, and the data in motion. In each pair, one module is where the data lives and the other is what you do with it, and the two are priced and scaled separately, because a workload rarely stresses both at once.

The data you analyzeLakehousewhere data livesevent history,model outputsComputeSQL over itGoverned Databasejoined in one queryThe data in motionStreamevent buspublish, subscribe,replayFeedlive deliveryDashboard or appin the browserKibo custody and the region pin apply across all of it
Pulse pairs where data lives with what you do with it: Lakehouse and Compute for analysis, Stream and Feed for data in motion.

The data you analyze​

Lakehouse is where your analytical data lives — event history, model outputs, exported records, anything you want to keep and query later. It's cheap columnar storage with a declared shape, so it stays a queryable table rather than a folder of forgotten files.

Compute is how you ask questions of it. The same SQL you already run against your operational database reaches your Lakehouse data too — and, in a single query, joins the two. You can ask "which of the customers who churned last quarter opened a ticket this week" without moving data between systems first.

The data in motion​

Stream is an event bus your project owns: publish events, subscribe to them, replay history. It's how the parts of your application — and your agents — tell each other what just happened.

Feed is how those events reach a screen. When a dashboard or an app holds a connection open to watch something update live, that's Feed doing the last-mile delivery to the browser.

How Pulse fits the rest of the platform​

Pulse doesn't stand apart from Kilimanjaro — it composes with it:

  • Your live operational data and your analytical history answer to one query. A Lakehouse dataset and a Governed Database collection can be joined in the same SQL statement, with row-level security still enforced on the operational side.
  • Kibo holds the keys. If a dataset lives in your own cloud storage account, or an event stream needs a credential, that custody runs through Kibo like every other secret on the platform.
  • Region pin holds everywhere. A dataset, a stream, and the collections they join against all stay inside the jurisdiction you chose for the project.

One thing Pulse is not: it doesn't run your agent's own code. Executing an agent's reasoning loop in a sealed sandbox is a Kilimanjaro Core capability, built on Firelock — the open-source execution shell Kilimanjaro maintains — and documented with Firelock, not here. Pulse is the data your agent works with, not the place it runs.

Start here​

  • Lakehouse — store analytical data as governed, queryable datasets.
  • Compute — query your data, and join analytical and operational data in one statement.
  • Stream — publish and subscribe to your own real-time event streams.
  • Feed — deliver live updates to dashboards and apps.