Skip to content

What is Catalyst

Catalyst is one workspace for getting real work done with AI. You chat with a model, let it use tools to act on your behalf, and — when a task is worth repeating — turn that conversation into a workflow that runs the same way every time.

The idea behind it: build the work once, swap the model whenever. The model you start with today won’t be the best one next quarter. Catalyst keeps your prompts, tools, data connections, and workflows separate from any single provider, so changing models is a setting — not a rewrite.

The Catalyst chat home: a greeting, a message box with a model picker and tools, and recent sessions to pick up where you left off.

Chat with any model

Talk to Claude, GPT, Gemini, or a model you host yourself — and switch between them mid-project. The interface, your tools, and your history stay the same.

Let it take action

In agentic mode the model can search the web, run code, generate images, and call your connected tools in a loop until the task is done — not just reply with text.

Build artifacts

Code, documents, diagrams, and live HTML render in a side panel you can preview, revise, and download — versioned as the model iterates.

Automate with workflows

Wire models, code, and tools into a visual workflow that runs on demand or on a schedule, and delivers results by email.

Connect your own tools & data

Bring your own integrations over MCP and connect your databases and files, so the model works with your systems, not a generic sandbox.

It remembers you

A profile and a searchable memory let Catalyst recall what matters across sessions, so you don’t re-explain yourself every time.

Most people start in chat and grow into the rest:

  1. Chat is the front door — ask a question, attach a file, switch models, turn on agentic mode when you want the model to do something rather than just answer.
  2. Workflows are chats made repeatable — once you’ve worked something out by hand, you (or the model) can capture it as a workflow that runs the same way every time.
  3. Connections — integrations (MCP), data sources, and your own models — are what let both chat and workflows touch the real world.
  4. Memory, skills, and prompts shape how the model behaves and what it knows about you, everywhere it runs.

You don’t need all of it on day one. The Quickstart gets you to a useful first result in a few minutes; Core concepts explains the mental model so the rest of the docs make sense.

Start the Quickstart