Krater Agent does the work for you, not just talks about it Most AI chatbots answer your question and stop. The Krater Agent keeps going. You give it an outcome you want, and it figures out the ste...
Most AI chatbots answer your question and stop. The Krater Agent keeps going. You give it an outcome you want, and it figures out the steps, does them, and hands you the finished result.
It’s the most capable AI agent built into one app: it browses the web, does research, generates images, videos, audio, 3D, slides, documents and diagrams, writes and ships working code, and acts on your behalf inside the apps you already use (Slack, Gmail, Notion, Linear, GitHub, HubSpot and more — over 1,400 app connections) via Additions. One chat box. Everything in.
The Agent is on by default in every new chat. You don’t have to switch anything on. If you’ve been chatting with Krater, you’ve already been using it.
Think of regular AI chat as a friend who’s great at conversation. You ask, they answer.
Think of the Krater Agent as a friend who’s great at conversation and has a laptop in front of them with all your favourite apps already logged in. You ask, and they can:
/image, /video, /tts) for you, so you never have to remember themIt’s the difference between a friend who tells you a recipe and a friend who shows up at your kitchen and cooks dinner.
When you send a message, you’ll often see the Agent showing its work above the answer:
Underneath all that, you get the answer. The work happens up top so you can see what it did. The answer itself is what you came for.
This is the short version. In practice the Agent mixes and matches these into one smooth result.
Krater plugs into 1,400+ apps through Additions. Once you’ve connected one, the Agent can use it on your behalf, not just talk about it. A few of the things people do with it all day:
If there’s an app you already use, chances are the Agent can use it for you. See the full list on the Additions page.
You don’t have to pick a slash command yourself. Ask for an image, the Agent will use /image for you. Ask for a voiceover, it will use /tts. Ask it to transcribe a meeting recording, it will use /transcribe. They’re not a separate world. They’re tools the Agent reaches for when they fit.
Drop a file into the chat and the Agent treats it like a real document on a real computer. It can read it, clean it up, compare it to other files, pull data out of it, or turn it into something else (a CSV into a chart, a PDF into a summary, a recording into text).
Files the Agent makes for you also stay around for the whole conversation, so step two can use what step one produced.
The Agent can:
So “keep an eye on this and tell me when something changes” is a thing you can ask for, not just a feature description. Scheduled runs live under Tasks, where you can pause or delete them anytime.
For jobs with several independent parts, the Agent can run a few mini-helpers at the same time instead of doing everything one after the other. You don’t manage this. You just notice that complicated tasks come back faster than you expected.
These are real prompts that play to the Agent’s strengths. Copy one, change it to fit your world.
Research the top 5 electric delivery vans on sale in Europe this year. Pull the real specs, range, and prices from each manufacturer’s own site. Give me a comparison table, a chart of price vs range, and a one-page PDF I can share with the team.
Here’s a CSV of customer emails. Some are duplicated, some are typos. Clean it up, validate the addresses, and give me back a clean file plus a short note on what you removed and why.
Write a 90-second YouTube intro about why subscriptions are killing creativity. Generate a thumbnail in dark mode with warm accents, draft the script, record the voiceover, and give me the script and the audio as separate files.
Build a one-page landing site for a fictional product called “Lumen”. Modern, dark mode, hero section, three features, a pricing block. Host it and send me the link.
Every Monday at 9am Copenhagen time, check the front page of Hacker News for any stories about AI tools. Send me a one-paragraph summary of anything interesting.
| You want… | Best fit |
|---|---|
| A quick factual answer | Regular chat (the Agent will also handle it, just no extra steps needed) |
| A single image, voice clip, or transcription | A slash command like /image, /tts, /transcribe |
| Several steps blended into one result | The Agent |
| Real research with sources and a deliverable | The Agent |
| A working app, site, or document you can open | The Agent |
| Something the Agent should keep doing on a schedule | The Agent |
You never have to choose between them. The Agent uses slash commands and tools as needed. They’re not separate worlds.
The Agent is on by default, but you’re in charge.
Click the settings icon at the top of any chat. The Krater Agent switch is right there under AI Behavior. Turn it off if you just want a plain conversation in that one chat.
Open Settings → Data & Privacy → Chat Defaults. Flip Krater Agent off if you’d rather start every new chat in plain mode and turn it on when you need it. Most people leave it on.
Web search is automatic. When a question needs current information, the Agent can search the web and cite sources. You do not need to enable a separate setting, and ordinary answers are not forced to use web results.
The Agent doesn’t have its own price tag. You’re charged per message at Krater’s normal rate, based on the model you’re using.
What changes is how much one message gets done. An Agent reply that browses the web, generates an image, and writes you a document is doing the work of five or ten regular chat turns. So one Agent turn can use more Credits than one chat turn, but you got five turns’ worth of work for it.
A few tips that keep Credits efficient:
See Understanding Credits for the full picture on pricing.
Do I have to turn the Agent on? No. It’s on by default in every new chat. If you’ve been chatting with Krater, you’ve already been using it.
Is the Agent included in my plan? It’s available on every paid plan, Pro, Ultra and Max, including team seats on a paid workspace.
Does the Agent slow my chats down? Not for simple questions. If there’s nothing to plan or execute, you get a normal reply with no extra wait. Things only take longer when the Agent is actually doing real work for you, which is usually exactly what you want.
Can it see my files from older chats? No. Each conversation has its own private workspace. To use a file in a new chat, attach it there.
What happens if I close the tab while it’s working? The work keeps going. The Agent finishes the current step and pauses. Come back later and pick up where it left off.
Can the Agent run forever? No. There are limits in place to prevent runaway work, including a hard credit ceiling per run. For long-running projects, the right tool is a scheduled task, not an endless session.
Why did the toggle used to be called “auto-actions”? That was the original name when the Agent only handled media generation automatically. It can do much more now, so we renamed it. Same switch, much bigger product.
How is Krater Agent different from other AI agents? Most AI chatbots either talk well or run automation, not both. Krater Agent does it all in one chat box: it uses any of 500+ AI models to think, browses the web with real sources, runs code in a private workspace, generates images, videos, audio, 3D, slides, documents and apps, and acts inside the apps you already use through Additions — Slack, Gmail, Notion, Linear, GitHub, HubSpot and more. You don’t switch tools. You just ask.
Can the Agent send emails, post to Slack or update my Notion docs? Yes. Connect the app once on the Additions page and the Agent can act inside it on your behalf. Same for calendars, CRMs, project management, dev tools, design tools and more.
Where do the files the Agent makes go? Everything the Agent generates for you (images, videos, audio, code visuals, documents) is automatically saved to your personal Krater Keep library. It stays there until you decide to delete it.
Can the Agent use slash commands like /image or /video on its own?
Yes. You don’t have to pick a slash command yourself. The Agent reaches for the right one when it fits the task.
Who do I email if I get stuck? hey@krater.ai. We read every message.