Get started
Agents
Work that runs in Horison and leaves you a result: a memo, a screen, a comparison, or something you wrote yourself.
If Chat is the conversation, an agent is the job. You pick (or write) a brief, point it at a deal or a set of documents, and let it run. It keeps going if you leave the page. When it finishes you get a document you can read, edit, and share, with the sources still attached.
On the left rail this is labelled Agents. That is the name we use everywhere now. Older pages and some search results may still say workflows. They mean this same place.
What an agent is
An agent is a repeatable piece of work defined in Horison. The definition says what it is trying to produce, which tools it may use (your graph, your documents, Gmail, Attio, and so on), and when it should run. A run is one execution of that definition. History under Agents is the list of those runs. Chat history is a different list.
The useful ones are not generic chat prompts. They are the jobs a PE team already has names for: write the IC memo, screen this CIM, compare this target to deals we have already seen. Horison ships those as templates. You can also treat them as examples of a more general thing: an agent is a Markdown brief plus a bit of config for trigger, tools, and output.
- It runs inside Horison, as you, on the deals and files you can already see.
- It can use the same connected tools Chat can, including Gmail and Attio (writes in Attio still wait for an approval).
- It produces a result you can open later, not only a chat transcript.
What you can run today
Open Agents on the rail. You will see a catalogue of templates. They cover the obvious PE jobs: an IC memo, a CIM pass, a risk summary, a comparison against earlier deals, a quick screen. Pick one, choose the deal (or leave it firm-wide), and start a run.
Think of each template as a filled-in brief. The shape is the same as a custom agent: instructions, tools, a document at the end. The catalogue is the set we have written for you so you do not start from a blank page.
Writing your own
The next step is agents you write yourself. The brief lives in Markdown. Config around it says when it fires (by hand, every weekday, once a week), which model and tools it may use, and what the output should look like. A daily pipeline digest and an IC memo are the same kind of object. One of them happens to have shipped as a template first.
That builder is not in every workspace yet. Until it is, the catalogue is how you run an agent. The picture to hold is: a Markdown file that does a job, with a trigger and an output, not a separate product called workflows.
From Claude or ChatGPT, later
Today, connecting Claude or ChatGPT to Horison lets that client read your graph and documents. It cannot start an in-product agent or pull back a finished memo. That is coming: the same MCP connection will be able to trigger an agent and retrieve what it produced. Until then, start runs from Horison itself. See Connect an agent for the client setup, and for what is live versus what is next.
Not the same as Chat, or Claude
Three different things share the word “agent,” and mixing them up is the usual confusion.
- Chat is the conversation on the rail. Ask a question, follow up, use tools. Nothing has to be saved as a memo.
- Agents (this page) are jobs inside Horison that produce an output. Templates today. Your own Markdown definitions as that ships.
- Connect an agent is the other direction: Claude, ChatGPT, or Cursor signing into Horison over MCP, so those products can read your workspace. That client is not running the in-product catalogue.
They share your deals and documents. They do not share a screen, a session, or a tool list. Attio’s CRM tools stay in Horison Chat and Horison Agents. They are not on the hosted MCP server.