Everything *[NYC] 2026: see what we announced. →
Skip to content

Project Horizon

Self organizing agents that learn on the job

Your team. Your workstreams. Any model.


Coding agents had it easy.

Code gives agents a head start. They can see the repository, follow its rules and run tests to check their work.

Knowledge work depends on context. A campaign can be great for one brand and completely wrong for another. It depends on the audience, the brand and the moment. What works changes from team to team and month to month.

Much of that context lives with your team: what you tried last spring, what legal always flags, what your customers actually respond to. An agent won’t know any of that on day one.

It has to learn the way a new colleague does: on the job, with you.

That’s what Horizon does.

  • They learn on the job

    Horizon agents remember decisions, preferences and ways of working. They distill what matters over time, so you don’t have to keep explaining the same things.

  • They’re multiplayer

    The whole team works with the same agents, in the same space. Hand off work, review progress and steer together. Agents can take on roles, bring in other agents and work across documents, data, and images. Choose the model for each job.

  • They keep working when you’re not there

    Hand them a week’s work and come back to results. The agents in a space share a cloud computer, so they can pick up long jobs where they left off and check in when they need a decision.


Day one, you explain. By week three, they remember.

  • Day 1: “Draft the launch brief.” It asks: Who’s it for? Which template? Who signs off?
On day one, an agent asks who a launch brief is for before starting.
  • Week 3: “Draft the launch brief.” Your template, your audience. Flags the claim legal pushed back on and sends it to the right reviewer.
By week three, the agent drafts a detailed launch brief using the team’s remembered audience, claims and reviewers.

Horizon turns what each thread teaches it into lasting memory. Explain something once, and the whole team benefits.


You first. Then your team.

  1. You first. Your Home is your personal space, where you set up your own assistants for work. Give them your research, briefs and follow-ups, and they start learning how you work. Give your personal agents access to services you want them to work with.
  2. Set up a space for your team. Invite colleagues. The agents build up the team’s context and keep learning from everyone.
  3. Bring your own agents. Add the agents you already use alongside Horizon’s. Everyone works from the same plan and memory.

Get started with Project Horizon


Frequently asked questions

What is Horizon?

Horizon is a multiplayer workspace for the whole team. Its agents learn on the job, picking up what good looks like for your team, and carry on when you’re away.

How is it different from ChatGPT, Claude, or Copilot?

ChatGPT, Claude and Copilot are excellent assistants, with team and background features of their own. Horizon starts with the team. Its agents work in a shared space and build up the team’s context over time. They share the space’s cloud computer for longer jobs. You can use the models you already trust.

Who is it for?

Teams working on content operations, research, and analysis.

How does Horizon relate to Sanity?

Horizon is produced by Sanity, by the same people who build Sanity’s content platform. It’s a Sanity product, and it will be increasingly integrated with the rest of Sanity.