Context is like ‘historical facts’. Context is needed if you want to use agents and AI models to make instructions. If you have a context pipeline that constantly collects and updates ‘historical facts’, you have a ‘ground truth’. This is something for agents in generative environments to use. That ground truth can only come from you, because you feed your context pipeline!
Explore our whiteboard talks


Experiment in secure sandboxes
Deploy sandboxed applications
Once you understand how your context pipeline works and you know how it scales, you can get creative with ways to feed it.
You can deploy applications to anything running in your environment. These applications can be set to collect context in many ways.

Bet some tokens
Play the game
Learn where to take your ‘context’. Some context can be turned into instructions. The way the game is played is that you bet that you have good enough context, that each time the inference iterates with your context, the model will use it to build toward a result that is more valuable to you than the cost of getting the ‘context’ and the tokens used in the iterations. So you can see how you should bet. If the instructions you get really pay off, I mean can be executed by something that creates value. Put as much ‘context’ and as many ‘tokens’ on the table as you can.

Affordable instructions
Economics
Collect context, Use a model (inference), Get instructions, Get a machine to run instructions.