Hey folks,
An agent works in a loop, and every model call resends the conversation. So a big tool result that sits in the context gets billed again and again. Smart Context Management in Agent Mesh avoids that by saving large content as an artifact and giving the model a short reference and a preview, with tools to read more when it needs to.
What’s new is that you can see the effect. The Cost and Token Usage page has a Smart Context Savings card, marked Estimated, showing the input tokens kept out of the model and what they would have cost at each model’s price. It’s recorded per LLM call.
Two related changes help too. Tables with more than 20 rows are now rendered straight from their artifact instead of being typed out by the model. In a 40 turn benchmark that took the cost from $2.56 to $0.59. And every deployed agent sends a leaner fixed prompt, with shorter built-in tool descriptions and peers listed by name.
The video shows a run with a 150 row report:
Have you measured how much of your agent spend is resent context? I’d like to hear what you found. Reply below.
If you want to try it yourself, you can download Solace Agent Mesh. Check it out on Github: