Product
Introducing Agent Runtime 2.0
Faster cold starts, durable memory and step-level retries. Here is what changed under the hood.
What is new
Runtime 2.0 rewrites the scheduler that sits between your agents and the models they call. Cold starts dropped from seconds to a few hundred milliseconds, and every step now checkpoints its state, so a failed tool call retries from where it stopped instead of from the top.
Durable memory
Memory is now stored per agent, per team or per customer, and summarised automatically as it grows. Agents recall context across sessions without you managing a vector database.
Existing agents move to the new runtime automatically over the next two weeks. Nothing changes in your code.