Pi Agent Harness has released version 0.86.0 with a new Prompt Cache Warming feature aimed at reducing repeated token costs during long-running tool tasks. When an agent job runs long enough that a prompt cache is close to expiring, Pi can resend the previous request before the cache is cleared. The refresh generates just one output token while keeping the already cached long context alive.
The update targets a common issue across many model services, where prompt caches are dropped after a period without requests. If the task later resumes, the same long context may be billed again as a full input. Pi said the new mechanism is designed to avoid that kind of cache miss, but it does not keep caches alive indiscriminately.
Instead, the system performs a cost check first. A refresh is triggered only if the expected savings from avoiding a cache miss, after subtracting the refresh cost, still amount to at least $0.05. Pi also noted that each refresh is not free: users still pay for reading the full cached segment plus the cost of one output token.
Pi Agent Harness released version 0.86.0 with a new Prompt Cache Warming feature.
For long-running tool tasks handled by an agent, Pi will resend the previous request before the cache expires. It generates only one token, allowing the already cached long context to remain available.
Many model services clear Prompt Cache after a period with no requests. If the task continues later, the same long context may be billed again as a full input.
Pi said the new feature is meant to avoid that kind of cache miss, but it does not keep caches alive by default. The system first calculates the economics. It will refresh only when the expected cost avoided from a cache miss, minus the refresh cost, still saves at least $0.05.
Each refresh still carries a cost of its own, including the fee for reading the full cached segment and the cost of one output token.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
