Use case
Agent engineering teams, after a long-chain task run leaves a model bill over budget, work through the full execution trace to decide which model each step should use and bring cost back in range.
Hard-coding model names in code, roughly routing by task type, or using only cheap models and fixing quality by hand.
Every step of a multi-step agent calls a large model, so cost grows with trajectory length; teams only see the bill afterwards and lack a way to adjust model choice per trajectory.
xOcto's call
Demand is evidenced
Trend: agent cost is shifting from single calls to cumulative spend across a whole trajectory, making trajectory-level routing a new optimization point. Entry: start with long-chain agent teams in support, scraping or code review, charging a share of saved spend or per trajectory; public material is one line only, so pricing, customers and timing remain unverified.
Reason to use it
Why users would choose it
Inference: unlike hard-coded models or coarse task-level routing, it reads the full trajectory and re-picks models step by step, moving cost control from after-the-fact bill review into runtime, so high-volume long-chain agent teams would pick it when spend exceeds budget; no retention or repeat-use evidence yet.
Where the easy answer breaks down
The tension worth following
An English validation note will follow from the public evidence.
If this is your job
Worth trying. Inference: unlike hard-coded models or coarse task-level routing, it reads the full trajectory and re-picks models step by step, moving cost control from after-the-fact bill review into runtime, so high-volume long-chain agent teams would pick it when spend exceeds budget; no retention or repeat-use evidence yet.
Entry and what to borrow
Trend: agent cost is shifting from single calls to cumulative spend across a whole trajectory, making trajectory-level routing a new optimization point. Entry: start with long-chain agent teams in support, scraping or code review, charging a share of saved spend or per trajectory; public material is one line only, so pricing, customers and timing remain unverified.