Top AI Product

ACC (Agent Context Compilation) tackles a clean idea: when agents solve problems, they produce massive trajectories — tool calls, environment observations, dozens of turns. That data is usually thrown away after the task. ACC converts it into long-context training material instead.

## How it works

When an agent works a search, software-engineering, or database-querying task, the evidence ends up scattered across many turns, requiring the model to integrate distant context segments. ACC compiles these trajectories into long-context QA pairs — questions whose answers require pulling together information spread across the full trajectory. Train on those, and the model improves at exactly the long-range integration that long-context reasoning demands.

## Why it matters

The bottleneck in long-context training is data: synthetic long documents are artificial, real long documents are scarce. Agent trajectories are a renewable source of genuinely long, genuinely dependency-rich context — generated for free every time an agent runs. ACC turns the exhaust of agent execution into fuel for the next model. It’s data-flywheel logic applied to long-context specifically.