Skip to main content
As developers, we frequently perform modifications across multiple files. For an AI agent to execute these changes reliably, it needs a higher degree of planning autonomy. This release introduces advanced task queues and transaction safety.

​
The Task Scheduler

Hypr now tracks goals using a structured tree queue. The model schedules, splits, and updates goals on-the-fly:
  • Actor: formulating planning lists.
  • Critic: validating changes against compiler outputs.
This dynamic sidebar updates in real-time, displaying pending, running, completed, and failed states.

​
Atomic Rollback Transactions

When refactoring multiple files, if a compiler error occurs mid-way, the codebase is left in a broken state. We built apply_multi_diff to address this. It backs up target files, writes edits, and compiles the code via bun run build. If compilation checks fail, it reverts all files to their original states within 200ms, returning the build diagnostic logs to the model for self-healing.