Inside Project HydraFusion: multi-model orchestration in the GitHub Copilot CLI
Julia Kasper walks through Project HydraFusion, a new multi-model orchestration approach in GitHub Copilot that aims to remove the guesswork of choosing the “right” model for a coding task.
Overview
Project HydraFusion is presented as an orchestration layer for GitHub Copilot that:
- Evaluates user intent for a given prompt
- Routes the prompt through different execution strategies
- Uses multiple models behind the scenes to improve outcomes and efficiency
How to try HydraFusion today
HydraFusion is available as an experimental feature.
- Enable experimental mode in:
- GitHub Copilot CLI
- GitHub Copilot App
- VS Code
- Use the
/experimentalcommand to access experimental behavior.
HydraFusion announcement/discussion:
HydraFusion vs. Auto Model Select
The video contrasts HydraFusion with “Auto Model Select”, positioning HydraFusion as a more explicit orchestration approach that can choose between different execution patterns (not just a single automatic model choice).
HydraFusion execution strategies
HydraFusion routes prompts using different strategies depending on intent and context.
Single
- Runs the prompt through a single execution path.
Cascade
- Uses a staged approach where one step can feed into the next.
Critique
- Adds a critique/evaluation step intended to improve the final output quality.
Demos: efficiency and behavior in practice
The episode includes demonstrations focused on:
- Efficiency improvements when HydraFusion selects an execution strategy automatically
- How the different strategies behave in real usage
Transparency and avoiding user bias
A stated goal is to make model selection and orchestration less dependent on user guesswork, while improving transparency around what the system is doing and why.
What’s next
The final segment covers upcoming direction for HydraFusion and encourages viewers to try the experimental feature and share feedback.