Harbor

Model orchestration

Harbor showcases prompt orchestration workflows using multiple models.
Skipper runs a prompt across several models and synthesizes the best parts into one final answer.

Try it!

Skipper

What is Skipper?

A panel of models answers the same prompt, a judge compares what came back, and a synthesizer writes the final response. One model's answer is hard to trust on its own, so Skipper makes the disagreement between them visible instead of hiding it.

How it works

01

Send a prompt

Choose which models should answer and which model should produce the final response.

02

Review the responses

Gust runs the calls concurrently. Harbor keeps each response and checks where the models agree or disagree.

03

Create a final answer

The selected synthesizer receives the analysis and combines the useful parts into one response.

About

DAGs for model orchestration

Marcio, the developer of Harbor

Hi, this is Marcio. I created Gust and I am now experimenting with leveraging its capabilities to orchestrate models.

Running prompts through a DAG orchestrator brings scalability and resilience: steps run concurrently, failures retry on their own, and long-running prompts keep going even when they are combined with other tasks.

Gust on GitHub