-
Notifications
You must be signed in to change notification settings - Fork 0
FAQ
PromptPilot is an SLM-powered control plane for AI coding agents. It uses a small model as the harness layer around Codex/Claude-style agents to route prompts, detect ambiguity, preserve constraints, compress noisy context, and decide when to pass through unchanged.
Yes. That is why PromptPilot does not use the SLM as the coding model.
The SLM is used as a bounded harness for workflow decisions: routing, clarification, safe rewriting, constraint extraction, low-risk compression, and passthrough decisions.
The expensive coding agent still does the hard work: reasoning about code, editing files, debugging failures, and fixing tests.
If the SLM is uncertain or transformation risk is high, the correct behavior is passthrough. A slightly more expensive run is better than a cheap but wrong run.
No. Token savings are useful, but they are not the goal by themselves. PromptPilot optimizes for semantic-preserving context control: preserve critical facts, compress low-value noise, ask for clarification when intent is ambiguous, and pass through unchanged when transformation risk is high.
No. PromptPilot sits before Codex/Claude-style agents. The frontier coding agent still performs implementation, debugging, test repair, and deep code reasoning.