Scores repo files by relevance, blocks low-value context, summarizes bulky files, and builds a smaller Nova-ready context packet.
See exactly which files consumed the most tokens. Visualize context waste.
Pick only the files connected to the current bug, feature, or question.
Compare a noisy Nova run against a moonshot-routed run side by side.
Large repos dump irrelevant files into every prompt. moonshot identifies and removes these token sinks.


package-lock.json, yarn.lock, and auto-generated files consume massive token budgets with almost no value for most coding tasks.


Massive context dumps from debug logs, old documentation, and changelog files that often distract the model from the real bug.


Auth, admin, analytics, or payment code accidentally included in a checkout task. Different features bleeding into your prompt.


Reading entire directories when only 2-3 files matter. Bulk context that drowns out the signal in noise.

Enter a task like: "Fix the checkout bug where discounts are applied after tax."

moonshot scans the repo and scores files/snippets by relevance to your task.

Irrelevant files are blocked or summarized. Relevant context is assembled into a budgeted prompt.

Nova receives the optimized context and returns the fix, explanation, or patch.
Amazon Nova API, SWE-Bench style dataset, repo scanner, token estimator, and prompt builder.

Score and filter files before they reach Nova. TypeScript powered.
Full REST + WebSocket API. Stream agent outputs directly into your product.
Nova already handles inference. moonshot decides what context deserves to reach Nova. We optimize before the model call.
Filter and route context before it reaches Nova
Set limits and moonshot builds within them
See exactly why each file was included or blocked
Compare a baseline Nova run against a moonshot-routed run. Same task, cleaner context, smaller request.
Built for the Nova track. Designed for large-repo developer workflows.