Overview - ScaleDown

ScaleDown Agent is an MCP server and Claude Code plugin that finds AI API calls in your project and shows where ScaleDown models can replace or shrink them. It analyzes code locally; no source code is sent to ScaleDown.

scaledown-team/SLM_Agent

Source, issues and the full README.

What it does

Evaluates. Scans for OpenAI, Anthropic, LangChain, LlamaIndex, Cohere and other AI calls.

Plans. Maps each call site to the model that fits:

Model	Use it to	Savings
`sd_compress`	Cut context tokens 50-70% before your LLM call (RAG, long docs)	Fewer tokens
`sd_classify`	Replace LLM classification calls	~95% cheaper
`sd_extract`	Replace LLM entity extraction calls	~95% cheaper
`sd_summarize`	Replace LLM summarization calls	~90% cheaper

Migrates. Applies the changes to your code, with your approval.
Optimizes. Tunes the prompt you pass to a ScaleDown model on your own data and benchmarks it against the frontier-model output it replaced.

Supported providers

OpenAI, Anthropic and LangChain (Python and TypeScript/JS); LlamaIndex, Cohere and Google Generative AI (Python); Vercel AI SDK (TypeScript/JS).

Install

Claude Code, Cursor, VS Code.

Evaluate

Migration report and MCP tools.

Optimize

Tune prompts on your data.