ContextAPI · Reference

ContextAPI

ContextAPI sits in front of any LLM and returns a token-efficient version of your prompt. Two endpoints — optimize and parse — cover most workflows.

Base URL

All ContextAPI requests use the following base URL:

https://api.cosavu.com/v1/context

Optimize

Compress a prompt before sending to your LLM. Returns the optimised prompt along with token savings and the parsed PromptIR.

Request body

Example

const result = await cosavu.context.optimize({
  prompt: "Could you please kindly explain RAG in detail",
  budget: 512,
  mode: "auto",
})
 
console.log(result.optimizedPrompt)
console.log(result.tokensSaved)

Response

{
  "color:#3b82f6">"optimizedPrompt": "Explain RAG in detail.",
  "color:#3b82f6">"originalTokens": 84,
  "color:#3b82f6">"outputTokens": 32,
  "color:#3b82f6">"tokensSaved": 52,
  "color:#3b82f6">"compressionPct": 0.619,
  "color:#3b82f6">"latencyMs": 12,
  "color:#3b82f6">"ir": {
    "color:#3b82f6">"blocks": [
      { "color:#3b82f6">"blockType": "INSTRUCTION", ">"tokens": 32, ">"isCompressed": true }
    ],
    "color:#3b82f6">"messinessScore": 0.61
  }
}

Parse

Parse a prompt into its typed PromptIR blocks without optimising. Useful when you want to inspect prompt structure or build custom optimisation logic on top.

Example

const ir = await cosavu.context.parse({
  prompt: "You are an expert. Given these docs, answer concisely.",
})
 
for (const block of ir.blocks) {
  console.log(block.blockType, block.tokens)
}

Block types