Skip to main content

​
prompts_encode

Purpose-built encoder for chat-style prompts. It detects nested JSON, schemas, and metadata automatically.

​
Parameters

NameTypeRequiredDescription
promptAny✅Chat payload (messages, tools, metadata, etc.).
optionsEncodeOptions❌Same tuning knobs as compress.
auto_detect_jsonbool❌ (default True)Detect JSON fragments embedded inside strings.
schemasdict[str, Any]❌Supply JSON Schemas to help Kaizen segment structured fields.
metadatadict[str, Any]❌Arbitrary observability data echoed in responses.
token_modelslist[str]❌Request model-specific token stats.

​
Code example

prompt = {
    "messages": [
        {"role": "system", "content": "You compress things"},
        {"role": "user", "content": "Summarize the following JSON", "metadata": {"doc_id": "abc"}}
    ],
    "metadata": {"workflow": "support-reply"}
}

async with KaizenClient.from_env() as client:
    encoded = await client.prompts_encode({
        "prompt": prompt,
        "token_models": ["gpt-4o-mini"]
    })
    ktof_prompt = encoded["result"]
    stats = encoded["stats"]

​
Response example

{
  "operation": "prompts.encode",
  "status": "ok",
  "result": "KTOF:...",
  "stats": {
    "original_bytes": 4096,
    "compressed_bytes": 1024,
    "reduction_ratio": 0.25
  },
  "token_stats": {
    "gpt-4o-mini": {"original": 520, "compressed": 148}
  },
  "metadata": {"workflow": "support-reply"}
}

​
Errors

  • 400 → prompt missing or contains unsupported types.
  • 422 → Auto-detection failed to parse a JSON fragment (disable auto_detect_json as a fallback).

​
Notes

  • Use this method for every provider wrapper; the returned result is what you send upstream.
  • Store metadata (e.g., customer IDs, workflow names) to correlate savings or replay prompts for auditing.