Skip to main content

​
optimize

Run compression and token estimation in a single call. Ideal when you need byte savings plus model-specific token counts.

​
Parameters

NameTypeRequiredDescription
dataAny✅JSON payload to compress.
optionsEncodeOptions❌Same knobs as compress (indent, delimiter, length_marker).
token_modelslist[str]❌Model IDs (e.g., ["gpt-4o-mini", "claude-3-5-sonnet-20240620"]) for token stats.

​
Code example

payload = {
    "data": {
        "messages": [
            {"role": "user", "content": "Audit this JSON blob"},
        ],
        "metadata": {"team": "ops"},
    },
    "token_models": ["gpt-4o-mini", "claude-3-5-sonnet-20240620"]
}

async with KaizenClient.from_env() as client:
    optimized = await client.optimize(payload)
    print(optimized["stats"])
    print(optimized["token_stats"]["gpt-4o-mini"])

​
Response example

{
  "operation": "optimize",
  "status": "ok",
  "result": "KTOF:...",
  "stats": {
    "original_bytes": 5120,
    "compressed_bytes": 1184,
    "reduction_ratio": 0.231
  },
  "token_stats": {
    "gpt-4o-mini": {"original": 640, "compressed": 184},
    "claude-3-5-sonnet-20240620": {"original": 612, "compressed": 176}
  }
}

​
Errors

  • 400 → token_models must be a list of strings; payload must include data.
  • 429 → Token estimation rate limit exceeded (per model). Retry after Retry-After.

​
Notes

  • Token counts use Kaizen’s calibrated model tables; treat them as accurate estimates, not billing-grade numbers.
  • When you only care about prompt compression (no token stats), prefer compress to avoid the extra processing overhead.