jevbooks

← All projects

chakuho

taku-me/chakuho

Jev-compatible System One decision endpoint over a local LLM.

Agent Decisions90%Runner-up: Infra / SDKs / IntegrationsMany questions, one request

What it is

chakuho is a Python HTTP service that answers typed decision questions (choice, noul/boolean, score) about a given state, returning probability distributions instead of generated text. It targets developers who want Jev-style System One decisions from an already-running local instruct model.

How it uses Jev

It emulates Jev's System One model rather than using Jev itself. The caller declares questions with type, instructions and criteria; chakuho labels options A-Z/a-z, sends one prompt to an OpenAI-compatible backend with max_tokens=1, temperature=0 and logprobs, aggregates top logprobs onto the declared labels, and returns choice/noul/score with probabilities and coverage.

Primitives:choicenoulscore

Technique worth stealing

Read the next-token logprob distribution of a general instruct model and aggregate it onto caller-declared option labels, so no text is generated.

Try it

uv sync; uv run chakuho serve --host 0.0.0.0 --port 9750; curl localhost:9750/health

View on GitHub

judged by Jevjev-1.13.0

Evidence

Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.

  • Jev-centricyes0.89
  • Shows a System One patternunclear0.42
  • Handles uncertaintyno0.05
  • Measuredno0.06
  • Runnableyes0.98
  • Worth recommendingno0.29
  • Model replicayes0.89
  • Problem scopescore on a 0–2 scale1.01
  • About Jevyes0.97

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.