# PostHog/jeeves

> Jeeves is a reasoning model built on Qwen3.5-9B (LoRA + pointer head), trained with SFT and CISPO, that thinks before answering and returns calibrated probabilities for yes/no, multiple-choice and rating questions via a Jev-compatible API.

- Magnitude: 6.6 out of 10 — Early signal
- Stars: 71 total · +83 stars today, ≈ 119 by evening
- Star trust: star growth looks organic
- Category: Language models · Language: Python · License: MIT · Created: 2026-09-29 · Last push: 2026-09-29
- GitHub: https://github.com/PostHog/jeeves · Page: https://gitnova.dev/en/r/PostHog/jeeves

## Useful for

- Serve a local classification endpoint that answers in Jev's request format
- Swap Jev for a higher-accuracy model on out-of-domain tasks
- Inspect model confidence via probabilities and ECE on your own data

## Why it’s here

- 83 stars so far today, about 119 expected by the end of the day.
- The repository is 0 days old and already has 71 stars. With less than two weeks of history, there's no usual pace to compare the spike against yet.
- Hacker News: “Jeeves. Reasoning improves Jev-like decision models” — 68 points, 2 h ago.
- About 109 forks a day — people are taking the code.
- Recent forks include notable developers: @jonnydubowsky (299 followers).

## Star trust

Star growth looks organic.

Star-trust labels are heuristics based on the repository’s behavior, not a check of every stargazer.

## Numbers

- Forks: 9
- Issues and pull requests: 0
- Watchers: 0
- Average over the last week: 119 per day
- Usual pace: too little history (under two weeks)
- Stars in the last hour (measured): 31

## Stars per day, last 3 days (oldest → newest, today is partial)

2026-09-27 … 2026-09-29: 0, 0, 83

## Hacker News

- Jeeves. Reasoning improves Jev-like decision models — 68 points, 27 comments: https://news.ycombinator.com/item?id=49891290

## Spotted in now

- Spotted on Hacker News

## Similar by description

1. **jaredpalmer/kev** — 3.9 · Cooling · Language models · Python · +94 stars today, ≈ 179 by evening
   kev is a LoRA adapter with a small readout head on top of Qwen2.5-0.5B that answers many typed questions about a document in a single forward pass, returning calibrated probabilities instead of text.
   Full card: https://gitnova.dev/en/r/jaredpalmer/kev.md
2. **kshetrajna12/reflex** — 0.8 · Steady · Language models · Python · +2 stars today, ≈ 4 by evening
   An open decision model: given a state (text, JSON, or image) and typed questions, it returns calibrated probabilities over fixed answer options. Runs on top of Qwen3.5 and answers with numbers instead of free text.
   Full card: https://gitnova.dev/en/r/kshetrajna12/reflex.md
3. **InternLM/Intern-Decision** — 4.4 · Early signal · Language models · Python · +33 stars today, ≈ 61 by evening
   A multimodal decision model built on Qwen3.5 that turns a state, optional images, and typed questions (choice, score, yes/no) into decisions with probabilities. Ships with training, two inference backends, temperature calibration, and a…
   Full card: https://gitnova.dev/en/r/InternLM/Intern-Decision.md
4. **mode-io/vllm-jev** — 2.5 · Early signal · Language models · Python · +18 stars today, ≈ 29 by evening
   A vLLM-based server for Jev decision models: given a question and candidate answers, it returns a label and probability for each candidate. Runs on Linux with NVIDIA GPUs and on Apple Silicon via MLX/MPS.
   Full card: https://gitnova.dev/en/r/mode-io/vllm-jev.md
5. **allebee/jevk5** — 0.6 · Cooling · Language models · Python · +0 stars today, ≈ 0 by evening
   An open-weight alternative to TypeSafe Jev: given a state (ticket, log, policy, diff) and a yes/no, choice, or score question, it returns a probability for every option in one forward pass with zero generated tokens. Qwen3.5-4B weights…
   Full card: https://gitnova.dev/en/r/allebee/jevk5.md

---

Magnitude (0–10) measures how fast and how unusually interest in a repository is growing right now. It is not a quality score.
Days are UTC. “So far today” is a fact; “expected by the end of the day” is a forecast.
Summaries and use cases are written by an LLM (DeepSeek V4.1 Flash) from the README and may be inaccurate: verify specific claims (benchmarks, speed, hardware) in the repository itself.
Data as of 2026-09-29 13:28 UTC, updated every 30 minutes.