Releases·Tool Release
Rizzo AI Academy releases Rizzo Flow, open-source local implementation of TypeSafe's Jev typed-decision model
Rizzo AI Academy released Rizzo Flow, an open-source, local-first implementation of the idea behind Jev, TypeSafe's 'System One' model. The tool takes unstructured state as input and produces typed, probabilistic decisions from an LLM without generating a single token, running entirely on the user's own machine. It offers developers a way to get structured decision outputs locally without token generation costs.
240 engagement1 sourceMon, Sep 21, 2026, 12:00 AM
AI Overview
- Rizzo Flow, by Rizzo AI Academy, is an open-source, local-first implementation of TypeSafe's Jev 'System One' model concept
- The tool turns unstructured input state into typed, probabilistic decisions from an LLM without generating a single token
- It runs entirely on the user's own machine; detailed documentation is available in Italian
Related
Social·Community
TypeSafe's decision model, explained: how it differs from an LLM and how it fared on fraud
Blogger tests TypeSafe's Jev decision model on fraud use case
Blogger tests TypeSafe's Jev decision model on fraud use case
0 engagement1 sourcerssMon, Sep 28, 2026, 06:33 PM
Social·Community
I ran a $0.63 classifier over 150 intents + 1,000 out-of-scope utterances. The confidence number held up; the fallback bucket didn't.
An independent developer ran TypeSafe's Jev structured-output model over the full CLINC150 test split (4,500 in-scope plus 1,000 out-of-scope utterances) with a 151-way choice question. The returned confidence was well calibrated (ECE 0.025) and out-of-scope detection was solid, but 54 off-topic utterances were still assigned specific intents at confidence 0.9 or higher, so neither a fallback bucket nor a confidence gate fully catches misrouted input. The results include an escalation table showing the accuracy/human-review tradeoffs for confidence thresholds.
An independent developer ran TypeSafe's Jev structured-output model over the full CLINC150 test split (4,500 in-scope plus 1,000 out-of-scope utterances) with a 151-way choice question. The returned confidence was well calibrated (ECE 0.025) and out-of-scope detection was solid, but 54 off-topic utterances were still assigned specific intents at confidence 0.9 or higher, so neither a fallback bucket nor a confidence gate fully catches misrouted input. The results include an escalation table showing the accuracy/human-review tradeoffs for confidence thresholds.
4 engagement1 sourceredditSun, Sep 20, 2026, 05:28 PM
Products·concept
Claude Code skill that routes classification tasks to Jev decision-only model
A Claude Code skill that makes the coding agent use Jev, TypeSafe's decision-only model, for bucketing tasks instead of regex checks or prompt-and-parse loops. The agent declares the task shape, validates on 5-20 real items with a stdlib runner, and logs outcomes. Aimed at developers writing classification, routing, or triage logic with coding agents.
A Claude Code skill that makes the coding agent use Jev, TypeSafe's decision-only model, for bucketing tasks instead of regex checks or prompt-and-parse loops. The agent declares the task shape, validates on 5-20 real items with a stdlib runner, and logs outcomes. Aimed at developers writing classification, routing, or triage logic with coding agents.
1 engagement1 sourceredditFri, Sep 18, 2026, 02:03 PM
Products·launched
WaterSheep: open-source LLM that answers plain-text survey-style questions with option probabilities
WaterSheep is an open-source model that answers plain-text questions (yes/no, single choice, rating, multi-label) and returns a probability for every option, like a classification model. It was built as an open alternative to Jev by TypeSafe and accepts the same request format, so TypeSafe's Python SDK works against a local WaterSheep server. Code, model weights, results, and a paper are all open-sourced.
WaterSheep is an open-source model that answers plain-text questions (yes/no, single choice, rating, multi-label) and returns a probability for every option, like a classification model. It was built as an open alternative to Jev by TypeSafe and accepts the same request format, so TypeSafe's Python SDK works against a local WaterSheep server. Code, model weights, results, and a paper are all open-sourced.
2 engagement1 sourceredditFri, Oct 2, 2026, 10:35 AM
Social·Community
Jev AI for Developers: A Practical Guide to TypeSafe's Decision Model
Medium guide explains TypeSafe's decision model for small-scope AI integrations
Medium guide explains TypeSafe's decision model for small-scope AI integrations
0 engagement1 sourcerssFri, Oct 2, 2026, 11:52 PM