TE taste-engine
Open source

Taste Engine.

Ingest doctrine. Extract canon. Detect drift. Repair alignment. All local, all Ollama, no paid API.

Install

npm install -g @mcptoolshop/taste-engine

Init

taste init my-project && taste doctor

Review

taste ingest README.md && taste extract run && taste review run artifact.md

Features

From doctrine extraction to org-wide rollout.

Multi-Pass Extraction

8 specialized LLM passes extract thesis statements, design rules, anti-patterns, voice conventions, and more from your source docs.

Deterministic Verdicts

Rule-backed verdict ladder (aligned → contradiction) that the model cannot override. Rules always win.

4-Dimension Scoring

Every artifact is scored on thesis preservation, pattern fidelity, anti-pattern collision, and voice/naming fit.

3-Mode Repair

Patch edits for surface drift, structural repair for deep misalignment, goal redirection for irreparable artifacts.

Workflow Gate

Advisory, warn, or required enforcement modes with CI exit codes, override receipts, and promotion doctrine.

Portfolio Intelligence

Cross-repo drift families, graduation patterns, preset fit analysis, and adoption recommendations.

Org Control Plane

Promotion queues, demotion triggers, 7-category alerts, preview/apply/rollback actions with audit receipts.

Operator Workbench

Dark-theme React UI at localhost:3200 for daily operational awareness across your entire portfolio.

Ollama-First

Runs entirely against local Ollama. No cloud API, no paid tokens, no data leaves your machine.

Usage

Install

npm install -g @mcptoolshop/taste-engine

# Requires Ollama running locally
ollama pull qwen2.5:14b

Extract canon

taste init my-project
taste ingest README.md docs/architecture.md
taste extract run
taste curate queue
taste curate accept 

Review artifacts

taste curate freeze --tag v1
taste review run path/to/artifact.md
taste gate run

Org rollout

taste org matrix --dir ./portfolio
taste org queue --dir ./portfolio
taste workbench --dir ./portfolio