How is Tenbin different from a one-shot converter?

One-shot converters emit a request and stop. Tenbin adds lint, single and batch evaluation for threshold calibration, rerank, taxonomy walks, cost guards, and runbook procedures, all aimed at a measured design rather than a prompt.

Do I have to run the MCP server to get value?

No. Without the MCP registered the skill still designs questions and generates code, lints offline, and evaluates through scripts with plain Python and no SDK. Only reranking and taxonomy walks require the MCP.

What does a typical triage session produce?

Atomic Choice, Score, and Noul questions with lint results, raw answers on a few of your tickets, accuracy per confidence band on labelled rows, and production code with thresholds in one file.

Does a TypeSafe key gate the whole workflow, or only the paid steps?

Only the paid steps. Without a key, project-use suggestions, state and question generation, code generation, lint and cost estimates all still work, as do mocked tests; paid evaluation runs only when you request it. Node.js 20 or later is the stated prerequisite, and make targets handle the chmod 600 env file, the build, and MCP registration. Accuracy and thresholds stay unmeasured until you validate on labelled rows.

What does the question linter catch before I spend a call?

Lint findings include compound questions, counting, date math, degree-only levels, missing state paths and token budget issues. It runs offline through tenbin_lint_questions or scripts/lint_questions.py with no API call.

What are Tenbin's prerequisites and install steps?

Node.js 20 or later and a TypeSafe API key from console.typesafe.ai. Clone the repo, then make env-file creates ~/.config/tenbin/env with chmod 600, make build-mcp builds the MCP server and make register-claude registers it; make install-claude does all of it at once.

What endpoint, model and price does Tenbin document for Jev?

One endpoint, POST https://api.typesafe.ai/v1/systemone, model jev-1.13.0 with alias jev-latest. It is billed at $0.042 per million input tokens with output free, and answers in around 100 ms.

CommunityJev Practices & Patterns

Tenbin

Design-time toolkit for using Jev from a coding agent, combining docs, a stdio MCP server, and an agent skill for decomposing judgments into Choice, Score, and Noul.

See highlights

Open github.com

Stars: 4
Language: TypeScript
License: MIT
Published by: Community
Author: simota
Last verified: Oct 2026

simota/tenbin

User repository on GitHub

View on GitHub

MCP server and agent skill for the TypeSafe AI System One API (Jev): decompose a judgment into Choice / Score / Noul questions, lint them, measure on labelled data, and put calibrated thresholds in code

Forks: 2
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

A repository for using TypeSafe AI (System One API, model Jev) from a coding agent at design time. It has three parts. Operational procedures (setup, daily work, incident response, maintenance) are in docs/04-runbook.md. Conventions for agents are in AGENTS.md. Ask your coding agent for a judgment your code needs, and it comes back with a measured design, not a prompt: Everything runs at design time. The output is code and numbers you own; production calls the TypeSafe SDK directly and never depends on the MCP.

Quoted from the project's README. Read the full README.

Tagscommunity awesomejev catalog-import

Works forJev for Classification Jev for Ranking Jev for Tool calling

Used inJev for Software & DevTools

Languages: TypeScript

Highlights

MCP server offers question lint, single and batch evaluation for threshold calibration, rerank, taxonomy walk, and cost guard.
Skill turns triage, guardrails, reranking, taxonomy placement, and extractor checks into linted questions plus code.
Production calls the TypeSafe SDK directly, so shipped code never depends on the MCP at runtime.
Lint, cost estimates, code generation, and mocked tests work without a key; paid evaluation runs only on request.

Watch out

Model accuracy and thresholds stay unmeasured until validated on labelled rows, and reranking plus the taxonomy walk need the MCP. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

jevalyzer

Grades the coding agent sessions already sitting on your disk. Jevalyzer reads Claude Code, Codex CLI, opencode, Gemini CLI and Antigravity transcripts, scores each exchange, and renders an interactive HTML report or terminal dashboard.