lingo-loop 0.1.4

Local-first Claude Code language tutor

lingo-loop

Learn languages inside your AI coding assistant — local-first, no telemetry.

(Demo recording pending — see docs/internal/launch-checklist.md.)

Why

• Local-first. Your profile, vocabulary, and progress live in plain files on your machine.
• No telemetry. Nothing is sent anywhere except the LLM API call your host already makes.
• Works inside your IDE. Practice reading, writing, and vocabulary without leaving Claude Code, Codex, Hermes, or OpenClaw.

Features

• Reading practice with adaptive vocabulary surfacing.
• Short writing prompts with corrections from your assistant.
• Spaced-repetition vocabulary loop, scheduled across sessions.
• Lesson and transcript flows tailored to your native and target languages.
• Progress tracking you can inspect with a single command.
• Editable YAML profile — language pair, level, and preferences in one place.
• Works across four AI coding hosts with the same CLI underneath.

Quick start (Claude Code)

Install with uv (or clone and run uv pip install -e ".[dev]" for a dev setup — see CONTRIBUTING.md).

# 1. Install the CLI
uv tool install lingo-loop==0.1.3

# 2. Detect AI hosts and install tutor skills for the ones you use
tutor doctor --json
tutor init

# 3. Write your learner profile (native + target language)
tutor setup write --json '{"profile":{"native_language":"en","target_language":"uk"},"preferences":{}}'

# 4. Restart Claude if needed, then ask: "start a reading session"

tutor init detects Claude, Codex, Hermes, and OpenClaw and shows a keyboard menu: arrow keys move, Space toggles providers, and Enter continues/applies. All four providers receive the shared skills/<skill-name>/SKILL.md hub. Hermes and OpenClaw also receive their registration files. Rerun any time to repair drift — it never touches your learner profile, history, or secrets. Non-interactive form: tutor init --provider claude --yes (also --dry-run, --json).

If 0.1.3 has not propagated yet, use the source-tag fallback:

uv tool install "git+https://github.com/artemVeduta/lingo-loop@v0.1.2"

Full Claude install doc: docs/install/claude.md

Other hosts

• Codex — docs/install/codex.md
• Hermes — docs/install/hermes.md
• OpenClaw — docs/install/openclaw.md

How it works

┌──────────────┐       ┌────────────────┐       ┌─────────────────┐
│  AI Host     │──────▶│  Tutor skill   │──────▶│  tutor CLI      │
│  (Claude/    │       │  (markdown +   │       │  (Python)       │
│  Codex/...)  │       │  prompts)      │       │                 │
└──────────────┘       └────────────────┘       └─────────────────┘
                                                          │
                                          ┌───────────────┼────────────────┐
                                          ▼               ▼                ▼
                                   ┌──────────┐   ┌──────────────┐   ┌─────────┐
                                   │ profile  │   │  learning    │   │  data/  │
                                   │  YAML    │   │   SQLite     │   │ defaults│
                                   └──────────┘   └──────────────┘   └─────────┘

       All state local. No network calls except the host's own LLM API.

Privacy

Your learner profile and preferences live in editable YAML files. Transactional learning state (vocabulary intervals, session history, progress) stays in a local SQLite database under your user data directory. There is no telemetry, no account, no cloud sync, and no remote storage. The only network traffic is the LLM API call your AI host already makes on your behalf.

Roadmap

Tracked publicly on GitHub:

• Milestones — versioned scope.
• Projects — in-flight work.
• Discussions — propose ideas before they become issues.

Contributing

Pull requests welcome. See CONTRIBUTING.md for dev setup, branch naming, commit conventions, and the DCO sign-off policy.

License

MIT © 2026 Artem Veduta