Skip to content
FucciUnavailablePublic

About

AI coaching in your terminal: pluggable domain coaches, persistent goal tracking, both-sides reasoning. Works with Claude, OpenAI, or a local model.

Resources

Code of conduct

Contributing

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

masta 🥋

An AI sensei for your goals — not a chatbot that spits out a plan in a table.

masta is an open-source, AI-provider-agnostic coaching system. It helps you turn a goal into a real plan, tracks your progress over time, searches actual evidence, argues both sides of every suggestion, keeps you motivated when you fail — and reminds you to rest. Coaching, not just prompting.

It is built around pluggable coaches — domain experts (e.g. backend engineering, safe weight loss) that each carry their own persona, knowledge, tools, and safety guardrails.

Why it's different

This is a system, not a prompt:

  • 🔌 Provider-agnostic — Claude, OpenAI, Azure/AI Foundry, DeepSeek, or a local model (Ollama). Switch with config, not code. Powered by LiteLLM behind a thin adapter so we're never locked in.
  • 🧠 Persistent memory — your goals, milestones, check-ins, and reflections live in a local SQLite database, so the sensei has continuity and can actually track you.
  • 🧩 Pluggable coaches — a coach is data, not code: a folder with a persona, knowledge, declared tools, and guardrails. Ship your own.
  • ⚖️ Both-sides reasoning — every recommendation comes with a steelmanned pro/con view, as a first-class step rather than an afterthought.
  • 🛟 Safety guardrails — sensitive coaches (health, weight loss) enforce disclaimers, safe ranges, and "talk to a professional" triggers.
  • 😴 Rest matters — break/rest reminders and check-ins are part of the design, not bolted on.

Status

Early alpha. MVP scope: the coach engine + persistent memory + both-sides reasoning, driven from a terminal chat, with provider swapping. Evidence search, scheduling/notifications, failure-recovery, and rest enforcement are scaffolded and arriving next (see ARCHITECTURE.md).

Quickstart

# 1. Install (uv recommended)
uv sync

# 2. Configure at least one provider
cp .env.example .env
$EDITOR .env

# 3. See what's wired up
uv run masta providers

# 4. Talk to a coach
uv run masta coach list
uv run masta chat --coach backend-engineering

CLI overview

Command What it does
masta chat --coach <name> Start an interactive coaching session
masta coach list / masta coach use <name> List / select coaches
masta goals View and manage tracked goals
masta checkin Log progress / how you're doing
masta providers List configured LLM providers & profiles
masta config Show resolved configuration

Writing your own coach

A coach is a directory with a coach.yaml, a persona.md, and optional knowledge/. See docs/writing-a-coach.md and the examples in examples/coaches/.

Contributing

PRs welcome — especially new coaches. See CONTRIBUTING.md.

License

MIT

About

AI coaching in your terminal: pluggable domain coaches, persistent goal tracking, both-sides reasoning. Works with Claude, OpenAI, or a local model.

Resources

Code of conduct

Contributing

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages