dobi

An AI-powered personal scrum master that runs in your terminal. You chat naturally; the LLM proposes changes to your backlog, sprints, standups, and retros; you approve, edit, or reject via interactive cards. All state persists as markdown files in a local .dobi/ directory.

Quick start

npm install -g @plarotta/dobi

Then cd into any project and run:

dobi

On first launch, dobi will walk you through selecting an LLM provider (Anthropic, OpenAI, Google Gemini, xAI, Groq, or OpenRouter) and entering your API key. Config is saved to ~/.dobi/config.json.

After setup, dobi creates a .dobi/ directory in your project and helps you set up your first backlog items.

Supported providers

| Provider | Default model | Env var | | --- | --- | --- | | Anthropic | claude-sonnet-4-20250514 | ANTHROPIC_API_KEY | | OpenAI | gpt-4o | OPENAI_API_KEY | | Google Gemini | gemini-2.0-flash | GOOGLE_API_KEY | | xAI | grok-3-mini | XAI_API_KEY | | Groq | llama-3.3-70b-versatile | GROQ_API_KEY | | OpenRouter | anthropic/claude-sonnet-4 | OPENROUTER_API_KEY |

You can also skip the setup wizard by setting env vars directly:

export DOBI_PROVIDER=anthropic
export DOBI_API_KEY=sk-ant-...
# Optional: override the default model
export DOBI_MODEL=claude-sonnet-4-20250514

Or, for backwards compatibility, just set ANTHROPIC_API_KEY.

Usage

Interactive mode (default)

dobi

Opens a TUI with a chat interface. Talk to your scrum master in natural language:

• "Add implement caching to the backlog, 5 points, backend"
• "Let's plan the next sprint"
• "I finished the auth flow"
• "Let's do standup"
• "Close the sprint"

The LLM reads your scrum data, proposes changes via approval cards, and you accept (y), edit (e), or reject (n).

CLI mode

dobi --cmd "add caching layer to backlog, 5pts"

Runs a single-turn interaction, auto-approves all proposals, prints the result, and exits. Useful for scripting or quick updates.

Keyboard shortcuts

| Key | Action | | --- | --- | | Ctrl+D | Toggle dashboard (kanban board, velocity, logs) | | Ctrl+C | Save session and quit | | B/L/V/K/R | Switch dashboard panels (Board, Log, Velocity, Backlog, Retros) |

Slash commands

Type these in the chat input for quick actions:

| Command | Action | | --- | --- | | /plan | Start sprint planning | | /standup | Draft today's standup | | /retro | Run a retrospective | | /close | Close the current sprint | | /status | Get a status briefing | | /backlog | Show the backlog | | /velocity | Show velocity trend | | /help | Show available commands |

How it works

Every interaction follows the propose-review-commit pattern:

You say what you did or what you're thinking about
                    |
    LLM interprets, reads scrum state, proposes changes
                    |
       You approve, edit, or reject the proposal
                    |
              State persists to markdown

The LLM never writes to disk without your approval. All mutations go through approval cards that show exactly what will change.

Data storage

All scrum state lives in .dobi/ as human-readable markdown:

.dobi/
  backlog.md              # Product backlog
  sprints/
    sprint-001.md         # Sprint files with items and metadata
  standups/
    2026-04-08.md         # Daily standups
  retros/
    sprint-001-retro.md   # Sprint retrospectives
  sessions/
    current.json          # Conversation history
    archive/              # Archived sessions

Files are plain markdown you can read, edit, or version with git.

Item format

- [ ] `a3f1` **Build auth flow** (3 pts) [backend] -- OAuth2 login
- [~] `c4d5` **Vector store** (5 pts) [backend] -- In progress
- [x] `b7c2` **Setup CI** (2 pts) [infra] -- Done

Status: [ ] = todo, [~] = in progress, [x] = done.

Development

git clone https://github.com/plarotta/dobi.git
cd dobi
npm install
npm run build
npm start

npm run build              # Compile TypeScript
npm run dev                # Watch mode
npm run test               # Run all tests
npm run test:watch         # Watch mode for tests
npx vitest run tests/data/backlog.test.ts   # Single test file

Tech stack

• TypeScript (strict, ESM, ES2022)
• pi-ai -- multi-provider LLM routing
• pi-agent-core -- agent loop, tool calling, event streaming
• pi-tui -- terminal UI framework
• vitest -- testing

License

MIT