Asoba Asoba Open Source
GitHub Asoba.co Contact

Getting Started with Nehanda CLI

Install and configure Nehanda CLI, then connect it to a provider and run your first agent session.

Requirements

1. Installation

git clone https://github.com/AsobaCloud/nehanda-cli.git
cd nehanda-cli
npm install

2. First Launch

npm start

This launches the interactive Ink TUI. On first run, the engine initializes the SQLite database at ~/.config/nehanda/ona-session.db and creates all tables.

You’ll see the REPL prompt:

❯

3. Connect a Provider

Nehanda CLI works out of the box with multiple providers. Choose one:

Option A: Nehanda Cloud (Default)

The CLI defaults to https://nehanda-ml.asoba.co/v1. If an API key is required:

❯ /key
New Nehanda API key: <your-key>
✓ Key saved.

Option B: Anthropic Claude

Set your API key as an environment variable, then switch provider:

export ANTHROPIC_API_KEY=sk-ant-...

❯ /config set model_config.provider claude_code_subscription
❯ /config set model_config.model_id claude-sonnet-4-20250514

Option C: LM Studio (Local)

  1. Download LM Studio and load a model
  2. Start the local server (default port 1234 or 8000)
  3. Switch provider:
❯ /config set model_config.provider lm_studio_local
❯ /model

Option D: Ollama (Local or Remote)

# Local Ollama
ollama serve  # starts on port 11434

❯ /config set model_config.provider ollama
❯ /config set model_config.base_url http://localhost:11434/v1
❯ /model llama3

For a remote Ollama instance on your LAN:

❯ /config set model_config.provider ollama
❯ /config set model_config.base_url http://my-server.local:11434/v1
❯ /model deepseek-coder-v2:latest

Option E: Any OpenAI-Compatible API

export OPENAI_API_KEY=sk-...

❯ /config set model_config.provider openai_compatible
❯ /config set model_config.base_url https://your-endpoint.com/v1
❯ /model your-model-name

4. Your First Session

Once connected, type a task:

❯ Read package.json and summarize the project

The engine will:

  1. Send your message to the provider
  2. The model will call the Read tool
  3. The permission gate may ask for confirmation (press y)
  4. The tool result is returned to the model
  5. The model responds with a summary

Try the SDLC Workflow

For a more structured workflow, try a task that triggers the full SDLC:

❯ Add input validation to the username field

The engine will ask: Use SDLC workflow (plan → implement → test)? [Y/n]

Press Enter (or y) to start the governed workflow:

  1. Explore — The model reads relevant files
  2. Planning — The model writes a plan (you approve or reject)
  3. Implement — The model makes code changes
  4. Test — The model runs tests and mandatory safety checks

See SDLC Workflow for the full walkthrough.

5. Useful First Commands

❯ /help            # Show all commands
❯ /model           # List available models
❯ /config          # Show current settings
❯ /mcp status      # Check MCP server status
❯ /clear           # Reset conversation
❯ /exit            # Quit

6. Optional: Configure MCP Servers

Add external tool servers by creating mcp.json in your project root:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "."]
    }
  }
}

Verify with /mcp list. See MCP Integration for the full guide.

7. Optional: Settings File

Create .ona/settings.json in your project root for project-specific configuration:

{
  "model_config": {
    "provider": "claude_code_subscription",
    "model_id": "claude-sonnet-4-20250514"
  },
  "permissions": {
    "defaultMode": "acceptEdits"
  }
}

See Configuration for the full schema.

Next Steps