nomad.ai

nomad.ai gives any Drive page access to a local AI agent. The underlying model runs in an external OpenAI-compatible server (Ollama, LM Studio, etc.) configured in Nomad settings.

Configuration

Global settings (Nomad → Settings → AI Settings):

• ai_base_url — base URL of your OpenAI-compatible runtime, e.g. http://localhost:11434/v1
• ai_default_model — fallback model name when a Drive does not specify one, e.g. llama3.2:3b

Per-Drive opt-in — add an ai key to your Drive’s /index.json:

{ "ai": { "model": "llama3.2:3b" } }

Or delegate to another Drive’s AI Config:

{ "ai": "hyper://abc123..." }

AI Config — a Drive can include /ai/system.md (system prompt) and /ai/tools/ (tool definitions) to customise behaviour.

Resolution order: Drive’s /index.json → Space default → global fallback (bare inference, no system prompt).

nomad.ai.chat(messages)

Send a conversation to the AI and stream the response back as text chunks.

• messages Array<{role: string, content: string}>. Conversation history in OpenAI message format ("user", "assistant" roles).
• Returns AsyncIterator<string>. Yields text chunks as they arrive. The system prompt from the Drive’s AI Config is prepended automatically.

const messages = [{ role: 'user', content: 'Summarise this Drive.' }]

for await (const chunk of nomad.ai.chat(messages)) {
  process.stdout.write(chunk)
}

Built-in tools available to the model (called opaquely — the page only receives text output):