OwnwareDocs
OwnwareEngine
View source
concept

Models

Ownware is provider-agnostic. A model is always one string — provider:model — set in the profile ("model" in agent.json), overridden per run (the model field on POST /api/v1/run), or picked automatically by the client from GET /api/v1/models.

For AI agentscondensed build recipe

keyless path = install Ollama, ollama pull llama3.2, use "ollama:llama3.2" — no env var needed (OLLAMA_HOST optional). Cloud paths = set exactly one of ANTHROPIC_API_KEY / OPENAI_API_KEY / GOOGLE_API_KEY / OPENROUTER_API_KEY in the environment before the gateway starts. GET /api/v1/models returns [{id, hasCredentials, default?}] — filter on hasCredentials to find what's usable right now.

Providers at a glance

Provider Env var Keyless? Example model string
Ollama (local) — (OLLAMA_HOST optional) ✅ free, private ollama:llama3.2
Anthropic ANTHROPIC_API_KEY — anthropic:claude-sonnet-4-6
OpenAI OPENAI_API_KEY — openai:gpt-5.5
Google GOOGLE_API_KEY — google:gemini-2.5-flash
OpenRouter (300+ models) OPENROUTER_API_KEY — openrouter:kimi-k2.7-code
bash
brew install ollama && ollama pull llama3.2       # macOS
curl -fsSL https://ollama.com/install.sh | sh && ollama pull llama3.2   # Linux

Then use "model": "ollama:llama3.2" — fully local, free, private. The quickstart chat detects a running Ollama automatically.

How clients discover models

GET /api/v1/models reports every known model with hasCredentials — true when a key is set or a local Ollama is reachable right now. default: true is flagged per provider (so several models carry it), so don't do models.find(m => m.default). The quickstart's chat.mjs instead takes the first usable model (preferring one flagged default). Build your clients the same way and "no config" stays true.

Keys can also be saved through the gateway's credential vault instead of the environment; either way they live on your machine (~/.ownware/) and never leave it.

Next steps