Getting Started
Configure a Provider
OPIDE supports 10 AI providers. Each agent profile can be assigned its own model and provider. API keys are stored in the OS keychain — never written to disk in plaintext.
Opening provider settings#
On first launch the provider setup panel opens automatically. After that, open it at any time via Cmd+, (Settings) then navigate to Providers.
Each provider needs an API key (except Ollama, which runs locally). Add as many providers as you want and switch between them per-agent.
Supported providers#
| Provider | Type | Notes |
|---|---|---|
| Anthropic | Native API | Claude 3.5, Claude 3 Opus, Sonnet, Haiku |
Native API | Gemini 2.0, Gemini 1.5 Pro and Flash | |
| OpenAI | Native API | GPT-4o, GPT-4 Turbo, o1, o3 |
| DeepSeek | OpenAI-compatible | DeepSeek-V3, DeepSeek-R1 |
| Moonshot (Kimi) | OpenAI-compatible | Kimi K2 and other Moonshot models |
| xAI (Grok) | OpenAI-compatible | Grok 2, Grok 2 Vision |
| Mistral | OpenAI-compatible | Mistral Large, Codestral |
| OpenRouter | OpenAI-compatible | Access to 100+ models through one key |
| Ollama | Local | Any model pulled via Ollama. No API key required. |
| Custom | OpenAI-compatible | Any endpoint implementing the OpenAI API spec |
Connecting Ollama for local models#
Ollama runs models locally on your machine. No API key, no data leaving your network. Install Ollama separately, pull a model, and OPIDE connects automatically.
- ◆Install Ollama from ollama.com
- ◆Pull a model:
ollama pull llama3.2or any other supported model - ◆In OPIDE Settings, select Ollama as a provider — the endpoint defaults to
http://localhost:11434 - ◆Select your pulled model from the dropdown and assign it to an agent profile
qwen2.5-coder:32b and deepseek-coder-v2:16b perform well. Run on a machine with at least 16GB RAM.Model routing#
Model assignment follows a 5-level fallback hierarchy. At each level, OPIDE checks for an override before falling through to the next:
- ◆Per-request override — set programmatically for a single call
- ◆Agent profile model — the model assigned to the specific agent (Coder, Researcher, etc.)
- ◆Specialty model — a model assigned to all agents of a given specialty
- ◆Provider default model — the default model for the selected provider
- ◆Global fallback — the model set in global Settings
API key security#
OPIDE never writes API keys to disk in plaintext. All keys are stored in the OS keychain: macOS Keychain on macOS, secret-service on Linux, Windows Credential Manager on Windows. A single unlock gives the agent access for the session.