Model providers

20+ supported LLM providers with streaming, tool calling, and extended reasoning

OpsCode works with 20+ model providers. You can switch models mid-session without restarting, use different models for different subagents, and run fully offline with Ollama.

Model format

Models are referenced using provider:model-name:

anthropic:claude-opus-4-7
openai:gpt-4.1
google_genai:gemini-2.5-pro
deepseek:deepseek-reasoner
ollama:llama3.3

You can set the model in several ways:

CLI flag: ops -M anthropic:claude-opus-4-7
Persistent default: ops --default-model openai:gpt-4.1
In-session: /model (opens the picker) or /model anthropic:claude-opus-4-7
Config file: Set [model].default in ~/.opscode/config.toml
Subagent override: Add model: provider:model-name in the subagent's AGENTS.md frontmatter

Supported providers

Provider	ID	Auth	Highlights
Anthropic	`anthropic`	`ANTHROPIC_API_KEY`	Claude 3.5/3.7, Extended Thinking, Vision
OpenAI	`openai`	`OPENAI_API_KEY`	GPT-4.1/4o, o1/o3-mini Reasoning, Vision
Google GenAI	`google_genai`	`GOOGLE_API_KEY`	Gemini 2.0/2.5 Pro/Flash, Flash Thinking
Google Vertex AI	`google_vertexai`	ADC (`GOOGLE_CLOUD_PROJECT`)	Enterprise Vertex AI endpoints
Azure OpenAI	`azure_openai`	`AZURE_OPENAI_API_KEY`	Azure-hosted OpenAI models
Groq	`groq`	`GROQ_API_KEY`	Ultra-low latency Llama, Qwen, DeepSeek
DeepSeek	`deepseek`	`DEEPSEEK_API_KEY`	DeepSeek V3, R1 Reasoning
Together AI	`together`	`TOGETHER_API_KEY`	Open-source foundation models
Fireworks AI	`fireworks`	`FIREWORKS_API_KEY`	High-speed function calling
OpenRouter	`openrouter`	`OPENROUTER_API_KEY`	Multi-provider routing gateway
Mistral AI	`mistralai`	`MISTRAL_API_KEY`	Mistral Large, Codestral, Pixtral
NVIDIA NIM	`nvidia`	`NVIDIA_API_KEY`	Accelerated NIM endpoints
Perplexity	`perplexity`	`PPLX_API_KEY`	Search-augmented Sonar models
Cohere	`cohere`	`COHERE_API_KEY`	Command R / R+
IBM watsonx	`ibm`	`WATSONX_APIKEY`	Enterprise Granite and Llama
HuggingFace	`huggingface`	`HUGGINGFACEHUB_API_TOKEN`	Dedicated Inference Endpoints
LiteLLM	`litellm`	`LITELLM_API_KEY`	Unified proxy for internal gateways
xAI	`xai`	`XAI_API_KEY`	Grok 2 / Grok 3
Baseten	`baseten`	`BASETEN_API_KEY`	Custom deployed models
Ollama	`ollama`	Optional	Fully local offline inference

Extended thinking

OpsCode supports extended reasoning for models like Claude 3.7 Thinking, OpenAI o1/o3-mini, and Gemini Flash Thinking. In the TUI, reasoning tokens are displayed in a collapsible thinking panel.

Set the reasoning effort:

/effort high       # Choices: low, medium, high

Or in ~/.opscode/config.toml:

[model]
reasoning_effort = "high"

Switch models

In a session

/model                              # Open the model picker
/model anthropic:claude-opus-4-7    # Switch immediately
/fast                               # Jump to your configured fast model
/effort high                        # Adjust reasoning effort

From the CLI

ops -M openai:gpt-4.1                           # One-time override
ops --default-model anthropic:claude-opus-4-7    # Set persistent default
ops --clear-default-model                        # Clear the default

OpsCode remembers your 10 most recently used models for quick access in the /model picker.

Local models with Ollama

Run completely offline:

# Pull a model
ollama pull llama3.3

# Launch OpsCode with it
ops -M ollama:llama3.3

In ~/.opscode/config.toml:

[providers.ollama]
enabled = true
base_url = "http://localhost:11434"
models = ["llama3.3", "qwen2.5-coder:32b"]

Model format​

Supported providers​

Extended thinking​

Switch models​

In a session​

From the CLI​

Local models with Ollama​