Models and Providers

The harness accepts model strings in provider:model format.

Agno Deep Agent keeps a few special cases for OpenAI Responses and Ollama, then delegates the rest to Agno’s native model-string resolver. That means the same format works in the CLI, /model, --compression-model, and the Python SDK.

If the string has no provider, for example gpt-5.2, the harness uses OpenAIResponses.

Common Providers

Provider	Example	Requirement
`openai-responses`	`openai-responses:gpt-5.2`	`OPENAI_API_KEY`
`openai`	`openai:gpt-4o`	`OPENAI_API_KEY`
`anthropic`	`anthropic:claude-sonnet-4-5`	`ANTHROPIC_API_KEY`
`google`	`google:gemini-3-flash-preview`	`GOOGLE_API_KEY` or Vertex AI
`groq`	`groq:llama-3.3-70b-versatile`	`GROQ_API_KEY`
`mistral`	`mistral:mistral-large-latest`	`MISTRAL_API_KEY`
`deepseek`	`deepseek:deepseek-chat`	`DEEPSEEK_API_KEY`
`xai`	`xai:grok-3`	`XAI_API_KEY`
`perplexity`	`perplexity:sonar-pro`	`PERPLEXITY_API_KEY`
`cohere`	`cohere:command-a-03-2025`	`CO_API_KEY`
`together`	`together:meta-llama/Llama-3-70b-chat-hf`	`TOGETHER_API_KEY`
`fireworks`	`fireworks:accounts/fireworks/models/llama-v3p1-70b-instruct`	`FIREWORKS_API_KEY`
`openrouter`	`openrouter:anthropic/claude-3.5-sonnet`	`OPENROUTER_API_KEY`
`litellm`	`litellm:gpt-4o`	Provider-specific key
`azure-ai-foundry`	`azure-ai-foundry:gpt-4o`	Azure AI Foundry credentials
`ollama`	`ollama:gemma4:e4b`	Local Ollama
`ollama-responses`	`ollama-responses:gpt-oss:20b`	Local Ollama Responses
`ollama-cloud`	`ollama-cloud:devstral-2`	`OLLAMA_API_KEY`

Inside the interactive CLI, use:

/models
/m anthropic:claude-sonnet-4-5

The /model ... command saves the selection in .deep-agent/config.json for the active workspace. Later runs use that saved model until another /model ... command changes it. The --model ... flag remains a one-run override, and DEEP_AGENT_MODEL is used when no workspace model has been saved.

Agno supports additional native, local, cloud, gateway, and aggregator providers. See the official Agno model-string docs and Agno provider index for the current list and provider-specific setup.

Provider Dependencies

The base install includes the default OpenAI and Ollama SDKs. Some providers need their own SDK package in the active environment, for example:

pip install anthropic google-genai groq mistralai cohere together litellm

If a provider backend is missing, the CLI prints a targeted install hint when one is known, or points you back to the Agno provider docs.

Local Ollama By Default

In Agno, the Ollama model class can use Ollama Cloud when OLLAMA_API_KEY is present and no host is provided.

To avoid surprises, Agno Deep Agent defines:

ollama:* -> http://localhost:11434

Cloud must be explicit:

ollama-cloud:devstral-2

Check Local Models

Use:

ollama list

The model name passed to the harness must match the NAME column exactly.

Example:

gemma4:e4b

should be used as:

agno-deep-agent --model ollama:gemma4:e4b "Say hello"

Custom Host

agno-deep-agent --model ollama:gemma4:e4b \
  --ollama-host http://192.168.1.10:11434 \
  "Analyze the project"

When To Use `ollama-responses`

Use ollama-responses:* when you want Ollama’s OpenAI-compatible /v1/responses endpoint. It requires an Ollama version that supports that endpoint.