agentforge-litellm

agentforge-litellm

LiteLLM router-based LLM provider for AgentForge.

Wraps LiteLLM's unified interface so a single AgentForge agent can route to 100+ underlying providers (OpenAI, Anthropic, Bedrock, Vertex, Azure, Mistral, Groq, Fireworks, Together, ...) by changing the model string — no per-provider sister package needed.

When to use this vs. a native sister package:

• Prefer native (agentforge-anthropic, agentforge-openai, ...) for production paths where you want first-class capability surface (caching, thinking, streaming with full type-checked stream events).
• Use LiteLLM as a unified gateway, for tail providers AgentForge doesn't ship native packages for yet, or to experiment with new models behind a single config.

Install

pip install "agentforge-litellm[litellm]"

Use

from agentforge import Agent

# Model string is "litellm:<litellm-model-string>". The second
# half is whatever LiteLLM accepts.
agent = Agent(model="litellm:gpt-4o-mini")
agent = Agent(model="litellm:anthropic/claude-sonnet-4-7")
agent = Agent(model="litellm:bedrock/anthropic.claude-sonnet-4-7")

Capabilities

{"tools"} only. LiteLLM normalises tools across underlying providers but doesn't expose a uniform caching / thinking / vision / streaming surface — those vary per backend. If you need those, use the matching native sister package.

License

Apache-2.0. See LICENSE.