diff --git a/pyproject.toml b/pyproject.toml
index b9682d2..b523f47 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -121,6 +121,9 @@ distributed = [
   "pyspark<4.1",
   "ray==2.48",
 ]
+litellm = [
+  "litellm>=1.80.0,<1.87.0",
+]
 
 [project.scripts]
 timecopilot = "timecopilot._cli:main"
diff --git a/timecopilot/agent.py b/timecopilot/agent.py
index f24e053..0e14203 100644
--- a/timecopilot/agent.py
+++ b/timecopilot/agent.py
@@ -8,6 +8,7 @@
 from pydantic_ai import Agent, ModelRetry, RunContext
 from pydantic_ai.agent import AgentRunResult
 from pydantic_ai.models import Model
+from pydantic_ai.models.openai import OpenAIModel
 from rich.console import Console
 from rich.panel import Panel
 from rich.table import Table
@@ -34,6 +35,35 @@
 
 from .forecaster import Forecaster, TimeCopilotForecaster
 from .models.adapters.sktime import SKTimeAdapter
+
+
+def _resolve_llm(llm: str | Model) -> str | Model:
+    """Resolve LLM string, adding LiteLLM support for 100+ providers.
+
+    If the string starts with 'litellm:', creates a pydantic-ai OpenAI model
+    backed by litellm SDK (no proxy needed). LiteLLM reads provider API keys
+    from environment variables (ANTHROPIC_API_KEY, OPENAI_API_KEY, etc.).
+
+    Examples:
+        --llm litellm:anthropic/claude-sonnet-4-6
+        --llm litellm:groq/llama-3.3-70b-versatile
+        --llm litellm:deepseek/deepseek-chat
+    """
+    if not (isinstance(llm, str) and llm.startswith("litellm:")):
+        return llm
+
+    model_name = llm[len("litellm:"):]
+    try:
+        import litellm
+
+        litellm.drop_params = True
+        client = litellm.AsyncOpenAI()
+    except ImportError:
+        raise ImportError(
+            "litellm package required for litellm: models. "
+            "Install with: pip install 'timecopilot[litellm]'"
+        )
+    return OpenAIModel(model_name, openai_client=client)
 from .models.prophet import Prophet
 from .models.stats import (
     ADIDA,
@@ -541,7 +571,7 @@ def __init__(
                 "model is not allowed to be passed as a keyword argument"
                 "use `llm` instead"
             )
-        self.llm = llm
+        self.llm = _resolve_llm(llm)
 
         self.forecasting_agent = Agent(
             deps_type=ExperimentDataset,