feat: Add LiteLLM provider with automatic prompt caching support

crates/goose-llm/tests/providers_complete.rs

@@ -157,7 +157,7 @@ impl ProviderTester {
             .content
             .iter()
             .filter_map(|message| message.as_tool_request())
-            .last()
+            .next_back()
             .expect("got tool request")
             .id;
 

crates/goose/src/model.rs

@@ -275,6 +275,7 @@ mod tests {
     }
 
     #[test]
+    #[serial_test::serial]
     fn test_model_config_context_limit_env_vars() {
         use temp_env::with_vars;
 

crates/goose/src/providers/base.rs

@@ -41,6 +41,8 @@ pub struct ModelInfo {
     pub output_token_cost: Option<f64>,
     /// Currency for the costs (default: "$")
     pub currency: Option<String>,
+    /// Whether this model supports cache control
+    pub supports_cache_control: Option<bool>,
 }
 
 impl ModelInfo {
@@ -52,6 +54,7 @@ impl ModelInfo {
             input_token_cost: None,
             output_token_cost: None,
             currency: None,
+            supports_cache_control: None,
         }
     }
 
@@ -68,6 +71,7 @@ impl ModelInfo {
             input_token_cost: Some(input_cost),
             output_token_cost: Some(output_cost),
             currency: Some("$".to_string()),
+            supports_cache_control: None,
         }
     }
 }
@@ -115,6 +119,7 @@ impl ProviderMetadata {
                     input_token_cost: None,
                     output_token_cost: None,
                     currency: None,
+                    supports_cache_control: None,
                 })
                 .collect(),
             model_doc_link: model_doc_link.to_string(),
@@ -290,6 +295,11 @@ pub trait Provider: Send + Sync {
         false
     }
 
+    /// Check if this provider supports cache control
+    fn supports_cache_control(&self) -> bool {
+        false
+    }
+
     /// Create embeddings if supported. Default implementation returns an error.
     async fn create_embeddings(&self, _texts: Vec<String>) -> Result<Vec<Vec<f32>>, ProviderError> {
         Err(ProviderError::ExecutionError(
@@ -435,6 +445,7 @@ mod tests {
             input_token_cost: None,
             output_token_cost: None,
             currency: None,
+            supports_cache_control: None,
         };
         assert_eq!(info.context_limit, 1000);
 
@@ -445,6 +456,7 @@ mod tests {
             input_token_cost: None,
             output_token_cost: None,
             currency: None,
+            supports_cache_control: None,
         };
         assert_eq!(info, info2);
 
@@ -455,6 +467,7 @@ mod tests {
             input_token_cost: None,
             output_token_cost: None,
             currency: None,
+            supports_cache_control: None,
         };
         assert_ne!(info, info3);
     }

crates/goose/src/providers/factory.rs

@@ -12,6 +12,7 @@ use super::{
     google::GoogleProvider,
     groq::GroqProvider,
     lead_worker::LeadWorkerProvider,
+    litellm::LiteLLMProvider,
     ollama::OllamaProvider,
     openai::OpenAiProvider,
     openrouter::OpenRouterProvider,
@@ -50,6 +51,7 @@ pub fn providers() -> Vec<ProviderMetadata> {
         // GithubCopilotProvider::metadata(),
         GoogleProvider::metadata(),
         GroqProvider::metadata(),
+        LiteLLMProvider::metadata(),
         OllamaProvider::metadata(),
         OpenAiProvider::metadata(),
         OpenRouterProvider::metadata(),
@@ -158,6 +160,7 @@ fn create_provider(name: &str, model: ModelConfig) -> Result<Arc<dyn Provider>>
         "databricks" => Ok(Arc::new(DatabricksProvider::from_env(model)?)),
         "gemini-cli" => Ok(Arc::new(GeminiCliProvider::from_env(model)?)),
         "groq" => Ok(Arc::new(GroqProvider::from_env(model)?)),
+        "litellm" => Ok(Arc::new(LiteLLMProvider::from_env(model)?)),
         "ollama" => Ok(Arc::new(OllamaProvider::from_env(model)?)),
         "openrouter" => Ok(Arc::new(OpenRouterProvider::from_env(model)?)),
         "gcp_vertex_ai" => Ok(Arc::new(GcpVertexAIProvider::from_env(model)?)),