diff --git a/crates/mofa-foundation/src/llm/agent.rs b/crates/mofa-foundation/src/llm/agent.rs
index 65f78a976..0c674ec09 100644
--- a/crates/mofa-foundation/src/llm/agent.rs
+++ b/crates/mofa-foundation/src/llm/agent.rs
@@ -29,9 +29,7 @@ use super::client::{ChatSession, LLMClient};
 use super::provider::{ChatStream, LLMProvider};
 use super::tool_executor::ToolExecutor;
 use super::types::{ChatMessage, LLMError, LLMResult, Tool};
-use crate::llm::{
-    AnthropicConfig, AnthropicProvider, GeminiConfig, GeminiProvider, OllamaConfig, OllamaProvider,
-};
+use crate::llm::{AnthropicConfig, AnthropicProvider, GeminiConfig, GeminiProvider};
 use crate::prompt;
 use futures::{Stream, StreamExt};
 use mofa_kernel::agent::AgentMetadata;
@@ -3663,19 +3661,19 @@ impl LLMAgentBuilder {
                 Arc::new(GeminiProvider::with_config(cfg))
             }
             "ollama" => {
-                let mut ollama_config = OllamaConfig::new();
-                ollama_config = ollama_config.with_base_url(&provider.api_base);
-                ollama_config = ollama_config.with_model(&agent.model_name);
+                let mut openai_config = OpenAIConfig::new("ollama")
+                    .with_base_url(&provider.api_base)
+                    .with_model(&agent.model_name);
 
                 if let Some(temp) = agent.temperature {
-                    ollama_config = ollama_config.with_temperature(temp);
+                    openai_config = openai_config.with_temperature(temp);
                 }
 
                 if let Some(max_tokens) = agent.max_completion_tokens {
-                    ollama_config = ollama_config.with_max_tokens(max_tokens as u32);
+                    openai_config = openai_config.with_max_tokens(max_tokens as u32);
                 }
 
-                Arc::new(OllamaProvider::with_config(ollama_config))
+                Arc::new(OpenAIProvider::with_config(openai_config))
             }
             other => {
                 return Err(LLMError::Other(format!(
diff --git a/crates/mofa-foundation/src/llm/mod.rs b/crates/mofa-foundation/src/llm/mod.rs
index 2308bdf8d..a80f52026 100644
--- a/crates/mofa-foundation/src/llm/mod.rs
+++ b/crates/mofa-foundation/src/llm/mod.rs
@@ -368,7 +368,7 @@ pub use anthropic::{AnthropicConfig, AnthropicProvider};
 // Re-export Google Gemini Provider
 pub use google::{GeminiConfig, GeminiProvider};
 // Re-export Ollama Provider
-pub use ollama::{OllamaConfig, OllamaProvider};
+pub use ollama::OllamaProvider;
 
 // Re-export 高级 API
 // Re-export Advanced API
diff --git a/crates/mofa-sdk/src/lib.rs b/crates/mofa-sdk/src/lib.rs
index d46cc2441..578f552a5 100644
--- a/crates/mofa-sdk/src/lib.rs
+++ b/crates/mofa-sdk/src/lib.rs
@@ -504,7 +504,7 @@ pub mod llm {
     pub use crate::llm_tools::ToolPluginExecutor;
     pub use mofa_foundation::llm::anthropic::{AnthropicConfig, AnthropicProvider};
     pub use mofa_foundation::llm::google::{GeminiConfig, GeminiProvider};
-    pub use mofa_foundation::llm::ollama::{OllamaConfig, OllamaProvider};
+    pub use mofa_foundation::llm::ollama::OllamaProvider;
     pub use mofa_foundation::llm::openai::{OpenAIConfig, OpenAIProvider};
     pub use mofa_foundation::llm::*;
 
@@ -542,13 +542,39 @@ pub mod llm {
         Ok(OpenAIProvider::with_config(config))
     }
 
-    /// Create an Ollama provider from environment variables (no API key required).
+    /// Create an Ollama-backed [`OpenAIProvider`] from environment variables (no API key required).
     ///
     /// Reads:
-    /// - `OLLAMA_BASE_URL`: base URL without `/v1` suffix, e.g. `http://localhost:11434` (optional)
-    /// - `OLLAMA_MODEL`: model name, e.g. `llama3` (optional)
-    pub fn ollama_from_env() -> Result<OllamaProvider, crate::llm::LLMError> {
-        Ok(crate::llm::OllamaProvider::from_env())
+    /// - `OLLAMA_HOST`: optional host (default `localhost:11434`); may be `host:port` or a full `http://` URL
+    /// - `OLLAMA_BASE_URL`: optional base URL (e.g. `http://localhost:11434`); used if `OLLAMA_HOST` is unset
+    /// - `OLLAMA_MODEL`: model name, e.g. `llama3` (optional, default `llama3`)
+    pub fn ollama_from_env() -> Result<OpenAIProvider, crate::llm::LLMError> {
+        let model = std::env::var("OLLAMA_MODEL").unwrap_or_else(|_| "llama3".to_string());
+        let mut cfg = OpenAIConfig::new("ollama");
+        let base_url = if let Ok(host) = std::env::var("OLLAMA_HOST") {
+            let h = host.trim();
+            let base = if h.starts_with("http://") || h.starts_with("https://") {
+                h.trim_end_matches('/').to_string()
+            } else {
+                format!("http://{}", h.trim_end_matches('/'))
+            };
+            if base.ends_with("/v1") {
+                base
+            } else {
+                format!("{}/v1", base)
+            }
+        } else if let Ok(base_url) = std::env::var("OLLAMA_BASE_URL") {
+            let base = base_url.trim().trim_end_matches('/');
+            if base.ends_with("/v1") {
+                base.to_string()
+            } else {
+                format!("{}/v1", base)
+            }
+        } else {
+            "http://localhost:11434/v1".to_string()
+        };
+        cfg = cfg.with_base_url(base_url).with_model(model);
+        Ok(OpenAIProvider::with_config(cfg))
     }
 }