diff --git a/crates/erp-ai/src/provider/mod.rs b/crates/erp-ai/src/provider/mod.rs
index c5d7aeb..fd68e14 100644
--- a/crates/erp-ai/src/provider/mod.rs
+++ b/crates/erp-ai/src/provider/mod.rs
@@ -1,4 +1,5 @@
 pub mod claude;
+pub mod openai;
 pub mod registry;
 
 use async_trait::async_trait;
diff --git a/crates/erp-ai/src/provider/openai.rs b/crates/erp-ai/src/provider/openai.rs
new file mode 100644
index 0000000..7eb48de
--- /dev/null
+++ b/crates/erp-ai/src/provider/openai.rs
@@ -0,0 +1,309 @@
+use async_stream::stream;
+use async_trait::async_trait;
+use futures::{Stream, StreamExt};
+use reqwest::Client;
+use serde::{Deserialize, Serialize};
+use std::pin::Pin;
+
+use super::AiProvider;
+use crate::dto::GenerateRequest;
+use crate::error::{AiError, AiResult};
+
+#[derive(Debug, Clone)]
+pub struct OpenAIProvider {
+    client: Client,
+    api_key: String,
+    base_url: String,
+    default_model: String,
+}
+
+impl OpenAIProvider {
+    pub fn new(api_key: String, base_url: String, default_model: String) -> Self {
+        Self {
+            client: Client::new(),
+            api_key,
+            base_url,
+            default_model,
+        }
+    }
+}
+
+#[derive(Serialize)]
+struct ChatRequest {
+    model: String,
+    max_tokens: u32,
+    temperature: f32,
+    messages: Vec<ChatMessage>,
+    stream: bool,
+}
+
+#[derive(Serialize)]
+struct ChatMessage {
+    role: String,
+    content: String,
+}
+
+#[derive(Deserialize)]
+struct ChatResponse {
+    choices: Vec<ChatChoice>,
+    usage: Option<ChatUsage>,
+}
+
+#[derive(Deserialize)]
+struct ChatChoice {
+    message: ChatMessageResp,
+}
+
+#[derive(Deserialize)]
+struct ChatMessageResp {
+    content: Option<String>,
+}
+
+#[derive(Deserialize)]
+struct ChatUsage {
+    prompt_tokens: u32,
+    completion_tokens: u32,
+}
+
+#[derive(Deserialize)]
+struct StreamChunk {
+    choices: Vec<StreamChoice>,
+}
+
+#[derive(Deserialize)]
+struct StreamChoice {
+    delta: StreamDelta,
+}
+
+#[derive(Deserialize)]
+struct StreamDelta {
+    content: Option<String>,
+}
+
+#[async_trait]
+impl AiProvider for OpenAIProvider {
+    async fn stream_generate(
+        &self,
+        req: GenerateRequest,
+    ) -> AiResult<Pin<Box<dyn Stream<Item = AiResult<String>> + Send>>> {
+        let model = if req.model.is_empty() {
+            self.default_model.clone()
+        } else {
+            req.model
+        };
+
+        let chat_req = ChatRequest {
+            model,
+            max_tokens: req.max_tokens,
+            temperature: req.temperature,
+            messages: vec![
+                ChatMessage {
+                    role: "system".into(),
+                    content: req.system_prompt,
+                },
+                ChatMessage {
+                    role: "user".into(),
+                    content: req.user_prompt,
+                },
+            ],
+            stream: true,
+        };
+
+        let response = self
+            .client
+            .post(format!("{}/v1/chat/completions", self.base_url))
+            .header("Authorization", format!("Bearer {}", self.api_key))
+            .header("content-type", "application/json")
+            .json(&chat_req)
+            .send()
+            .await
+            .map_err(|e| AiError::ProviderError(format!("OpenAI API 请求失败: {e}")))?;
+
+        if !response.status().is_success() {
+            let status = response.status();
+            let body = response.text().await.unwrap_or_default();
+            return Err(AiError::ProviderError(format!(
+                "OpenAI API 错误 {status}: {body}"
+            )));
+        }
+
+        let s = Box::pin(stream! {
+            let mut stream = response.bytes_stream();
+            while let Some(chunk_result) = stream.next().await {
+                let bytes = match chunk_result {
+                    Ok(b) => b,
+                    Err(e) => {
+                        yield Err(AiError::ProviderError(format!("流读取错误: {e}")));
+                        break;
+                    }
+                };
+
+                let text = String::from_utf8_lossy(&bytes);
+                for line in text.lines() {
+                    if let Some(data) = line.strip_prefix("data: ") {
+                        if data == "[DONE]" {
+                            return;
+                        }
+                        if let Ok(chunk) = serde_json::from_str::<StreamChunk>(data) {
+                            for choice in chunk.choices {
+                                if let Some(content) = choice.delta.content {
+                                    yield Ok(content);
+                                }
+                            }
+                        }
+                    }
+                }
+            }
+        });
+
+        Ok(s)
+    }
+
+    async fn generate(&self, req: GenerateRequest) -> AiResult<crate::dto::GenerateResponse> {
+        let start = std::time::Instant::now();
+
+        let model = if req.model.is_empty() {
+            self.default_model.clone()
+        } else {
+            req.model.clone()
+        };
+
+        let chat_req = ChatRequest {
+            model: model.clone(),
+            max_tokens: req.max_tokens,
+            temperature: req.temperature,
+            messages: vec![
+                ChatMessage {
+                    role: "system".into(),
+                    content: req.system_prompt,
+                },
+                ChatMessage {
+                    role: "user".into(),
+                    content: req.user_prompt,
+                },
+            ],
+            stream: false,
+        };
+
+        let resp = self
+            .client
+            .post(format!("{}/v1/chat/completions", self.base_url))
+            .header("Authorization", format!("Bearer {}", self.api_key))
+            .header("content-type", "application/json")
+            .json(&chat_req)
+            .send()
+            .await
+            .map_err(|e| AiError::ProviderError(e.to_string()))?;
+
+        let status = resp.status();
+        let body = resp
+            .text()
+            .await
+            .map_err(|e| AiError::ProviderError(e.to_string()))?;
+
+        if !status.is_success() {
+            return Err(AiError::ProviderError(format!(
+                "OpenAI {status}: {body}"
+            )));
+        }
+
+        let parsed: ChatResponse = serde_json::from_str(&body)
+            .map_err(|e| AiError::ProviderError(format!("解析响应失败: {e}")))?;
+
+        let content = parsed
+            .choices
+            .first()
+            .and_then(|c| c.message.content.clone())
+            .unwrap_or_default();
+
+        let (input_tokens, output_tokens) = parsed
+            .usage
+            .map(|u| (u.prompt_tokens, u.completion_tokens))
+            .unwrap_or((0, 0));
+
+        Ok(crate::dto::GenerateResponse {
+            content,
+            model,
+            input_tokens,
+            output_tokens,
+            duration_ms: start.elapsed().as_millis() as u64,
+        })
+    }
+
+    fn name(&self) -> &str {
+        "openai"
+    }
+
+    async fn health_check(&self) -> AiResult<bool> {
+        let resp = self
+            .client
+            .get(format!("{}/v1/models", self.base_url))
+            .header("Authorization", format!("Bearer {}", self.api_key))
+            .send()
+            .await;
+
+        match resp {
+            Ok(r) => Ok(r.status().is_success()),
+            Err(_) => Ok(false),
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn openai_provider_construction() {
+        let provider = OpenAIProvider::new(
+            "sk-test".into(),
+            "https://api.openai.com".into(),
+            "gpt-4o".into(),
+        );
+        assert_eq!(provider.name(), "openai");
+        assert_eq!(provider.default_model, "gpt-4o");
+    }
+
+    #[test]
+    fn chat_request_serialization() {
+        let req = ChatRequest {
+            model: "gpt-4o".into(),
+            max_tokens: 1024,
+            temperature: 0.7,
+            messages: vec![
+                ChatMessage {
+                    role: "system".into(),
+                    content: "你是助手".into(),
+                },
+                ChatMessage {
+                    role: "user".into(),
+                    content: "你好".into(),
+                },
+            ],
+            stream: false,
+        };
+        let json = serde_json::to_value(&req).unwrap();
+        assert_eq!(json["model"], "gpt-4o");
+        assert_eq!(json["messages"].as_array().unwrap().len(), 2);
+    }
+
+    #[test]
+    fn chat_response_deserialization() {
+        let json = r#"{
+            "choices": [{"message": {"role": "assistant", "content": "你好！"}}],
+            "usage": {"prompt_tokens": 10, "completion_tokens": 5}
+        }"#;
+        let resp: ChatResponse = serde_json::from_str(json).unwrap();
+        assert_eq!(resp.choices[0].message.content.as_deref(), Some("你好！"));
+        assert_eq!(resp.usage.unwrap().prompt_tokens, 10);
+    }
+
+    #[test]
+    fn stream_chunk_deserialization() {
+        let json = r#"{
+            "choices": [{"delta": {"content": "Hello"}, "index": 0}]
+        }"#;
+        let chunk: StreamChunk = serde_json::from_str(json).unwrap();
+        assert_eq!(chunk.choices[0].delta.content.as_deref(), Some("Hello"));
+    }
+}