diff --git a/fireSeqSearch_addon/main.js b/fireSeqSearch_addon/main.js
index 2e413b2..93df9f7 100644
--- a/fireSeqSearch_addon/main.js
+++ b/fireSeqSearch_addon/main.js
@@ -142,72 +142,120 @@ function parseRawList(rawSearchResult) {
     return hits;
 }
 
-function createTitleBarDom(count) {
-    const titleBar = createElementWithText("div");
-    titleBar.classList.add('fireSeqSearchTitleBar');
-    const hitCount = `<span>We found <b>${count.toString()}</b> results in your logseq notebook</span>`;
-    titleBar.insertAdjacentHTML("afterbegin",hitCount);
-
-    function setSummaryState(cl, state) {
-        let prop = 'none';
-        if (state) { prop = ''; }
-        for (const el of document.querySelectorAll(cl)) {
-            el.style.display=prop;
+async function processLlmSummary(serverInfo, parsedSearchResult, fireDom) {
+
+    const doneListApi = "http://127.0.0.1:3030/llm_done_list";
+    let list = await fetch(doneListApi);
+    list = await list.text();
+    list = JSON.parse(list);
+
+    const findByTitle = function(title) {
+        const ul = fireDom.querySelector( ".fireSeqSearchHitList" );
+        if (ul === null)    return null;
+        for (const child of ul.children) {
+            const liTitle = child.firstChild.text;
+            if (title === liTitle) {
+                return child;
+            }
         }
-    }
-
-    let btn = document.createElement("button");
-    btn.classList.add("hideSummary");
-    let text = document.createTextNode("Hide Summary");
-    btn.appendChild(text);
-    btn.onclick = function () {
-        setSummaryState(".fireSeqSearchHitSummary", false);
-        setSummaryState(".fireSeqSearchLlmSummary", false);
-    };
-    titleBar.appendChild(btn);
-
-    btn = document.createElement("button");
-    btn.classList.add("showSummary");
-    text = document.createTextNode("Summary");
-    btn.appendChild(text);
-    btn.onclick = function () {
-        setSummaryState(".fireSeqSearchHitSummary", true);
-        setSummaryState(".fireSeqSearchLlmSummary", false);
+        return null;
     };
-    titleBar.appendChild(btn);
-
+    const setLlmResult = function (title, llmSummary) {
+        const targetRow = findByTitle(title);
+        if (targetRow === null) {
+            consoleLogForDebug("Error! Can't find dom for ", title);
+            return;
+        }
+        if (targetRow.querySelector( ".fireSeqSearchLlmSummary" ) != null) {
+            consoleLogForDebug("Skip. We have the summary for ", title);
+            return;
+        }
 
-    btn = document.createElement("button");
-    btn.classList.add("showLlm");
-    text = document.createTextNode("LLM");
-    btn.appendChild(text);
-    btn.onclick = function () {
-        setSummaryState(".fireSeqSearchHitSummary", false);
-        setSummaryState(".fireSeqSearchLlmSummary", true);
+        const summary = createElementWithText("span", "");
+        summary.innerHTML = llmSummary;
+        summary.classList.add('fireSeqSearchLlmSummary');
+        targetRow.appendChild(summary);
     };
-    titleBar.appendChild(btn);
+    for (const record of parsedSearchResult) {
+        const title = record.title;
+        if (!list.includes(title)) {
+            consoleLogForDebug("Not ready, skip" + title);
+            continue;
+        }
+        // TODO remove hard code port
+        const llm_api = "http://127.0.0.1:3030/summarize/" + title;
+        let sum = await fetch(llm_api);
+        sum = await sum.text();
+        setLlmResult(title, sum);
+    }
+}
 
 
-    return titleBar;
-}
-function createFireSeqDom(count) {
+function createFireSeqDom(serverInfo, parsedSearchResult) {
+    const count = parsedSearchResult.length;
     const div = document.createElement("div");
     div.setAttribute("id", fireSeqSearchDomId);
-    const bar = createTitleBarDom(count);
+
+    const createTitleBarDom = function () {
+        const titleBar = createElementWithText("div");
+        titleBar.classList.add('fireSeqSearchTitleBar');
+        const hitCount = `<span>We found <b>${count.toString()}</b> results in your logseq notebook</span>`;
+        titleBar.insertAdjacentHTML("afterbegin",hitCount);
+
+        function setSummaryState(cl, state) {
+            let prop = 'none';
+            if (state) { prop = ''; }
+            for (const el of document.querySelectorAll(cl)) {
+                el.style.display=prop;
+            }
+        }
+        let btn = document.createElement("button");
+        btn.classList.add("hideSummary");
+        let text = document.createTextNode("Hide Summary");
+        btn.appendChild(text);
+        btn.onclick = function () {
+            setSummaryState(".fireSeqSearchHitSummary", false);
+            setSummaryState(".fireSeqSearchLlmSummary", false);
+        };
+        titleBar.appendChild(btn);
+
+        btn = document.createElement("button");
+        btn.classList.add("showSummary");
+        text = document.createTextNode("Summary");
+        btn.appendChild(text);
+        btn.onclick = function () {
+            setSummaryState(".fireSeqSearchHitSummary", true);
+            setSummaryState(".fireSeqSearchLlmSummary", false);
+        };
+        titleBar.appendChild(btn);
+
+        btn = document.createElement("button");
+        btn.classList.add("showLlm");
+        text = document.createTextNode("LLM");
+        btn.appendChild(text);
+        btn.onclick = function () {
+            setSummaryState(".fireSeqSearchHitSummary", false);
+            setSummaryState(".fireSeqSearchLlmSummary", true);
+            processLlmSummary(serverInfo, parsedSearchResult, div);
+        };
+        titleBar.appendChild(btn);
+        return titleBar;
+    };
+    const bar = createTitleBarDom();
     div.appendChild(bar);
     return div;
 }
 
 async function appendResultToSearchResult(serverInfo, parsedSearchResult, dom) {
     const firefoxExtensionUserOption = await checkUserOptions();
-
     consoleLogForDebug('Loaded user option: ' + JSON.stringify(firefoxExtensionUserOption));
 
-
     function buildListItems(parsedSearchResult) {
         const hitList = document.createElement("ul");
+        hitList.classList.add('fireSeqSearchHitList');
         for (const record of parsedSearchResult) {
             const li =  createElementWithText("li", "");
+            li.classList.add('fireSeqSearchHitListItem');
             if (firefoxExtensionUserOption.ShowScore) {
                 const score = createElementWithText("span", String(record.score));
                 li.appendChild(score);
@@ -252,18 +300,6 @@ async function appendResultToSearchResult(serverInfo, parsedSearchResult, dom) {
     insertDivToWebpage(dom);
 }
 
-async function processLlmSummary(serverInfo, parsedSearchResult, dom) {
-    for (const record of parsedSearchResult) {
-        // TODO remove hard code port
-        const llm_api = "http://127.0.0.1:3030/summarize/" + record.title;
-        console.log("llm called");
-        console.log(record.title);
-        const response = await fetch(llm_api);
-        const text = await response.text();
-        console.log(text);
-    }
-}
-
 async function mainProcess(fetchResultArray) {
     consoleLogForDebug("main process");
 
@@ -272,18 +308,10 @@ async function mainProcess(fetchResultArray) {
     consoleLogForDebug(serverInfo);
     const parsedSearchResult = parseRawList(rawSearchResult);
 
-    console.log("in main");
-    console.log(rawSearchResult);
-    console.log(parsedSearchResult);
-
-    const fireDom = createFireSeqDom(parsedSearchResult.length);
+    const fireDom = createFireSeqDom(serverInfo, parsedSearchResult);
 
     appendResultToSearchResult(serverInfo, parsedSearchResult, fireDom);
 
-    if (serverInfo.llm_enabled) {
-        consoleLogForDebug("llm");
-        processLlmSummary(serverInfo, parsedSearchResult, fireDom);
-    }
 }
 
 
@@ -318,7 +346,6 @@ function getSearchParameterFromCurrentPage() {
 (function() {
     const searchParameter = getSearchParameterFromCurrentPage();
 
-
     addGlobalStyle(fireSeqSearchScriptCSS);
 
     //https://gomakethings.com/waiting-for-multiple-all-api-responses-to-complete-with-the-vanilla-js-promise.all-method/
@@ -328,9 +355,7 @@ function getSearchParameterFromCurrentPage() {
     ]).then(function (responses) {
         return Promise.all(responses.map(function (response) {return response.json();}));
     }).then(function (data) {
-        //consoleLogForDebug(data);
         mainProcess(data);
-        //return appendResultToSearchResult(data);
     }).then((_e) => {
         const highlightedItems = document.querySelectorAll('.fireSeqSearchHighlight');
         consoleLogForDebug(highlightedItems);
diff --git a/fire_seq_search_server/src/http_client/endpoints.rs b/fire_seq_search_server/src/http_client/endpoints.rs
index 6d78417..58f0899 100644
--- a/fire_seq_search_server/src/http_client/endpoints.rs
+++ b/fire_seq_search_server/src/http_client/endpoints.rs
@@ -1,5 +1,5 @@
 use std::sync::Arc;
-use log::debug;
+use log::{debug, info};
 
 use crate::query_engine::{QueryEngine, ServerInformation};
 use axum::Json;
@@ -30,6 +30,15 @@ pub async fn summarize(
     Html(r.await)
 }
 
+pub async fn get_llm_done_list(
+    State(engine_arc): State<Arc<QueryEngine>>
+    ) -> Html<String>{
+
+    info!("get list endpoint called");
+    let r = engine_arc.get_llm_done_list();
+    Html(r.await)
+}
+
 pub async fn generate_word_cloud(State(engine_arc): State<Arc<QueryEngine>>)
                                                     -> Html<String> {
     let div_id = "fireSeqSearchWordcloudRawJson";
diff --git a/fire_seq_search_server/src/local_llm/mod.rs b/fire_seq_search_server/src/local_llm/mod.rs
index 43f69a2..4a7c769 100644
--- a/fire_seq_search_server/src/local_llm/mod.rs
+++ b/fire_seq_search_server/src/local_llm/mod.rs
@@ -46,6 +46,13 @@ pub struct Usage {
     pub total_tokens: i64,
 }
 
+#[derive(Default, Debug, Clone, PartialEq, Serialize, Deserialize)]
+pub struct HealthCheck {
+    pub slots_idle: i64,
+    pub slots_processing: i64,
+    pub status: String,
+}
+
 // End genereated
 
 const LLM_SERVER_PORT: &str = "8081"; // TODO Remove this magic number
@@ -165,7 +172,6 @@ impl LlmEngine {
 
 impl LlmEngine{
     pub async fn summarize(&self, full_text: &str) -> String {
-        info!("summarize called");
         //http://localhost:8080/completion
         let ep = self.endpoint.to_owned() + "/v1/chat/completions";
         let data = Self::build_data(full_text);
@@ -175,17 +181,12 @@ impl LlmEngine{
             .send()
             .await
             .unwrap();
-        //info!(" response {:?}", &res);
         let content = res.text().await.unwrap();
-        //info!(" text {:?}", &content);
         let parsed: LlamaResponse = serde_json::from_str(&content).unwrap();
-        //info!(" parsed {:?}", &parsed);
         let v = parsed.choices;
         let v0 = v.into_iter().next().unwrap();
         v0.message.content
-
-
-            //TODO remove unwrap
+        //TODO remove unwrap
     }
 
     pub async fn post_summarize_job(&self, doc: DocData) {
@@ -196,13 +197,19 @@ impl LlmEngine{
     }
 
     pub async fn call_llm_engine(&self) {
+
+        let health = self.health().await.unwrap();
+        if health.slots_idle == 0 {
+            info!("No valid slot, continue");
+            return;
+        }
+
         let mut next_job: Option<DocData> = None;
 
         let mut jcache = self.job_cache.lock().await;//.unwrap();
         next_job = jcache.job_queue.pop_front();
         drop(jcache);
 
-
         let doc = match next_job {
             Some(x) => x,
             None => { return; },
@@ -216,14 +223,14 @@ impl LlmEngine{
         }
         drop(jcache);
 
+        info!("Start summarize job:  {}", &title);
         let summarize_result = self.summarize(&doc.body).await;
+        info!("Finished summarize job:  {}", &title);
 
         let mut jcache = self.job_cache.lock().await;//.unwrap();
         next_job = jcache.job_queue.pop_front();
-        info!("get summarize result {}", &title);
         jcache.done_job.insert(title, summarize_result);
         drop(jcache);
-
     }
 
     pub async fn quick_fetch(&self, title: &str) -> Option<String> {
@@ -231,16 +238,24 @@ impl LlmEngine{
         return jcache.done_job.get(title).cloned();
     }
 
-    pub async fn health(&self) -> Result<(), Box<dyn std::error::Error>>  {
-        info!("Calling health check");
-        let resp = reqwest::get(self.endpoint.to_owned() + "/health")
-                .await?
-                .headers().to_owned()
-                //.status()
-                //.text().await?
-                ;
-        info!("Health check: {:#?}", resp);
-        Ok(())
+    pub async fn get_llm_done_list(&self) -> Vec<String> {
+        let mut r = Vec::new();
+        let jcache = self.job_cache.lock().await;
+        for (title, _text) in &jcache.done_job {
+            info!("already done : {}", &title);
+            r.push(title.to_owned());
+        }
+        return r;
+    }
+
+    pub async fn health(&self) -> Result<HealthCheck, Box<dyn std::error::Error>>  {
+        let res = self.client.get(self.endpoint.to_owned() + "/health")
+            .send()
+            .await
+            .unwrap();
+        let content = res.text().await.unwrap();
+        let parsed: HealthCheck = serde_json::from_str(&content).unwrap();
+        Ok(parsed)
     }
 }
 
diff --git a/fire_seq_search_server/src/main.rs b/fire_seq_search_server/src/main.rs
index 800b644..a9c1ea4 100644
--- a/fire_seq_search_server/src/main.rs
+++ b/fire_seq_search_server/src/main.rs
@@ -78,12 +78,10 @@ async fn main() {
         engine.llm = Some(llm_arc);
 
         let poll_handle = tokio::spawn( async move {
-            info!("inside main loop");
             loop {
                 llm_poll.call_llm_engine().await;
                 let wait_llm = tokio::time::Duration::from_millis(500);
                 tokio::time::sleep(wait_llm).await;
-                info!("main loop: poll again");
             }
         });
 //        poll_handle.await;
@@ -97,6 +95,7 @@ async fn main() {
         .route("/server_info", get(endpoints::get_server_info))
         .route("/wordcloud", get(endpoints::generate_word_cloud))
         .route("/summarize/:title", get(endpoints::summarize))
+        .route("/llm_done_list", get(endpoints::get_llm_done_list))
         .with_state(engine_arc.clone());
 
     let listener = tokio::net::TcpListener::bind(&engine_arc.server_info.host)
diff --git a/fire_seq_search_server/src/query_engine/mod.rs b/fire_seq_search_server/src/query_engine/mod.rs
index 62fe961..cb7c021 100644
--- a/fire_seq_search_server/src/query_engine/mod.rs
+++ b/fire_seq_search_server/src/query_engine/mod.rs
@@ -166,6 +166,16 @@ impl QueryEngine {
             "LLM turned off".to_owned()
         }
     }
+    pub async fn get_llm_done_list(&self) -> String {
+        if cfg!(feature="llm") {
+            let llm = self.llm.as_ref().unwrap();
+            let result = &llm.get_llm_done_list().await;
+            let json = serde_json::to_string(&result).unwrap();
+            return json;
+        } else {
+            "LLM turned off".to_owned()
+        }
+    }
 }
 
 fn term_preprocess(term:String) -> String {