diff --git a/src-tauri/Cargo.toml b/src-tauri/Cargo.toml
index 0aeb6d2..4e0262e 100644
--- a/src-tauri/Cargo.toml
+++ b/src-tauri/Cargo.toml
@@ -31,6 +31,7 @@ tauri-plugin-log = "2"
 tauri-plugin-notification = "2"
 tauri-plugin-shell = "2"
 tauri-plugin-fs = "2"
+reqwest = { version = "0.12", default-features = false, features = ["json", "rustls-tls", "stream"] }
 
 [dev-dependencies]
 tempfile = "3"
diff --git a/src-tauri/src/commands.rs b/src-tauri/src/commands.rs
index cf96e71..42ad589 100644
--- a/src-tauri/src/commands.rs
+++ b/src-tauri/src/commands.rs
@@ -1,6 +1,6 @@
 use crate::config::{MigrationConfig, NotificationLevel, RobinConfig, SourceOs};
 use std::sync::Mutex;
-use tauri::State;
+use tauri::{Emitter, State};
 
 pub struct AppState {
     pub config: Mutex<RobinConfig>,
@@ -79,3 +79,50 @@ pub fn panel_opened() {
 pub fn panel_closed() {
     crate::notify::set_panel_open(false);
 }
+
+#[tauri::command]
+pub async fn chat(
+    message: String,
+    state: State<'_, AppState>,
+    app_handle: tauri::AppHandle,
+) -> Result<(), String> {
+    let (base_url, model, source_os, distro) = {
+        let cfg = state.config.lock().map_err(|e| e.to_string())?;
+        let base_url = cfg.ollama.base_url.clone();
+        let model = cfg.ollama.model.clone();
+        let (source_os, distro) = match cfg.migration.as_ref() {
+            Some(m) => {
+                let os = match m.source_os {
+                    crate::config::SourceOs::Macos => "macOS",
+                    crate::config::SourceOs::Windows => "Windows",
+                    crate::config::SourceOs::Linux => "Linux",
+                    crate::config::SourceOs::Unknown => "an unknown OS",
+                };
+                (os.to_string(), m.distro.clone())
+            }
+            None => ("an unknown OS".to_string(), "Linux".to_string()),
+        };
+        (base_url, model, source_os, distro)
+    };
+
+    let system_prompt = crate::llm::build_system_prompt(&source_os, &distro);
+    let messages = vec![
+        crate::llm::ChatMessage {
+            role: "system".into(),
+            content: system_prompt,
+        },
+        crate::llm::ChatMessage {
+            role: "user".into(),
+            content: message,
+        },
+    ];
+
+    tauri::async_runtime::spawn(async move {
+        if let Err(e) = crate::llm::chat_stream(&base_url, &model, messages, &app_handle).await {
+            log::error!("chat stream error: {e}");
+            let _ = app_handle.emit("robin:chat-error", e.to_string());
+        }
+    });
+
+    Ok(())
+}
diff --git a/src-tauri/src/lib.rs b/src-tauri/src/lib.rs
index 5da7576..086382f 100644
--- a/src-tauri/src/lib.rs
+++ b/src-tauri/src/lib.rs
@@ -1,6 +1,7 @@
 mod commands;
 mod config;
 mod distro;
+mod llm;
 mod notify;
 mod patterns;
 mod tray;
@@ -76,6 +77,7 @@ pub fn run() {
             commands::get_pending_events,
             commands::panel_opened,
             commands::panel_closed,
+            commands::chat,
         ])
         .run(tauri::generate_context!())
         .expect("error while running Robin");
diff --git a/src-tauri/src/llm.rs b/src-tauri/src/llm.rs
new file mode 100644
index 0000000..a1b8f20
--- /dev/null
+++ b/src-tauri/src/llm.rs
@@ -0,0 +1,156 @@
+use anyhow::Context;
+use serde::{Deserialize, Serialize};
+use tauri::{AppHandle, Emitter};
+
+#[derive(Debug, Clone, Serialize)]
+pub struct ChatMessage {
+    pub role: String,
+    pub content: String,
+}
+
+#[derive(Debug, Deserialize)]
+struct OllamaChunk {
+    message: Option<OllamaChunkMessage>,
+    done: bool,
+}
+
+#[derive(Debug, Deserialize)]
+struct OllamaChunkMessage {
+    content: String,
+}
+
+pub fn build_system_prompt(source_os: &str, distro: &str) -> String {
+    format!(
+        "You are Robin, a friendly Linux migration companion built into the user's desktop. \
+        The user migrated from {source_os} and is currently using {distro}. \
+        Help them with Linux questions. When concepts differ from their previous OS, \
+        explain the Linux equivalent clearly. Be concise, practical, and warm. \
+        Use plain language and avoid unexplained jargon. \
+        Never make the user feel bad for not knowing something."
+    )
+}
+
+pub async fn chat_stream(
+    base_url: &str,
+    model: &str,
+    messages: Vec<ChatMessage>,
+    app: &AppHandle,
+) -> anyhow::Result<()> {
+    let client = reqwest::Client::new();
+    let url = format!("{base_url}/api/chat");
+    let body = serde_json::json!({
+        "model": model,
+        "messages": messages,
+        "stream": true
+    });
+
+    let mut response = client
+        .post(&url)
+        .json(&body)
+        .send()
+        .await
+        .context("failed to reach Ollama — is it running at the configured URL?")?;
+
+    if !response.status().is_success() {
+        let status = response.status();
+        let text = response.text().await.unwrap_or_default();
+        anyhow::bail!("Ollama returned {status}: {text}");
+    }
+
+    let mut buffer: Vec<u8> = Vec::new();
+
+    loop {
+        match response.chunk().await {
+            Ok(Some(bytes)) => {
+                buffer.extend_from_slice(&bytes);
+                while let Some(nl) = buffer.iter().position(|&b| b == b'\n') {
+                    let line_bytes = buffer[..nl].to_vec();
+                    buffer.drain(..=nl);
+                    let line = String::from_utf8_lossy(&line_bytes);
+                    let line = line.trim();
+                    if line.is_empty() {
+                        continue;
+                    }
+                    match serde_json::from_str::<OllamaChunk>(line) {
+                        Ok(chunk) if chunk.done => {
+                            if let Err(e) = app.emit("robin:chat-done", ()) {
+                                log::warn!("llm: failed to emit chat-done: {e}");
+                            }
+                            return Ok(());
+                        }
+                        Ok(chunk) => {
+                            if let Some(msg) = chunk.message {
+                                if !msg.content.is_empty() {
+                                    if let Err(e) = app.emit("robin:chat-token", msg.content) {
+                                        log::warn!("llm: failed to emit chat-token: {e}");
+                                    }
+                                }
+                            }
+                        }
+                        Err(e) => {
+                            log::warn!("llm: failed to parse Ollama chunk: {e} — raw: {line}");
+                        }
+                    }
+                }
+            }
+            Ok(None) => break,
+            Err(e) => {
+                return Err(e).context("stream read error");
+            }
+        }
+    }
+
+    // Stream ended without a done:true line
+    if let Err(e) = app.emit("robin:chat-done", ()) {
+        log::warn!("llm: failed to emit chat-done: {e}");
+    }
+    Ok(())
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn prompt_includes_source_os_and_distro() {
+        let prompt = build_system_prompt("macOS", "cachyos");
+        assert!(prompt.contains("macOS"), "prompt must mention source OS");
+        assert!(prompt.contains("cachyos"), "prompt must mention distro");
+    }
+
+    #[test]
+    fn prompt_includes_robin_name() {
+        let prompt = build_system_prompt("Windows", "linuxmint");
+        assert!(prompt.contains("Robin"), "prompt must identify Robin as the assistant");
+    }
+
+    #[test]
+    fn parse_token_chunk() {
+        let json = r#"{"model":"llama3.2","created_at":"2024-01-01T00:00:00Z","message":{"role":"assistant","content":"Hello"},"done":false}"#;
+        let chunk: OllamaChunk = serde_json::from_str(json).unwrap();
+        assert!(!chunk.done);
+        assert_eq!(chunk.message.unwrap().content, "Hello");
+    }
+
+    #[test]
+    fn parse_done_chunk() {
+        let json = r#"{"model":"llama3.2","created_at":"2024-01-01T00:00:00Z","done":true,"done_reason":"stop"}"#;
+        let chunk: OllamaChunk = serde_json::from_str(json).unwrap();
+        assert!(chunk.done);
+        assert!(chunk.message.is_none());
+    }
+
+    #[test]
+    fn parse_empty_content_chunk_deserializes() {
+        let json = r#"{"model":"llama3.2","created_at":"2024-01-01T00:00:00Z","message":{"role":"assistant","content":""},"done":false}"#;
+        let chunk: OllamaChunk = serde_json::from_str(json).unwrap();
+        assert!(!chunk.done);
+        assert_eq!(chunk.message.unwrap().content, "");
+    }
+
+    #[test]
+    fn malformed_json_fails_to_parse() {
+        let result = serde_json::from_str::<OllamaChunk>("not valid json");
+        assert!(result.is_err(), "malformed JSON must fail to parse");
+    }
+}
diff --git a/src/components/ChatPanel.vue b/src/components/ChatPanel.vue
index 193c636..5ca174b 100644
--- a/src/components/ChatPanel.vue
+++ b/src/components/ChatPanel.vue
@@ -26,8 +26,11 @@
         class="chat-input"
         placeholder="Ask Robin something..."
         @keydown.enter="send"
+        :disabled="thinking"
       />
-      <button class="send-btn" @click="send">→</button>
+      <button class="send-btn" @click="send" :disabled="thinking">
+        {{ thinking ? '…' : '→' }}
+      </button>
     </footer>
   </div>
 </template>
@@ -42,13 +45,12 @@ interface RobinEvent { pattern_id: string; title: string; body: string; severity
 
 const messages = ref<Message[]>([])
 const input = ref('')
+const thinking = ref(false)
 const messagesEl = ref<HTMLElement | null>(null)
 let unlisten: UnlistenFn | null = null
+let activeCleanup: (() => void) | null = null
 
 onMounted(async () => {
-  // Signal backend: stop queuing events into PENDING while panel is open.
-  // Must happen before drain so new events that arrive during mount are not
-  // double-delivered on the next open cycle.
   try { await invoke('panel_opened') } catch {}
 
   try {
@@ -60,7 +62,6 @@ onMounted(async () => {
     console.warn('Robin: failed to drain pending events:', err)
   }
 
-  // Always set up the live listener, even if drain failed
   unlisten = await listen<RobinEvent>('robin:event', ({ payload }) => {
     pushRobinEvent(payload)
   })
@@ -69,6 +70,7 @@ onMounted(async () => {
 onUnmounted(() => {
   unlisten?.()
   invoke('panel_closed').catch(() => {})
+  activeCleanup?.()
 })
 
 function pushRobinEvent(e: RobinEvent) {
@@ -80,13 +82,55 @@ function pushRobinEvent(e: RobinEvent) {
 
 async function send() {
   const text = input.value.trim()
-  if (!text) return
+  if (!text || thinking.value) return
   input.value = ''
+  thinking.value = true
+
   messages.value.push({ role: 'user', content: text })
-  // M4: invoke('chat', { message: text }) and stream response
-  messages.value.push({ role: 'robin', content: '(LLM chat not yet connected — M4)' })
+  messages.value.push({ role: 'robin', content: '' })
+  const robinIdx = messages.value.length - 1
+
   await nextTick()
   messagesEl.value?.scrollTo({ top: messagesEl.value.scrollHeight, behavior: 'smooth' })
+
+  let unlistenToken: UnlistenFn | null = null
+  let unlistenDone: UnlistenFn | null = null
+  let unlistenError: UnlistenFn | null = null
+
+  function cleanup() {
+    unlistenToken?.()
+    unlistenDone?.()
+    unlistenError?.()
+    thinking.value = false
+    activeCleanup = null
+  }
+  activeCleanup = cleanup
+
+  unlistenToken = await listen<string>('robin:chat-token', ({ payload }) => {
+    messages.value[robinIdx].content += payload
+    nextTick(() => {
+      messagesEl.value?.scrollTo({ top: messagesEl.value.scrollHeight, behavior: 'smooth' })
+    })
+  })
+
+  unlistenDone = await listen('robin:chat-done', () => cleanup())
+
+  unlistenError = await listen<string>('robin:chat-error', ({ payload }) => {
+    if (!messages.value[robinIdx].content) {
+      messages.value[robinIdx].content =
+        `Robin couldn't reach Ollama. Make sure it's running at the configured URL. (${payload})`
+    }
+    cleanup()
+  })
+
+  try {
+    await invoke('chat', { message: text })
+  } catch (err) {
+    if (!messages.value[robinIdx].content) {
+      messages.value[robinIdx].content = `Something went wrong starting the chat. (${err})`
+    }
+    cleanup()
+  }
 }
 </script>
 
@@ -154,4 +198,6 @@ async function send() {
   cursor: pointer;
 }
 .send-btn:hover { opacity: 0.85; }
+.send-btn:disabled { opacity: 0.5; cursor: not-allowed; }
+.chat-input:disabled { opacity: 0.6; cursor: not-allowed; }
 </style>