diff --git a/backend/src/routes/chat.ts b/backend/src/routes/chat.ts
index cb47fe7..c172db5 100644
--- a/backend/src/routes/chat.ts
+++ b/backend/src/routes/chat.ts
@@ -78,7 +78,23 @@ function inlineTextFiles(messages: UIMessage[]): UIMessage[] {
   });
 }
 
-function systemPrompt(veilActive: boolean, providerLabel: string): string {
+// A short directive for the clinician's chosen "situation" mode, appended to the
+// base prompt. Chat mode adds nothing.
+function modeDirective(mode: string | undefined): string {
+  if (mode === "analysis") {
+    return "Mode — Analysis: the clinician wants interpretation, not just retrieval. After fetching a patient's data, surface patterns and correlations across their problems, labs and visits (e.g. recurring complaints, trends, likely links) and call out anything notable. Stay grounded in the tool results.";
+  }
+  if (mode === "graph") {
+    return "Mode — Graph: the clinician wants to see how a patient's problems and visits connect. When they reference a patient, call getPatient (its record graph renders automatically) and briefly describe the key relationships between illnesses and encounters.";
+  }
+  return "";
+}
+
+function systemPrompt(
+  veilActive: boolean,
+  providerLabel: string,
+  mode?: string,
+): string {
   return [
     "You are temetro, a clinical assistant that helps clinicians retrieve,",
     "organize, and add patient information. You operate over a real patient",
@@ -130,6 +146,7 @@ function systemPrompt(veilActive: boolean, providerLabel: string): string {
     veilActive
       ? `Privacy: this conversation runs on an external provider (${providerLabel}). Patient identifiers are de-identified as tokens like [PATIENT_1] / [MRN_1]; refer to patients generically ("this patient") rather than repeating tokens.`
       : "",
+    modeDirective(mode),
   ]
     .filter(Boolean)
     .join("\n");
@@ -137,10 +154,11 @@ function systemPrompt(veilActive: boolean, providerLabel: string): string {
 
 chatRouter.post("/", async (req, res, next) => {
   try {
-    const { messages, model: requestedModel } = req.body as {
+    const { messages, model: requestedModel, mode } = req.body as {
       messages: UIMessage[];
       model?: string;
       effort?: string;
+      mode?: string;
     };
     if (!Array.isArray(messages)) {
       res.status(400).json({ error: "messages must be an array." });
@@ -171,7 +189,7 @@ chatRouter.post("/", async (req, res, next) => {
     };
 
     const modelMessages = await convertToModelMessages(inlineTextFiles(messages));
-    const system = systemPrompt(veil.active, resolved.providerLabel);
+    const system = systemPrompt(veil.active, resolved.providerLabel, mode);
 
     const stream = createUIMessageStream({
       execute: async ({ writer }) => {
diff --git a/frontend/components/chat/chat-input.tsx b/frontend/components/chat/chat-input.tsx
index 1fcc16f..fa5071a 100644
--- a/frontend/components/chat/chat-input.tsx
+++ b/frontend/components/chat/chat-input.tsx
@@ -11,19 +11,17 @@ import {
 } from "react";
 import { useTranslation } from "react-i18next";
 
-import { ModelPicker } from "@/components/chat/model-picker";
+import { ModePicker } from "@/components/chat/mode-picker";
 import { PatientFormDialog } from "@/components/chat/patient-form-dialog";
-import type { Effort } from "@/lib/ai-models";
+import type { ChatMode } from "@/lib/chat-modes";
 import { cn } from "@/lib/utils";
 
 type ChatInputProps = {
   onSubmit: (text: string, files: File[]) => void;
   status: ChatStatus;
   onStop?: () => void;
-  model: string;
-  effort: Effort;
-  onModelChange: (model: string) => void;
-  onEffortChange: (effort: Effort) => void;
+  mode: ChatMode;
+  onModeChange: (mode: ChatMode) => void;
 };
 
 const iconButton =
@@ -37,10 +35,8 @@ export function ChatInput({
   onSubmit,
   status,
   onStop,
-  model,
-  effort,
-  onModelChange,
-  onEffortChange,
+  mode,
+  onModeChange,
 }: ChatInputProps) {
   const { t } = useTranslation();
 
@@ -181,11 +177,9 @@ export function ChatInput({
           </div>
 
           <div className="flex shrink-0 items-center gap-1">
-            <ModelPicker
-              effort={effort}
-              model={model}
-              onEffortChange={onEffortChange}
-              onModelChange={onModelChange}
+            <ModePicker
+              mode={mode}
+              onModeChange={onModeChange}
               triggerClassName={cn(pillButton, "mr-1")}
             />
             <button
diff --git a/frontend/components/chat/chat-panel.tsx b/frontend/components/chat/chat-panel.tsx
index d3b2e69..dec37c1 100644
--- a/frontend/components/chat/chat-panel.tsx
+++ b/frontend/components/chat/chat-panel.tsx
@@ -61,6 +61,7 @@ import { InventoryListCard } from "@/components/chat/inventory-list-card";
 import { ImportPreviewCard } from "@/components/chat/import-preview-card";
 import { LabChartCard } from "@/components/chat/lab-chart-card";
 import { PatientResult } from "@/components/chat/patient-cards";
+import { RecordGraph } from "@/components/graph/record-graph";
 import {
   AppointmentListCard,
   PrescriptionListCard,
@@ -79,6 +80,7 @@ import {
   type Effort,
   getModel,
 } from "@/lib/ai-models";
+import { type ChatMode, DEFAULT_MODE } from "@/lib/chat-modes";
 import type { ActionPreviewData, TemetroUIMessage } from "@/lib/ai-chat";
 import {
   getThread,
@@ -114,6 +116,9 @@ export function ChatPanel() {
   const { t } = useTranslation();
   const [model, setModel] = useState<string>(DEFAULT_MODEL_ID);
   const [effort, setEffort] = useState<Effort>(DEFAULT_EFFORT);
+  // The clinician-facing "situation" mode (Chat / Analysis / Graph). The model
+  // itself comes from Settings → AI; this shapes what the assistant does.
+  const [mode, setMode] = useState<ChatMode>(DEFAULT_MODE);
 
   // Veil consent: cloud models de-identify + send data externally. We ask once
   // per session before the first such send — inline (no modal). `pendingConsent`
@@ -185,10 +190,10 @@ export function ChatPanel() {
       const fileParts = await Promise.all(files.map(fileToPart));
       sendMessage(
         { text, files: fileParts },
-        { body: { model: modelId, effort, threadId: threadIdRef.current } },
+        { body: { model: modelId, effort, mode, threadId: threadIdRef.current } },
       );
     },
-    [sendMessage, effort],
+    [sendMessage, effort, mode],
   );
 
   const send = useCallback(
@@ -223,7 +228,11 @@ export function ChatPanel() {
             id: nanoid(),
             role: "assistant",
             parts: patient
-              ? [{ type: "data-patientCard", data: patient }]
+              ? [
+                  mode === "graph"
+                    ? { type: "data-recordGraph", data: patient }
+                    : { type: "data-patientCard", data: patient },
+                ]
               : [
                   {
                     type: "text",
@@ -245,6 +254,7 @@ export function ChatPanel() {
     [
       consented,
       isCloudModel,
+      mode,
       model,
       pendingConsent,
       runAgentWith,
@@ -353,10 +363,8 @@ export function ChatPanel() {
 
   const promptInput = (
     <ChatInput
-      effort={effort}
-      model={model}
-      onEffortChange={setEffort}
-      onModelChange={setModel}
+      mode={mode}
+      onModeChange={setMode}
       onStop={stop}
       onSubmit={send}
       status={status}
@@ -545,6 +553,26 @@ export function ChatPanel() {
                 />
               );
             }
+            if (part.type === "data-recordGraph") {
+              return (
+                <div
+                  className="w-full overflow-hidden rounded-2xl border bg-card/30"
+                  key={key}
+                >
+                  <div className="flex items-center justify-between gap-2 px-4 pt-3">
+                    <span className="font-medium text-foreground text-sm">
+                      {part.data.name}
+                    </span>
+                    <span className="text-muted-foreground text-xs">
+                      {t("chat.graphCard.label")}
+                    </span>
+                  </div>
+                  <div className="p-3">
+                    <RecordGraph patient={part.data} />
+                  </div>
+                </div>
+              );
+            }
             if (part.type === "data-labCard") {
               return <LabChartCard data={part.data} key={key} />;
             }
diff --git a/frontend/components/chat/mode-picker.tsx b/frontend/components/chat/mode-picker.tsx
new file mode 100644
index 0000000..e7eafd5
--- /dev/null
+++ b/frontend/components/chat/mode-picker.tsx
@@ -0,0 +1,77 @@
+"use client";
+
+import { ChevronDown } from "lucide-react";
+import { useTranslation } from "react-i18next";
+
+import {
+  Menu,
+  MenuPopup,
+  MenuRadioGroup,
+  MenuRadioItem,
+  MenuTrigger,
+} from "@/components/ui/menu";
+import { type ChatMode, CHAT_MODES, getMode } from "@/lib/chat-modes";
+
+type ModePickerProps = {
+  mode: ChatMode;
+  onModeChange: (mode: ChatMode) => void;
+  triggerClassName: string;
+};
+
+/**
+ * The situation-mode control in the chat composer (Chat / Analysis / Graph).
+ * Replaces the old model picker — clinicians pick what they want the assistant
+ * to do, not which LLM runs (that's configured once in Settings → AI).
+ */
+export function ModePicker({
+  mode,
+  onModeChange,
+  triggerClassName,
+}: ModePickerProps) {
+  const { t } = useTranslation();
+  const selected = getMode(mode);
+  const SelectedIcon = selected.icon;
+
+  return (
+    <Menu>
+      <MenuTrigger
+        render={
+          <button
+            aria-label={t("chat.input.mode")}
+            className={triggerClassName}
+            type="button"
+          />
+        }
+      >
+        <SelectedIcon className="size-4 opacity-80" />
+        <span className="truncate font-medium text-foreground">
+          {t(selected.labelKey)}
+        </span>
+        <ChevronDown className="size-4 opacity-70" />
+      </MenuTrigger>
+      <MenuPopup align="end" className="min-w-64">
+        <MenuRadioGroup
+          onValueChange={(value) => onModeChange(value as ChatMode)}
+          value={mode}
+        >
+          {CHAT_MODES.map((m) => {
+            const Icon = m.icon;
+            return (
+              <MenuRadioItem key={m.id} value={m.id}>
+                <span className="flex items-center gap-2">
+                  <Icon className="size-4 opacity-80" />
+                  <span className="flex flex-col">
+                    <span className="text-foreground">{t(m.labelKey)}</span>
+                    <span className="text-muted-foreground text-xs">
+                      {t(m.descriptionKey)}
+                    </span>
+                  </span>
+                </span>
+              </MenuRadioItem>
+            );
+          })}
+        </MenuRadioGroup>
+      </MenuPopup>
+    </Menu>
+  );
+}
diff --git a/frontend/lib/ai-chat.ts b/frontend/lib/ai-chat.ts
index b7ba2b8..060c974 100644
--- a/frontend/lib/ai-chat.ts
+++ b/frontend/lib/ai-chat.ts
@@ -69,6 +69,9 @@ export type ActionPreviewData = {
 // prefixed with `data-` (e.g. `data-patientCard`), per the AI SDK convention.
 export type TemetroDataParts = {
   patientCard: Patient;
+  // The same patient rendered as an Obsidian-style problems↔visits graph (Graph
+  // mode). Carries the full record so the graph renders client-side.
+  recordGraph: Patient;
   labCard: LabCardData;
   importPreview: ImportPreviewData;
   veilNotice: VeilNoticeData;
diff --git a/frontend/lib/chat-modes.ts b/frontend/lib/chat-modes.ts
new file mode 100644
index 0000000..8de1c30
--- /dev/null
+++ b/frontend/lib/chat-modes.ts
@@ -0,0 +1,44 @@
+// The chat "situation" modes shown in the composer — what the clinician wants
+// the assistant to do, rather than which LLM runs (the model/provider is set
+// once in Settings → AI). The mode travels with each send so the backend can
+// shape its system prompt, and Graph mode also renders the record graph for the
+// `/patient` fast-path.
+
+import { MessageSquare, Network, Sparkles } from "lucide-react";
+
+export type ChatMode = "chat" | "analysis" | "graph";
+
+export const DEFAULT_MODE: ChatMode = "chat";
+
+export type ChatModeOption = {
+  id: ChatMode;
+  icon: typeof MessageSquare;
+  // i18n keys under `chat.input.modes.*`.
+  labelKey: string;
+  descriptionKey: string;
+};
+
+export const CHAT_MODES: ChatModeOption[] = [
+  {
+    id: "chat",
+    icon: MessageSquare,
+    labelKey: "chat.input.modes.chat.label",
+    descriptionKey: "chat.input.modes.chat.description",
+  },
+  {
+    id: "analysis",
+    icon: Sparkles,
+    labelKey: "chat.input.modes.analysis.label",
+    descriptionKey: "chat.input.modes.analysis.description",
+  },
+  {
+    id: "graph",
+    icon: Network,
+    labelKey: "chat.input.modes.graph.label",
+    descriptionKey: "chat.input.modes.graph.description",
+  },
+];
+
+export function getMode(id: string): ChatModeOption {
+  return CHAT_MODES.find((m) => m.id === id) ?? CHAT_MODES[0];
+}
diff --git a/frontend/lib/i18n/locales/en/translation.json b/frontend/lib/i18n/locales/en/translation.json
index 7c14715..2238875 100644
--- a/frontend/lib/i18n/locales/en/translation.json
+++ b/frontend/lib/i18n/locales/en/translation.json
@@ -888,6 +888,21 @@
       "send": "Send",
       "stop": "Stop",
       "model": "Model",
+      "mode": "Mode",
+      "modes": {
+        "chat": {
+          "label": "Chat",
+          "description": "Ask and retrieve patient information."
+        },
+        "analysis": {
+          "label": "Analysis",
+          "description": "Interpret patterns across a patient's history."
+        },
+        "graph": {
+          "label": "Graph",
+          "description": "See how problems and visits connect."
+        }
+      },
       "moreModels": "More models",
       "effort": "Effort",
       "effortOptions": {
@@ -911,6 +926,9 @@
     "thinking": "Thinking…",
     "steps": "Steps",
     "reasoning": "Reasoning",
+    "graphCard": {
+      "label": "Record graph"
+    },
     "history": {
       "title": "Chats",
       "untitled": "New chat",