diff --git a/qitos/qita/_cli_app.py b/qitos/qita/_cli_app.py
index e1cb9a9..b23bc41 100644
--- a/qitos/qita/_cli_app.py
+++ b/qitos/qita/_cli_app.py
@@ -1466,6 +1466,61 @@ def _render_run_html(payload: Dict[str, Any], embedded: bool) -> str:
   if(!thought) return (summary || '<div class="muted">No explicit thought.</div>');
   return '<div style="white-space:pre-wrap;line-height:1.6;background:#0b1220;border:1px solid #1c2b44;border-radius:8px;padding:10px;color:#cfe6ff">'+esc(thought)+'</div>' + summary;
 }}
+function modelInputPayloads(events){{
+  const out = [];
+  const es = Array.isArray(events) ? events : [];
+  for(const e of es){{
+    const payload = e && typeof e.payload === 'object' ? e.payload : null;
+    if(payload && String(payload.stage || '') === 'model_input') out.push(payload);
+  }}
+  return out;
+}}
+function renderModelMessagePreview(messages){{
+  const msgs = Array.isArray(messages) ? messages : [];
+  if(!msgs.length) return '<div class="muted">No message list recorded.</div>';
+  const blocks = [];
+  for(let i=0;i<msgs.length;i+=1){{
+    const msg = msgs[i] && typeof msgs[i] === 'object' ? msgs[i] : {{}};
+    const role = msg.role !== undefined ? String(msg.role) : 'message';
+    const content = msg.content !== undefined ? msg.content : '';
+    blocks.push(
+      '<details class="raw">' +
+      '<summary class="muted">message ' + i + ' · ' + esc(role) + '</summary>' +
+      '<pre>' + esc(typeof content === 'string' ? content : JSON.stringify(content, null, 2)) + '</pre>' +
+      '</details>'
+    );
+  }}
+  return blocks.join('');
+}}
+function renderModelInput(events){{
+  const inputs = modelInputPayloads(events);
+  if(!inputs.length){{
+    return '<div class="muted">No model input recorded for this step. This usually means the step was handled by the harness or a tool path without calling the model.</div>';
+  }}
+  const blocks = [];
+  for(let i=0;i<inputs.length;i+=1){{
+    const payload = inputs[i] || {{}};
+    const ctx = payload.context && typeof payload.context === 'object' ? payload.context : {{}};
+    const rows = [];
+    rows.push(kvRow('model input event', i + 1));
+    if(payload.history_message_count !== undefined) rows.push(kvRow('history messages', payload.history_message_count));
+    if(Array.isArray(payload.messages)) rows.push(kvRow('messages sent', payload.messages.length));
+    if(ctx.input_tokens_total !== undefined) rows.push(kvRow('input tokens', ctx.input_tokens_total));
+    if(ctx.prepared_tokens !== undefined) rows.push(kvRow('prepared tokens', ctx.prepared_tokens));
+    if(ctx.history_tokens !== undefined) rows.push(kvRow('history tokens', ctx.history_tokens));
+    if(ctx.occupancy_ratio !== undefined) rows.push(kvRow('context used', ((Number(ctx.occupancy_ratio) || 0) * 100).toFixed(1) + '%'));
+    const prepared = payload.prepared_full || payload.prepared || '';
+    const preparedBlock = prepared
+      ? '<div style="margin-top:8px"><div class="muted" style="margin-bottom:4px">prepared text from agent.prepare(state)</div><pre>' + esc(String(prepared)) + '</pre></div>'
+      : '<div class="muted" style="margin-top:8px">No prepared text recorded.</div>';
+    const messagesBlock =
+      '<details class="raw" style="margin-top:8px"><summary class="muted">Actual messages sent to model</summary>' +
+      renderModelMessagePreview(payload.messages) +
+      '</details>';
+    blocks.push('<div class="item">' + kvBlock(rows) + preparedBlock + messagesBlock + '</div>');
+  }}
+  return '<div class="list">' + blocks.join('') + '</div>';
+}}
 function renderAction(actions){{
   const label = firstActionLabel(actions);
   if(!label) return '<div class="muted">No action.</div>';
@@ -1870,6 +1925,7 @@ def _render_run_html(payload: Dict[str, Any], embedded: bool) -> str:
     let h = '<div class="card-head"><div class="step">STEP ' + it.sid + '</div><div class="muted">events ' + it.events.length + '</div></div>';
     if(showObs) h += sectionHtml('State', renderState(obsInput), obsInput, 'state', collapsedAll);
     h += sectionHtml('Prompt', renderPromptMetadata(it.step.prompt_metadata || {{}}), it.step.prompt_metadata || {{}}, 'prompt', collapsedAll);
+    h += sectionHtml('Model Input', renderModelInput(it.events), modelInputPayloads(it.events), 'model_input', collapsedAll);
     h += sectionHtml('Visual Assets', renderVisualAssets(it.step), {{visual_assets: it.step.visual_assets || [], observation_modalities: it.step.observation_modalities || [], model_input_modalities: it.step.model_input_modalities || [], model_input_visual_count: it.step.model_input_visual_count || 0}}, 'visual_assets', collapsedAll);
     h += sectionHtml('Thought', renderThought(d, it.events, it.step), d, 'thought', collapsedAll);
     h += sectionHtml('Parser Diagnostics', renderParserDiagnostics(it.step.parser_diagnostics || {{}}), it.step.parser_diagnostics || {{}}, 'parser', collapsedAll);
diff --git a/tests/test_qita_cli.py b/tests/test_qita_cli.py
index 9230e85..9c1f502 100644
--- a/tests/test_qita_cli.py
+++ b/tests/test_qita_cli.py
@@ -100,6 +100,7 @@ def _make_run(root: Path, run_id: str) -> Path:
     )
     (run / "events.jsonl").write_text(
         '{"step_id":0,"phase":"INIT","ok":true,"ts":"x"}\n'
+        '{"step_id":0,"phase":"DECIDE","ok":true,"ts":"x1","payload":{"stage":"model_input","prepared":"Task: inspect the run","prepared_full":"Task: inspect the run\\nStep: 0","history_message_count":1,"messages":[{"role":"system","content":"You are an agent."},{"role":"user","content":"Task: inspect the run"}],"context":{"input_tokens_total":22,"prepared_tokens":6,"history_tokens":8,"occupancy_ratio":0.12}}}\n'
         '{"step_id":0,"phase":"DECIDE","ok":true,"ts":"y","payload":{"stage":"model_output","raw_output":"Thought: inspect the run","model_response":{"text":"Thought: inspect the run","usage":{"prompt_tokens":10,"completion_tokens":4,"total_tokens":14},"finish_reason":"stop","tool_calls":[{"id":"call_1","type":"function","function":{"name":"visit_url","arguments":"{\\"url\\":\\"https://example.com\\"}"}}],"model_name":"demo-model","provider":"demo-provider","metadata":{}},"context":{"input_tokens_total":3200,"occupancy_ratio":0.74}}}\n',
         encoding="utf-8",
     )
@@ -251,6 +252,9 @@ def test_render_pages(tmp_path: Path):
     assert "decision_source" in view
     assert "native_tool_call_used" in view
     assert "tool_delivery" in view
+    assert "Model Input" in view
+    assert "prepared text from agent.prepare(state)" in view
+    assert "Actual messages sent to model" in view
     assert "Visual Assets" in view
     assert "grounding metadata" in view
     assert "critic retries" in view