cascade-ai 0.2.11 → 0.3.0

package/dist/cli.js

@@ -83,7 +83,7 @@ var __toESM = (mod, isNodeMode, target) => (target = mod != null ? __create(__ge
 var CASCADE_VERSION, CASCADE_CONFIG_FILE, CASCADE_DB_FILE, CASCADE_DASHBOARD_SECRET_FILE, GLOBAL_CONFIG_DIR, GLOBAL_DB_FILE, GLOBAL_KEYSTORE_FILE, GLOBAL_RUNTIME_DB_FILE, DEFAULT_DASHBOARD_PORT, DEFAULT_CONTEXT_LIMIT, DEFAULT_AUTO_SUMMARIZE_AT, MODELS, T1_MODEL_PRIORITY, T2_MODEL_PRIORITY, T3_MODEL_PRIORITY, VISION_MODEL_PRIORITY, COMPLEXITY_T2_COUNT, THEME_NAMES, DEFAULT_THEME, OLLAMA_BASE_URL, LM_STUDIO_BASE_URL, AZURE_BASE_URL_TEMPLATE, TOOL_NAMES, DEFAULT_APPROVAL_REQUIRED;
 var init_constants = __esm({
   "src/constants.ts"() {
-    CASCADE_VERSION = "0.2.11";
+    CASCADE_VERSION = "0.3.0";
     CASCADE_CONFIG_FILE = ".cascade/config.json";
     CASCADE_DB_FILE = ".cascade/memory.db";
     CASCADE_DASHBOARD_SECRET_FILE = ".cascade/dashboard-secret";
@@ -522,17 +522,38 @@ var init_anthropic = __esm({
           messages,
           tools: tools?.length ? tools : void 0
         });
+        let isThinking = false;
         for await (const event of stream) {
-          if (event.type === "content_block_delta" && event.delta.type === "text_delta") {
-            const text = event.delta.text;
-            fullContent += text;
-            onChunk({ text, finishReason: null });
+          if (event.type === "content_block_delta") {
+            if (event.delta.type === "thinking_delta") {
+              if (!isThinking) {
+                isThinking = true;
+                fullContent += "<think>\n";
+                onChunk({ text: "<think>\n", finishReason: null });
+              }
+              const text = event.delta.thinking;
+              fullContent += text;
+              onChunk({ text, finishReason: null });
+            } else if (event.delta.type === "text_delta") {
+              if (isThinking) {
+                isThinking = false;
+                fullContent += "\n</think>\n\n";
+                onChunk({ text: "\n</think>\n\n", finishReason: null });
+              }
+              const text = event.delta.text;
+              fullContent += text;
+              onChunk({ text, finishReason: null });
+            }
           } else if (event.type === "message_delta" && event.usage) {
             outputTokens = event.usage.output_tokens;
           } else if (event.type === "message_start" && event.message.usage) {
             inputTokens = event.message.usage.input_tokens;
           }
         }
+        if (isThinking) {
+          fullContent += "\n</think>\n\n";
+          onChunk({ text: "\n</think>\n\n", finishReason: null });
+        }
         const finalMessage = await stream.finalMessage();
         const toolCalls = finalMessage.content.filter((b) => b.type === "tool_use").map((b) => ({
           id: b.id,
@@ -716,9 +737,25 @@ var init_openai = __esm({
           }
         }
         const toolCallsMap = {};
+        let isThinking = false;
         for await (const chunk of stream) {
           const delta = chunk.choices[0]?.delta;
+          const reasoningContent = delta?.reasoning_content;
+          if (reasoningContent) {
+            if (!isThinking) {
+              isThinking = true;
+              fullContent += "<think>\n";
+              onChunk({ text: "<think>\n", finishReason: null });
+            }
+            fullContent += reasoningContent;
+            onChunk({ text: reasoningContent, finishReason: null });
+          }
           if (delta?.content) {
+            if (isThinking) {
+              isThinking = false;
+              fullContent += "\n</think>\n\n";
+              onChunk({ text: "\n</think>\n\n", finishReason: null });
+            }
             fullContent += delta.content;
             onChunk({ text: delta.content, finishReason: null });
           }
@@ -741,6 +778,10 @@ var init_openai = __esm({
             outputTokens = chunk.usage.completion_tokens;
           }
         }
+        if (isThinking) {
+          fullContent += "\n</think>\n\n";
+          onChunk({ text: "\n</think>\n\n", finishReason: null });
+        }
         const toolCalls = Object.values(toolCallsMap).map((tc) => {
           let input = {};
           try {
@@ -1834,8 +1875,8 @@ Original error: ${err.message}`
   upsertRuntimeNode(node) {
     this.enqueueWrite(() => {
       this.db.prepare(`
-        INSERT INTO runtime_nodes (tier_id, session_id, parent_id, role, label, status, current_action, progress_pct, updated_at, workspace_path, is_global)
-        VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+        INSERT INTO runtime_nodes (tier_id, session_id, parent_id, role, label, status, current_action, progress_pct, updated_at, workspace_path, is_global, output)
+        VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
         ON CONFLICT(tier_id) DO UPDATE SET
           session_id = excluded.session_id,
           parent_id = excluded.parent_id,
@@ -1846,7 +1887,8 @@ Original error: ${err.message}`
           progress_pct = excluded.progress_pct,
           updated_at = excluded.updated_at,
           workspace_path = excluded.workspace_path,
-          is_global = excluded.is_global
+          is_global = excluded.is_global,
+          output = excluded.output
       `).run(
         node.tierId,
         node.sessionId,
@@ -1858,7 +1900,8 @@ Original error: ${err.message}`
         node.progressPct ?? null,
         node.updatedAt,
         node.workspacePath ?? null,
-        node.isGlobal ? 1 : 0
+        node.isGlobal ? 1 : 0,
+        node.output ?? null
       );
     });
   }
@@ -1879,14 +1922,15 @@ Original error: ${err.message}`
       progressPct: row.progress_pct ?? void 0,
       updatedAt: row.updated_at,
       workspacePath: row.workspace_path ?? void 0,
-      isGlobal: row.is_global === 1
+      isGlobal: row.is_global === 1,
+      output: row.output ?? void 0
     }));
   }
   addRuntimeNodeLog(log) {
     this.enqueueWrite(() => {
       this.db.prepare(`
-        INSERT INTO runtime_node_logs (id, session_id, tier_id, role, label, status, current_action, progress_pct, timestamp, workspace_path, is_global)
-        VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+        INSERT INTO runtime_node_logs (id, session_id, tier_id, role, label, status, current_action, progress_pct, timestamp, workspace_path, is_global, output)
+        VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
       `).run(
         log.id,
         log.sessionId,
@@ -1898,7 +1942,8 @@ Original error: ${err.message}`
         log.progressPct ?? null,
         log.timestamp,
         log.workspacePath ?? null,
-        log.isGlobal ? 1 : 0
+        log.isGlobal ? 1 : 0,
+        log.output ?? null
       );
       this.db.prepare(`
         DELETE FROM runtime_node_logs
@@ -1941,7 +1986,8 @@ Original error: ${err.message}`
       progressPct: row.progress_pct ?? void 0,
       timestamp: row.timestamp,
       workspacePath: row.workspace_path ?? void 0,
-      isGlobal: row.is_global === 1
+      isGlobal: row.is_global === 1,
+      output: row.output ?? void 0
     }));
   }
   // ── Messages ──────────────────────────────────
@@ -2268,7 +2314,8 @@ Original error: ${err.message}`
         progress_pct INTEGER,
         updated_at TEXT NOT NULL,
         workspace_path TEXT,
-        is_global INTEGER NOT NULL DEFAULT 0
+        is_global INTEGER NOT NULL DEFAULT 0,
+        output TEXT
       );
 
       CREATE INDEX IF NOT EXISTS idx_runtime_nodes_session ON runtime_nodes(session_id);
@@ -2285,7 +2332,8 @@ Original error: ${err.message}`
         progress_pct INTEGER,
         timestamp TEXT NOT NULL,
         workspace_path TEXT,
-        is_global INTEGER NOT NULL DEFAULT 0
+        is_global INTEGER NOT NULL DEFAULT 0,
+        output TEXT
       );
 
       CREATE TABLE IF NOT EXISTS model_cache (
@@ -2314,6 +2362,14 @@ Original error: ${err.message}`
 
       CREATE INDEX IF NOT EXISTS idx_file_snapshots_session ON file_snapshots(session_id);
     `);
+    try {
+      this.db.exec("ALTER TABLE runtime_nodes ADD COLUMN output TEXT");
+    } catch {
+    }
+    try {
+      this.db.exec("ALTER TABLE runtime_node_logs ADD COLUMN output TEXT");
+    } catch {
+    }
   }
   // ── Deserializers ─────────────────────────────
   deserializeSession(row, messages) {
@@ -2576,6 +2632,7 @@ var ConfigManager = class {
     }
   }
   async injectEnvKeys() {
+    const isFirstRun = this.config.providers.length === 0;
     const envProviders = [
       { env: "ANTHROPIC_API_KEY", type: "anthropic" },
       { env: "OPENAI_API_KEY", type: "openai" },
@@ -2586,10 +2643,13 @@ var ConfigManager = class {
       const key = process.env[env];
       if (!key) continue;
       const existing = this.config.providers.find((p) => p.type === type);
-      if (!existing) this.config.providers.push({ type, apiKey: key });
-      else if (!existing.apiKey) existing.apiKey = key;
+      if (!existing && isFirstRun) {
+        this.config.providers.push({ type, apiKey: key });
+      } else if (existing && !existing.apiKey) {
+        existing.apiKey = key;
+      }
     }
-    if (!this.config.providers.find((p) => p.type === "ollama")) {
+    if (isFirstRun && !this.config.providers.find((p) => p.type === "ollama")) {
       this.config.providers.push({ type: "ollama" });
     }
   }
@@ -3538,7 +3598,7 @@ var BaseTier = class extends EventEmitter {
   getStatus() {
     return this.status;
   }
-  setStatus(status) {
+  setStatus(status, output) {
     this.status = status;
     const timestamp = (/* @__PURE__ */ new Date()).toISOString();
     const event = {
@@ -3547,7 +3607,8 @@ var BaseTier = class extends EventEmitter {
       parentId: this.parentId,
       label: this.label,
       status,
-      timestamp
+      timestamp,
+      output
     };
     this.emit("status", event);
     this.emit("tier:status", event);
@@ -3573,7 +3634,8 @@ var BaseTier = class extends EventEmitter {
       status: this.status,
       currentAction: update.currentAction,
       progressPct: update.progressPct,
-      timestamp
+      timestamp,
+      output: update.output
     });
   }
   buildMessage(type, to, payload) {
@@ -3954,16 +4016,17 @@ Now execute your subtask using this context where relevant.`
           return this.buildResult("ESCALATED", output, { checksRun, passed, failed }, issues, correctionAttempts);
         }
       }
-      this.setStatus("COMPLETED");
-      this.sendStatusUpdate({ progressPct: 100, currentAction: "Subtask complete", status: "IN_PROGRESS" });
+      this.setStatus("COMPLETED", output);
+      this.sendStatusUpdate({ progressPct: 100, currentAction: "Subtask complete", status: "IN_PROGRESS", output });
       this.peerBus?.publish(this.id, assignment.subtaskId, output, "COMPLETED");
       return this.buildResult("COMPLETED", output, { checksRun, passed, failed }, issues, correctionAttempts);
     } catch (err) {
       const errMsg = err instanceof Error ? err.message : String(err);
       issues.push(`Execution error: ${errMsg}`);
-      this.setStatus("FAILED");
-      this.peerBus?.publish(this.id, assignment.subtaskId, errMsg, "FAILED");
-      return this.buildResult("ESCALATED", output || errMsg, { checksRun, passed, failed }, issues, correctionAttempts);
+      const finalOutput = output || errMsg;
+      this.setStatus("FAILED", finalOutput);
+      this.peerBus?.publish(this.id, assignment.subtaskId, finalOutput, "FAILED");
+      return this.buildResult("ESCALATED", finalOutput, { checksRun, passed, failed }, issues, correctionAttempts);
     }
   }
   sendToPeer(toId, content) {
@@ -4646,8 +4709,9 @@ var T2Manager = class extends BaseTier {
       const summary = await this.aggregateResults(assignment, t3Results);
       const issues = t3Results.filter((r) => r.status !== "COMPLETED").flatMap((r) => r.issues);
       const overallStatus = this.determineStatus(t3Results);
-      this.setStatus(overallStatus === "COMPLETED" ? "COMPLETED" : "FAILED");
-      this.sendStatusUpdate({ progressPct: 100, currentAction: "Section complete", status: "IN_PROGRESS" });
+      const isOk = overallStatus === "COMPLETED" || overallStatus === "PARTIAL";
+      this.setStatus(isOk ? "COMPLETED" : "FAILED", summary);
+      this.sendStatusUpdate({ progressPct: 100, currentAction: "Section complete", status: "IN_PROGRESS", output: summary });
       const result = {
         sectionId: assignment.sectionId,
         sectionTitle: assignment.sectionTitle,
@@ -4660,7 +4724,7 @@ var T2Manager = class extends BaseTier {
       return result;
     } catch (err) {
       const errMsg = err instanceof Error ? err.message : String(err);
-      this.setStatus("FAILED");
+      this.setStatus("FAILED", errMsg);
       const failedResult = {
         sectionId: assignment.sectionId,
         sectionTitle: assignment.sectionTitle,
@@ -5225,8 +5289,8 @@ Create a CORRECTION PLAN that contains only the new sections needed to fix the i
       status: "IN_PROGRESS"
     });
     const output = await this.compileFinalOutput(userPrompt, plan, allT2Results);
-    this.setStatus("COMPLETED");
-    this.sendStatusUpdate({ progressPct: 100, currentAction: "Task complete", status: "IN_PROGRESS" });
+    this.setStatus("COMPLETED", output);
+    this.sendStatusUpdate({ progressPct: 100, currentAction: "Task complete", status: "IN_PROGRESS", output });
     return { output, t2Results: allT2Results, taskId: this.taskId, complexity: plan.complexity };
   }
   getEscalations() {
@@ -8055,11 +8119,6 @@ var SlashCommandRegistry = class {
       description: "Show active models per tier",
       handler: (_args, ctx) => ({ output: ctx.onModelInfo(), handled: true })
     });
-    this.register({
-      command: "/models",
-      description: "Browse available models by provider",
-      handler: (_args, ctx) => ({ output: ctx.onModelsInfo(), handled: true })
-    });
     this.register({
       command: "/providers",
       description: "Show configured providers",
@@ -8188,35 +8247,69 @@ function AgentTree({ root, theme }) {
 }
 function T2Row({ node, theme, isLast }) {
   const connector = isLast ? "\u2514\u2500 " : "\u251C\u2500 ";
-  const t3Active = countByRoleAndStatus(node, "T3", "ACTIVE");
-  const t3Total = countByRole(node, "T3");
-  const workerSuffix = t3Total > 0 ? `  T3\xD7${t3Total}` : "";
+  const t3Nodes = (node.children ?? []).filter((c) => c.role === "T3");
+  const t3ActiveCount = t3Nodes.filter((c) => c.status === "ACTIVE").length;
   const label = stripRolePrefix(node.label, node.role);
   const action = node.currentAction ? `  ${node.currentAction.slice(0, 38)}` : "";
+  return /* @__PURE__ */ jsxs(Box, { flexDirection: "column", children: [
+    /* @__PURE__ */ jsxs(Box, { children: [
+      /* @__PURE__ */ jsxs(Text, { color: theme.colors.muted, children: [
+        "  ",
+        connector
+      ] }),
+      /* @__PURE__ */ jsx(Text, { color: theme.colors.t2Color, bold: true, children: "[T2]" }),
+      /* @__PURE__ */ jsxs(Text, { color: theme.colors.foreground, children: [
+        " ",
+        label
+      ] }),
+      node.status === "ACTIVE" && /* @__PURE__ */ jsxs(Fragment, { children: [
+        action ? /* @__PURE__ */ jsx(Text, { color: theme.colors.muted, children: action }) : null,
+        /* @__PURE__ */ jsx(Text, { children: " " }),
+        /* @__PURE__ */ jsx(Spinner, { type: "dots" }),
+        t3ActiveCount > 0 ? /* @__PURE__ */ jsxs(Text, { color: theme.colors.muted, children: [
+          " (",
+          t3ActiveCount,
+          " running)"
+        ] }) : null
+      ] }),
+      node.status === "COMPLETED" && /* @__PURE__ */ jsx(Text, { color: theme.colors.success, children: " \u2714" }),
+      node.status === "FAILED" && /* @__PURE__ */ jsx(Text, { color: theme.colors.error, children: " \u2718" }),
+      node.status === "ESCALATED" && /* @__PURE__ */ jsx(Text, { color: theme.colors.warning, children: " \u25B2" })
+    ] }),
+    t3Nodes.map((t3, idx) => /* @__PURE__ */ jsx(
+      T3Row,
+      {
+        node: t3,
+        theme,
+        isLast: idx === t3Nodes.length - 1,
+        parentIsLast: isLast
+      },
+      t3.id
+    ))
+  ] });
+}
+function T3Row({ node, theme, isLast, parentIsLast }) {
+  const indent = parentIsLast ? "      " : "  \u2502   ";
+  const connector = isLast ? "\u2514\u2500 " : "\u251C\u2500 ";
+  const label = stripRolePrefix(node.label, node.role);
+  const action = node.currentAction ? ` ${node.currentAction.slice(0, 42)}` : "";
   return /* @__PURE__ */ jsxs(Box, { children: [
     /* @__PURE__ */ jsxs(Text, { color: theme.colors.muted, children: [
-      "  ",
+      indent,
       connector
     ] }),
-    /* @__PURE__ */ jsx(Text, { color: theme.colors.t2Color, bold: true, children: "[T2]" }),
-    /* @__PURE__ */ jsxs(Text, { color: theme.colors.foreground, children: [
+    /* @__PURE__ */ jsx(Text, { color: theme.colors.t3Color, children: "[T3]" }),
+    /* @__PURE__ */ jsxs(Text, { color: theme.colors.muted, children: [
       " ",
       label
     ] }),
-    workerSuffix ? /* @__PURE__ */ jsx(Text, { color: theme.colors.muted, children: workerSuffix }) : null,
     node.status === "ACTIVE" && /* @__PURE__ */ jsxs(Fragment, { children: [
       action ? /* @__PURE__ */ jsx(Text, { color: theme.colors.muted, children: action }) : null,
       /* @__PURE__ */ jsx(Text, { children: " " }),
-      /* @__PURE__ */ jsx(Spinner, { type: "dots" }),
-      t3Active > 0 ? /* @__PURE__ */ jsxs(Text, { color: theme.colors.muted, children: [
-        " (",
-        t3Active,
-        " running)"
-      ] }) : null
+      /* @__PURE__ */ jsx(Spinner, { type: "dots" })
     ] }),
     node.status === "COMPLETED" && /* @__PURE__ */ jsx(Text, { color: theme.colors.success, children: " \u2714" }),
-    node.status === "FAILED" && /* @__PURE__ */ jsx(Text, { color: theme.colors.error, children: " \u2718" }),
-    node.status === "ESCALATED" && /* @__PURE__ */ jsx(Text, { color: theme.colors.warning, children: " \u25B2" })
+    node.status === "FAILED" && /* @__PURE__ */ jsx(Text, { color: theme.colors.error, children: " \u2718" })
   ] });
 }
 function hasActiveOrFailed(node) {
@@ -8828,7 +8921,7 @@ function Repl({ config, workspacePath, themeName, initialPrompt, identityName })
     treeNodesRef.current.set(event.tierId, node);
     const store = storeRef.current;
     if (store) {
-      const runtimeNode = { tierId: node.id, sessionId: sessionIdRef.current, parentId: node.parentId, role: node.role, label: node.label, status: node.status, currentAction: node.currentAction, progressPct: node.progressPct, updatedAt: (/* @__PURE__ */ new Date()).toISOString(), workspacePath, isGlobal: false };
+      const runtimeNode = { tierId: node.id, sessionId: sessionIdRef.current, parentId: node.parentId, role: node.role, label: node.label, status: node.status, currentAction: node.currentAction, progressPct: node.progressPct, updatedAt: (/* @__PURE__ */ new Date()).toISOString(), workspacePath, isGlobal: false, output: event.output };
       store.upsertRuntimeNode(runtimeNode);
       globalStoreRef.current?.upsertRuntimeNode({ ...runtimeNode, isGlobal: true });
     }
@@ -8839,7 +8932,6 @@ function Repl({ config, workspacePath, themeName, initialPrompt, identityName })
   useEffect(() => {
     const originalWarn = console.warn;
     const originalLog = console.log;
-    process.stdout.write("\x1Bc");
     console.warn = (...args) => {
       const msg = args.join(" ");
       if (msg.includes("non-text parts") || msg.includes("functionCall")) return;
@@ -9356,8 +9448,8 @@ Use /identity <name|id> to switch.`;
   const statusHeight = agentTreeHeight + timelineHeight;
   const costHeight = state.showCost ? 6 : 0;
   const approvalHeight = state.approvalRequest ? 12 : 0;
-  const slashVisibleCount = Math.min(SLASH_PAGE_SIZE, slashEntries.length);
-  const slashHeight = slashVisibleCount > 0 ? slashVisibleCount + 2 : 0;
+  Math.min(SLASH_PAGE_SIZE, slashEntries.length);
+  const slashHeight = isTypingCommand ? SLASH_PAGE_SIZE + 2 : 0;
   const chromeHeight = statusHeight + costHeight + approvalHeight + slashHeight + 7;
   const availableHeight = Math.max(4, height - chromeHeight);
   const allLines = formatToLines(
@@ -9769,8 +9861,21 @@ function wizardReducer(state, action) {
         addingAnotherCompat: false
       };
     }
+    case "ADD_OLLAMA": {
+      const newEntry = {
+        id: randomUUID(),
+        type: "ollama",
+        label: `Ollama endpoint ${state.entries.filter((e) => e.type === "ollama").length + 1}`
+      };
+      return {
+        ...state,
+        entries: [...state.entries, newEntry],
+        currentEntryIdx: state.entries.length,
+        addingAnotherOllama: false
+      };
+    }
     case "SKIP_MORE":
-      return { ...state, addingAnotherAzure: false, addingAnotherCompat: false, step: "FETCH_MODELS", currentEntryIdx: 0 };
+      return { ...state, addingAnotherAzure: false, addingAnotherCompat: false, addingAnotherOllama: false, step: "FETCH_MODELS", currentEntryIdx: 0 };
     case "GO_FETCH":
       return { ...state, step: "FETCH_MODELS", fetchLog: [], fetchedModels: [] };
     case "SET_FETCH_LOG":
@@ -9803,6 +9908,7 @@ function SetupWizard({ workspacePath, onComplete }) {
     currentEntryIdx: 0,
     addingAnotherAzure: false,
     addingAnotherCompat: false,
+    addingAnotherOllama: false,
     fetchedModels: [],
     fetchLog: [],
     tierT1: "auto",
@@ -9902,13 +10008,16 @@ function SetupWizard({ workspacePath, onComplete }) {
   }, [state.step, state.entries, state.tierT1, state.tierT2, state.tierT3, workspacePath, onComplete, exit]);
   useInput((_input, key) => {
     if (state.step === "PROVIDER_SELECT") {
-      if (key.upArrow) setProviderCursor((p) => Math.max(0, p - 1));
-      if (key.downArrow) setProviderCursor((p) => Math.min(providerOrder.length - 1, p + 1));
+      if (key.upArrow) setProviderCursor((p) => p <= 0 ? providerOrder.length - 1 : p - 1);
+      if (key.downArrow) setProviderCursor((p) => p >= providerOrder.length - 1 ? 0 : p + 1);
       if (_input === " ") dispatch({ type: "TOGGLE_PROVIDER", provider: providerOrder[providerCursor] });
       if (_input === "a") dispatch({ type: "TOGGLE_ALL" });
       if (_input === "i") dispatch({ type: "INVERT_SELECTION" });
       if (key.return) {
-        if (state.selectedTypes.size === 0) return;
+        if (state.selectedTypes.size === 0) {
+          dispatch({ type: "SET_ERROR", error: "Please select at least one provider using <space> before pressing <enter>." });
+          return;
+        }
         dispatch({ type: "CONFIRM_PROVIDERS" });
         const firstType = [...state.selectedTypes][0];
         setFieldStage(firstType === "azure" ? "deploymentName" : firstType === "openai-compatible" ? "label" : firstType === "ollama" ? "baseUrl" : "apiKey");
@@ -9965,11 +10074,7 @@ function SetupWizard({ workspacePath, onComplete }) {
     } else if (currentEntry.type === "ollama") {
       dispatch({ type: "SET_ENTRY_FIELD", field: "baseUrl", value: val || "http://localhost:11434" });
       setFieldBuffer("");
-      const nextEntry = state.entries[state.currentEntryIdx + 1];
-      if (nextEntry) {
-        setFieldStage(nextEntry.type === "azure" ? "deploymentName" : nextEntry.type === "openai-compatible" ? "label" : nextEntry.type === "ollama" ? "baseUrl" : "apiKey");
-      }
-      dispatch({ type: "NEXT_ENTRY" });
+      setFieldStage("askMore");
     } else {
       dispatch({ type: "SET_ENTRY_FIELD", field: "apiKey", value: val });
       setFieldBuffer("");
@@ -10010,7 +10115,7 @@ function SetupWizard({ workspacePath, onComplete }) {
       return /* @__PURE__ */ jsxs(Box, { flexDirection: "column", paddingX: 2, paddingY: 1, children: [
         /* @__PURE__ */ jsxs(Box, { marginBottom: 1, children: [
           /* @__PURE__ */ jsx(Text, { color: "magenta", bold: true, children: "? " }),
-          /* @__PURE__ */ jsx(Text, { bold: true, children: isAzure ? "Add another Azure deployment? (y/n)" : "Add another custom endpoint? (y/n)" })
+          /* @__PURE__ */ jsx(Text, { bold: true, children: isAzure ? "Add another Azure deployment? (y/n)" : isOllama ? "Add another Ollama endpoint? (y/n)" : "Add another custom endpoint? (y/n)" })
         ] }),
         /* @__PURE__ */ jsx(Box, { children: /* @__PURE__ */ jsx(
           SelectInput,
@@ -10024,8 +10129,9 @@ function SetupWizard({ workspacePath, onComplete }) {
             onSelect: (item) => {
               if (item.value === "yes") {
                 if (isAzure) dispatch({ type: "ADD_AZURE" });
+                else if (isOllama) dispatch({ type: "ADD_OLLAMA" });
                 else dispatch({ type: "ADD_COMPAT" });
-                setFieldStage(isAzure ? "deploymentName" : "label");
+                setFieldStage(isAzure ? "deploymentName" : isOllama ? "baseUrl" : "label");
                 setFieldBuffer("");
               } else {
                 const nextEntry = state.entries[state.currentEntryIdx + 1];
@@ -10041,7 +10147,7 @@ function SetupWizard({ workspacePath, onComplete }) {
       ] });
     }
     const prompt = isAzure && fieldStage === "deploymentName" ? `Azure deployment name (${currentEntry.label})` : isAzure && fieldStage === "baseUrl" ? `Azure endpoint URL` : isCompat && fieldStage === "label" ? `Name for this endpoint (e.g. Groq)` : isCompat && fieldStage === "baseUrl" ? `Base URL (e.g. https://api.groq.com/openai/v1)` : isOllama ? `Ollama URL (Enter for http://localhost:11434)` : `${currentEntry.label} API Key`;
-    const isMasked = fieldStage === "apiKey";
+    const isMasked = fieldStage === "apiKey" && !isOllama;
     return /* @__PURE__ */ jsxs(Box, { flexDirection: "column", paddingX: 2, paddingY: 1, children: [
       /* @__PURE__ */ jsxs(Box, { marginBottom: 1, children: [
         /* @__PURE__ */ jsx(Text, { color: "magenta", bold: true, children: "? " }),
@@ -10935,8 +11041,8 @@ var DashboardServer = class {
       void (async () => {
         const cascade = new Cascade(this.config, this.workspacePath, this.store);
         cascade.on("stream:token", (e) => {
-          this.socket.broadcast("stream:token", { sessionId, token: e.text });
-          this.socket.broadcastToRoom(`session:${sessionId}`, "stream:token", { sessionId, token: e.text });
+          this.socket.broadcast("stream:token", { sessionId, tierId: e.tierId, text: e.text });
+          this.socket.broadcastToRoom(`session:${sessionId}`, "stream:token", { sessionId, tierId: e.tierId, text: e.text });
         });
         cascade.on("tier:status", (e) => {
           this.socket.broadcast("tier:status", { sessionId, ...e });
@@ -11372,6 +11478,9 @@ async function startRepl(options) {
     await cm.load();
     config = cm.getConfig();
   }
+  if (process.stdout.isTTY) {
+    process.stdout.write("\x1B[2J\x1B[H");
+  }
   const { waitUntilExit } = render(
     React5.createElement(Repl, {
       config,