@freesyntax/notch-cli 0.5.22 → 0.5.23

package/dist/index.js

@@ -3,15 +3,17 @@ import "./chunk-4HPRBCSY.js";
 import {
   MCPClient,
   buildToolMap,
-  describeTools,
+  describeToolSchemas,
+  disconnectMCPServers,
   drainTailNotifications,
+  initMCPServers,
   listBuiltinAgents,
   nextSubagentId,
   parseMCPConfig,
   pollPendingAgents,
   setCurrentSurface,
   spawnSubagent
-} from "./chunk-EPSOOCNB.js";
+} from "./chunk-474TAHDN.js";
 import {
   Rollout,
   generateSessionId,
@@ -29,13 +31,12 @@ import "./chunk-O6AKZ4OH.js";
 import {
   loadConfig,
   persistConfigPatch
-} from "./chunk-J66N6AFH.js";
+} from "./chunk-UHK6SI4H.js";
 import "./chunk-KCAR5DOB.js";
 import {
   ByokMissingApiKeyError,
   ByokMissingBaseUrlError,
   MODEL_CATALOG,
-  MODEL_IDS,
   MissingApiKeyError,
   findByokProvider,
   isByokRef,
@@ -43,9 +44,11 @@ import {
   listByokProviders,
   modelSupportsImages,
   parseByokRef,
+  recordShadowUsage,
   resolveModel,
   validateConfig
-} from "./chunk-JXQ4HZ47.js";
+} from "./chunk-JVFOAPYV.js";
+import "./chunk-GFVLHUSS.js";
 import "./chunk-6CZCFY6H.js";
 import "./chunk-6U3ZAGYA.js";
 import "./chunk-FFB7GK3Y.js";
@@ -618,6 +621,304 @@ function microcompact(messages, opts2 = {}) {
   return { messages: out, cleared, savedChars };
 }
 
+// src/agent/correction.ts
+var CORRECTION_PRESETS = {
+  /** No correction — for capable models (Claude, GPT-4, etc.) */
+  disabled: () => ({
+    intentGate: false,
+    maxCorrectionAttempts: 0,
+    useTemplateFormat: false
+  }),
+  /** Light correction — for mid-tier models (Qwen 14B+, Llama 70B) */
+  capable: () => ({
+    intentGate: false,
+    maxCorrectionAttempts: 1,
+    useTemplateFormat: false
+  }),
+  /** Full correction — for small models (7B-14B) */
+  smallModel: () => ({
+    intentGate: true,
+    maxCorrectionAttempts: 2,
+    useTemplateFormat: false
+  }),
+  /** Maximum scaffolding — for tiny models (sub-7B) */
+  tinyModel: () => ({
+    intentGate: true,
+    maxCorrectionAttempts: 3,
+    useTemplateFormat: true
+  })
+};
+var INTENT_GATE_PROMPT = `You are a classifier. Given the user's message and the list of available tools, decide: does this message require calling a tool to answer properly?
+
+Reply with EXACTLY one word: YES or NO
+
+Available tools: {tool_names}
+
+User message: {user_message}`;
+async function intentGate(model, userMessage, toolNames) {
+  const { generateText: generateText3 } = await import("ai");
+  const prompt = INTENT_GATE_PROMPT.replace("{tool_names}", toolNames.join(", ")).replace("{user_message}", userMessage.slice(0, 500));
+  try {
+    const result = await generateText3({
+      model,
+      messages: [{ role: "user", content: prompt }],
+      maxTokens: 10,
+      temperature: 0
+    });
+    const answer = result.text.trim().toUpperCase();
+    if (answer.startsWith("YES")) return "yes";
+    if (answer.startsWith("NO")) return "no";
+    return "unclear";
+  } catch {
+    return "unclear";
+  }
+}
+function looksLikeFailedToolCall(text, toolNames) {
+  if (!text || text.length < 10) return false;
+  let weakSignals = 0;
+  let strongSignals = 0;
+  const lowerText = text.toLowerCase();
+  for (const name of toolNames) {
+    if (lowerText.includes(name.toLowerCase())) {
+      weakSignals++;
+      break;
+    }
+  }
+  const jsonPatterns = [
+    /\{\s*"name"\s*:/,
+    /\{\s*"tool"\s*:/,
+    /\{\s*"function"\s*:/,
+    /"arguments"\s*:\s*\{/,
+    /"parameters"\s*:\s*\{/
+  ];
+  for (const pat of jsonPatterns) {
+    if (pat.test(text)) {
+      strongSignals++;
+      break;
+    }
+  }
+  const intentPatterns = [
+    /\b(?:i'll|let me|i need to|i should|i want to|i will)\s+(?:use|call|invoke|run|execute)\b/i,
+    /\b(?:calling|using|invoking|running)\s+(?:the\s+)?(?:tool|function)\b/i
+  ];
+  for (const pat of intentPatterns) {
+    if (pat.test(text)) {
+      weakSignals++;
+      break;
+    }
+  }
+  const xmlPatterns = [
+    /<tool_call/i,
+    /<function_call/i,
+    /<tool_use/i,
+    /<\|tool▁call\|>/,
+    /<\|plugin\|>/
+  ];
+  for (const pat of xmlPatterns) {
+    if (pat.test(text)) {
+      strongSignals++;
+      break;
+    }
+  }
+  if (/```(?:json|tool)?\s*\n?\s*\{/.test(text)) {
+    weakSignals++;
+  }
+  return strongSignals >= 1 || weakSignals >= 2;
+}
+var CORRECTION_PROMPTS = [
+  // Attempt 1: XML format (most models have seen this in training)
+  `Your previous response tried to call a tool but the format was wrong.
+
+Your output was:
+---
+{raw_output}
+---
+
+To call a tool, you MUST use this exact format:
+<tool_call>
+{"name": "tool_name", "arguments": {"param1": "value1"}}
+</tool_call>
+
+Available tools: {tool_names}
+
+Rewrite your response with the correct tool call format. Output ONLY the tool call, nothing else.`,
+  // Attempt 2: bare JSON (simpler)
+  `Your output was not a valid tool call. Rewrite it as a single JSON object:
+
+{"name": "TOOL_NAME", "arguments": {"key": "value"}}
+
+Available tools: {tool_names}
+Your failed output: {raw_output_short}
+
+Reply with ONLY the JSON object.`,
+  // Attempt 3: ask directly (simplest possible)
+  `Which tool do you want to call? Reply with the tool name and arguments as JSON.
+Tools: {tool_names}
+JSON:`
+];
+async function correctionPass(model, rawOutput, toolNames, conversationMessages, maxAttempts) {
+  const { generateText: generateText3 } = await import("ai");
+  for (let i = 0; i < Math.min(maxAttempts, CORRECTION_PROMPTS.length); i++) {
+    const promptTemplate = CORRECTION_PROMPTS[i];
+    if (!promptTemplate) continue;
+    const prompt = promptTemplate.replace("{raw_output}", rawOutput.slice(0, 1500)).replace("{raw_output_short}", rawOutput.slice(0, 500)).replace("{tool_names}", toolNames.join(", "));
+    try {
+      const result = await generateText3({
+        model,
+        messages: [
+          ...conversationMessages.slice(-4),
+          // keep recent context
+          { role: "user", content: prompt }
+        ],
+        maxTokens: 500,
+        temperature: 0
+      });
+      const parsed = extractToolCallsFromText(result.text, toolNames);
+      if (parsed.length > 0) return parsed;
+    } catch {
+    }
+  }
+  return null;
+}
+function extractTemplateToolCalls(text, toolNames) {
+  const calls = [];
+  const toolNameSet = new Set(toolNames.map((n) => n.toLowerCase()));
+  const lines = text.split("\n");
+  let currentTool = null;
+  let currentArgs = {};
+  for (const line of lines) {
+    const trimmed = line.trim();
+    const toolMatch = trimmed.match(/^TOOL:\s*(.+)$/i);
+    if (toolMatch && toolMatch[1]) {
+      if (currentTool && toolNameSet.has(currentTool.toLowerCase())) {
+        calls.push({
+          id: `correction-${Date.now()}-${calls.length}`,
+          name: currentTool,
+          args: currentArgs
+        });
+      }
+      currentTool = toolMatch[1].trim();
+      currentArgs = {};
+      continue;
+    }
+    const argMatch = trimmed.match(/^ARG_(\w+):\s*(.+)$/i);
+    if (argMatch && currentTool && argMatch[1] && argMatch[2]) {
+      const key = argMatch[1];
+      const rawValue = argMatch[2].trim();
+      currentArgs[key] = coerceValue(rawValue);
+    }
+  }
+  if (currentTool && toolNameSet.has(currentTool.toLowerCase())) {
+    calls.push({
+      id: `correction-${Date.now()}-${calls.length}`,
+      name: currentTool,
+      args: currentArgs
+    });
+  }
+  return calls;
+}
+function coerceValue(raw) {
+  if (raw === "true") return true;
+  if (raw === "false") return false;
+  if (raw === "null") return null;
+  const num = Number(raw);
+  if (!Number.isNaN(num) && raw !== "") return num;
+  if (raw.startsWith("{") && raw.endsWith("}") || raw.startsWith("[") && raw.endsWith("]")) {
+    try {
+      return JSON.parse(raw);
+    } catch {
+    }
+  }
+  return raw;
+}
+function extractToolCallsFromText(text, toolNames) {
+  const toolNameSet = new Set(toolNames.map((n) => n.toLowerCase()));
+  const xmlRegex = /<tool_call>\s*([\s\S]*?)\s*<\/tool_call>/gi;
+  let match2;
+  const xmlCalls = [];
+  while ((match2 = xmlRegex.exec(text)) !== null) {
+    const captured = match2[1];
+    if (captured) {
+      const parsed = tryParseToolJson(captured, toolNameSet);
+      if (parsed) xmlCalls.push(parsed);
+    }
+  }
+  if (xmlCalls.length > 0) return xmlCalls;
+  const codeBlockRegex = /```(?:json|tool)?\s*\n?([\s\S]*?)\n?\s*```/g;
+  const codeCalls = [];
+  while ((match2 = codeBlockRegex.exec(text)) !== null) {
+    const captured = match2[1];
+    if (captured) {
+      const parsed = tryParseToolJson(captured, toolNameSet);
+      if (parsed) codeCalls.push(parsed);
+    }
+  }
+  if (codeCalls.length > 0) return codeCalls;
+  const jsonRegex = /\{[^{}]*"name"\s*:\s*"[^"]+"\s*[,}][\s\S]*?\}/g;
+  const jsonCalls = [];
+  while ((match2 = jsonRegex.exec(text)) !== null) {
+    const parsed = tryParseToolJson(match2[0], toolNameSet);
+    if (parsed) jsonCalls.push(parsed);
+  }
+  if (jsonCalls.length > 0) return jsonCalls;
+  const templateCalls = extractTemplateToolCalls(text, toolNames);
+  if (templateCalls.length > 0) return templateCalls;
+  return [];
+}
+function tryParseToolJson(jsonStr, toolNameSet) {
+  try {
+    const obj = JSON.parse(jsonStr.trim());
+    const name = obj.name ?? obj.tool ?? obj.function;
+    if (!name || typeof name !== "string") return null;
+    if (!toolNameSet.has(name.toLowerCase())) return null;
+    const args = obj.arguments ?? obj.parameters ?? obj.args ?? {};
+    return {
+      id: `correction-${Date.now()}-${Math.random().toString(36).slice(2, 8)}`,
+      name,
+      args: typeof args === "object" ? args : {}
+    };
+  } catch {
+    return null;
+  }
+}
+var CAPABLE_PATTERNS = [
+  /claude/i,
+  /gpt-4/i,
+  /gpt-3\.5/i,
+  /o[1-4]/i,
+  /gemini-(?:pro|ultra|2)/i,
+  /deepseek-(?:v3|r1)/i
+];
+var SMALL_PATTERNS = [
+  /qwen.*(?:7b|14b|32b)/i,
+  /llama.*(?:8b|13b|70b)/i,
+  /mistral.*(?:7b|8x7b)/i,
+  /gemma.*(?:7b|9b|27b)/i,
+  /codestral/i,
+  /deepseek-coder/i
+];
+var TINY_PATTERNS = [
+  /qwen.*(?:0\.5b|1\.5b|3b|4b)/i,
+  /llama.*(?:1b|3b)/i,
+  /gemma.*(?:2b|4b)/i,
+  /phi.*(?:1|2|3)/i,
+  /tinyllama/i,
+  /stablelm/i
+];
+function autoDetectPreset(modelId) {
+  if (!modelId) return CORRECTION_PRESETS.capable();
+  for (const pat of CAPABLE_PATTERNS) {
+    if (pat.test(modelId)) return CORRECTION_PRESETS.disabled();
+  }
+  for (const pat of TINY_PATTERNS) {
+    if (pat.test(modelId)) return CORRECTION_PRESETS.tinyModel();
+  }
+  for (const pat of SMALL_PATTERNS) {
+    if (pat.test(modelId)) return CORRECTION_PRESETS.smallModel();
+  }
+  return CORRECTION_PRESETS.capable();
+}
+
 // src/agent/loop.ts
 function getErrorSignature(toolName, result) {
   return {
@@ -644,7 +945,18 @@ async function runAgentLoop(messages, config) {
   let wasCompressed = false;
   const recentErrors = [];
   const MAX_REPEATED_ERRORS = 3;
+  const correctionCfg = config.correction ?? autoDetectPreset(config.modelId ?? "");
+  let totalCorrectionAttempts = 0;
+  let intentResult = "unclear";
   let history = [...messages];
+  const toolNames = Object.keys(tools);
+  const templateFormatHint = correctionCfg.useTemplateFormat && toolNames.length > 0 ? [
+    "",
+    "If native tool calling fails, use this fallback format with no extra prose:",
+    "TOOL: <tool name>",
+    "ARG_<argument_name>: <argument value>",
+    `Available tool names: ${toolNames.join(", ")}`
+  ].join("\n") : "";
   await config.toolContext.runHook?.("pre-compact", { messageCount: history.length });
   history = await autoCompress(history, config.model, contextWindow, () => {
     wasCompressed = true;
@@ -653,6 +965,13 @@ async function runAgentLoop(messages, config) {
   if (wasCompressed) {
     await config.toolContext.runHook?.("post-compact", { messageCount: history.length });
   }
+  if (correctionCfg.intentGate && toolNames.length > 0) {
+    const lastUserMsg = [...history].reverse().find((m) => m.role === "user");
+    const userText = typeof lastUserMsg?.content === "string" ? lastUserMsg.content : "";
+    if (userText) {
+      intentResult = await intentGate(config.model, userText, toolNames);
+    }
+  }
   while (iterations < maxIter) {
     iterations++;
     const mc = microcompact(history, { keepLastN: 6, threshold: 50 });
@@ -666,7 +985,7 @@ async function runAgentLoop(messages, config) {
     let streamUsage = null;
     const result = streamText({
       model: config.model,
-      system: config.systemPrompt,
+      system: `${config.systemPrompt}${templateFormatHint}`,
       messages: history,
       tools,
       maxSteps: 1
@@ -705,8 +1024,67 @@ async function runAgentLoop(messages, config) {
     if (streamUsage) {
       totalPromptTokens += streamUsage.promptTokens ?? 0;
       totalCompletionTokens += streamUsage.completionTokens ?? 0;
+      void recordShadowUsage(
+        config.model,
+        streamUsage
+      );
     }
     totalToolCalls += toolCalls.length;
+    if (toolCalls.length === 0 && fullText && correctionCfg.maxCorrectionAttempts > 0 && toolNames.length > 0) {
+      const directExtracted = extractToolCallsFromText(fullText, toolNames);
+      const templateExtracted = correctionCfg.useTemplateFormat ? extractTemplateToolCalls(fullText, toolNames) : [];
+      const shouldCorrect = looksLikeFailedToolCall(fullText, toolNames) || directExtracted.length > 0 || templateExtracted.length > 0 || intentResult === "yes" && totalToolCalls === 0;
+      if (shouldCorrect) {
+        const corrected = directExtracted.length > 0 ? directExtracted : templateExtracted.length > 0 ? templateExtracted : await correctionPass(
+          config.model,
+          fullText,
+          toolNames,
+          history,
+          correctionCfg.maxCorrectionAttempts
+        );
+        totalCorrectionAttempts++;
+        if (corrected && corrected.length > 0) {
+          config.onCorrection?.(totalCorrectionAttempts, true);
+          for (const tc of corrected) {
+            toolCalls.push({
+              toolCallId: tc.id,
+              toolName: tc.name,
+              args: tc.args
+            });
+          }
+          totalToolCalls += corrected.length;
+          for (const tc of corrected) {
+            const toolDef = tools[tc.name];
+            if (!toolDef) {
+              toolResults.push({
+                toolCallId: tc.id,
+                result: { content: `Tool "${tc.name}" not found.`, isError: true }
+              });
+              continue;
+            }
+            try {
+              const result2 = await toolDef.execute(tc.args);
+              toolResults.push({ toolCallId: tc.id, result: result2 });
+              config.onToolCall?.(tc.name, tc.args);
+              config.onToolResult?.(
+                tc.name,
+                result2?.content ?? String(result2),
+                result2?.isError ?? false
+              );
+            } catch (err) {
+              const msg = err instanceof Error ? err.message : String(err);
+              toolResults.push({
+                toolCallId: tc.id,
+                result: { content: `Error: ${msg}`, isError: true }
+              });
+              config.onToolResult?.(tc.name, `Error: ${msg}`, true);
+            }
+          }
+        } else {
+          config.onCorrection?.(totalCorrectionAttempts, false);
+        }
+      }
+    }
     if (toolCalls.length > 0) {
       let hasRepeatedError = false;
       for (const tr of toolResults) {
@@ -772,6 +1150,7 @@ async function runAgentLoop(messages, config) {
       iterations,
       toolCallCount: totalToolCalls,
       compressed: wasCompressed,
+      correctionAttempts: totalCorrectionAttempts,
       usage: {
         promptTokens: totalPromptTokens,
         completionTokens: totalCompletionTokens,
@@ -787,6 +1166,7 @@ async function runAgentLoop(messages, config) {
     iterations,
     toolCallCount: totalToolCalls,
     compressed: wasCompressed,
+    correctionAttempts: totalCorrectionAttempts,
     usage: {
       promptTokens: totalPromptTokens,
       completionTokens: totalCompletionTokens,
@@ -794,7 +1174,7 @@ async function runAgentLoop(messages, config) {
     }
   };
 }
-async function buildSystemPrompt(projectRoot, _modelId) {
+async function buildSystemPrompt(projectRoot, _modelId, toolContext) {
   const sections = [
     // =========================================================
     // CACHEABLE PREFIX — stable across turns, sessions, users.
@@ -805,7 +1185,7 @@ async function buildSystemPrompt(projectRoot, _modelId) {
       () => [
         "You are Notch, an expert AI coding assistant built by Driftrail.",
         "You help developers write, debug, refactor, and understand code.",
-        "You have access to tools for reading/writing files, running shell commands, searching code, and git operations."
+        "You have access to the tools listed in the Tool Schemas section."
       ].join("\n")
     ),
     safeSection(
@@ -841,8 +1221,11 @@ async function buildSystemPrompt(projectRoot, _modelId) {
         "- When the user asks for a fix, fix the root cause, not the symptom."
       ].join("\n")
     ),
-    safeSection("tools-list", () => `## Available Tools
-${describeTools()}`),
+    DANGEROUS_uncachedSystemPromptSection(
+      "tool-schemas",
+      () => describeToolSchemas(toolContext),
+      "The active tool set depends on coordinator mode, MCP servers, plugins, and permissions for this session."
+    ),
     safeSection("builtin-agents", () => {
       try {
         const agents = listBuiltinAgents();
@@ -1671,8 +2054,7 @@ ${toolContext.cwd}
 ## Repository
 ${repoContext}
 
-## Available Tools
-${describeTools()}
+${describeToolSchemas(toolContext)}
 
 ## Instructions
 1. Read relevant files to understand the current state
@@ -5403,20 +5785,20 @@ async function ensureGitignoreEntries(p, entries, sectionTitle, ctx) {
   ctx.log(chalk8.green(`  Updated .gitignore (+${missing.length})`));
 }
 var DEFAULT_CONFIG = {
-  model: "notch-pyre",
+  model: "openrouter/anthropic/claude-sonnet-4-6",
   temperature: 0.3,
   maxIterations: 25,
   useRepoMap: true,
   renderMarkdown: true,
   permissionMode: "auto",
-  // BYOK (Bring-Your-Own-Key) passthrough — disabled by default.
-  // Uncomment `byok` below (and remove the "_" prefix) to route every
+  // Provider passthrough — disabled by default.
+  // Uncomment `provider` below (and remove the "_" prefix) to route every
   // request to an OpenAI-compatible endpoint (OpenAI / Anthropic /
-  // OpenRouter / Together / Fireworks / Groq / Ollama / vLLM / LM Studio
+  // OpenRouter / Google / DeepSeek / Together / Fireworks / Groq / Ollama / vLLM / LM Studio
   // / custom). Providers read their key from the corresponding env var —
   // e.g. OPENROUTER_API_KEY, ANTHROPIC_API_KEY, OPENAI_API_KEY.
   // Run `notch --list-providers` to see all built-in ids.
-  _byokExample: {
+  _providerExample: {
     provider: "openrouter",
     model: "anthropic/claude-sonnet-4-6"
   }
@@ -7766,8 +8148,7 @@ function buildCompleter(cwd) {
     }
     if (line.startsWith("/model ")) {
       const partial = line.slice(7);
-      const modelNames = MODEL_IDS.map((id) => id.replace("notch-", ""));
-      const allNames = [...MODEL_IDS, ...modelNames];
+      const allNames = listByokProviders().filter((p) => p.defaultModel).map((p) => `${p.id}/${p.defaultModel}`);
       const matches = allNames.filter((m) => m.startsWith(partial));
       return [matches.map((m) => `/model ${m}`), line];
     }
@@ -7818,10 +8199,9 @@ var SLASH_COMMANDS = [
   { name: "/quit", description: "Exit Notch", category: "Core" },
   // Model & Status
   { name: "/model", description: "Switch or list models", category: "Model" },
-  { name: "/model download", description: "Download a Notch model locally", category: "Model" },
-  { name: "/downloads", description: "Show local model download progress", category: "Model" },
   { name: "/status", description: "Check API endpoint health", category: "Model" },
-  { name: "/sync-keys", description: "Pull BYOK keys from freesyntax.dev", category: "Model" },
+  { name: "/providers", description: "List model providers", category: "Model" },
+  { name: "/sync-keys", description: "Pull provider keys from freesyntax.dev", category: "Model" },
   // Session
   { name: "/save", description: "Save current session", category: "Session" },
   { name: "/sessions", description: "List saved sessions", category: "Session" },
@@ -8352,22 +8732,21 @@ import fs20 from "fs/promises";
 import { createRequire as createRequire2 } from "module";
 var _require2 = createRequire2(import.meta.url);
 var VERSION = _require2("../package.json").version;
-var modelChoices = MODEL_IDS.join(", ");
 if (process.argv[2] === "update") {
   await runUpdateCli(process.argv.slice(3));
   process.exit(process.exitCode ?? 0);
 }
 if (process.argv[2] === "ollama") {
-  const { runOllamaCli } = await import("./ollama-launch-P5KBK7AJ.js");
+  const { runOllamaCli } = await import("./ollama-launch-3IKB2A3Z.js");
   const code = await runOllamaCli(process.argv.slice(3), process.cwd());
   process.exit(code);
 }
 if (process.argv[2] === "config") {
-  const { runConfigCli } = await import("./config-set-3IWEVZQ4.js");
+  const { runConfigCli } = await import("./config-set-5F4VK7IT.js");
   const code = await runConfigCli(process.argv.slice(3), process.cwd());
   process.exit(code);
 }
-var program = new Command().name("notch").description("Notch CLI \u2014 AI-powered coding assistant by Driftrail").version(VERSION).argument("[prompt...]", "One-shot prompt (runs once and exits)").option(`-m, --model <model>`, `Notch model (${modelChoices}) or BYOK ref like openrouter:anthropic/claude-sonnet-4-6`).option("--base-url <url>", "Override the backend base URL (Notch or BYOK)").option("--api-key <key>", "API key for the backend (prefer the env var: NOTCH_API_KEY / OPENAI_API_KEY / ANTHROPIC_API_KEY / OPENROUTER_API_KEY / ...)").option("--provider <id>", "BYOK provider id (openai, anthropic, openrouter, together, fireworks, groq, ollama, lmstudio, vllm, custom). Run --list-providers to see them all.").option("--list-providers", "List built-in BYOK providers and their API-key env vars, then exit").option("--no-repo-map", "Disable automatic repository mapping").option("--no-markdown", "Disable markdown rendering in output").option("--max-iterations <n>", "Max tool-call rounds per turn", "25").option("-y, --yes", "Auto-confirm destructive actions").option("--trust", "Trust mode \u2014 auto-allow all tool calls").option("--theme <theme>", `UI color theme (${THEME_IDS.join(", ")})`).option("--resume", "Resume the last session for this project").option("--session <id>", "Resume a specific session by ID").option("--cwd <dir>", "Set working directory").option("--json", "Emit JSONL event stream on stdout (headless/CI mode)").option("--output-schema <file>", "Path to JSON Schema constraining the final structured output").option("--output-last-message <file>", "Write the final assistant message to this file on exit").option("--guardian", "Enable Guardian: independent Solace-Lite risk scoring before every prompt-level tool call").option("--coordinator", "Coordinator mode: top-level agent can only spawn/continue/stop workers (plus read/grep/glob). All real work is delegated.").option("--no-auto-dream", "Disable the background memory-consolidation daemon (default: enabled in REPL)").option("--no-update", "Disable the background update check on launch (equivalent to NOTCH_AUTO_UPDATE=0)").option("--update-channel <name>", "npm dist-tag to follow for updates (latest | next | beta)").option(
+var program = new Command().name("notch").description("Notch CLI \u2014 AI-powered coding assistant by Driftrail").version(VERSION).argument("[prompt...]", "One-shot prompt (runs once and exits)").option("-m, --model <model>", "Model id, preferably provider/model (for example openrouter/anthropic/claude-sonnet-4-6)").option("--base-url <url>", "Override the provider base URL (for custom OpenAI-compatible endpoints)").option("--api-key <key>", "API key for the active provider (prefer provider env vars such as OPENROUTER_API_KEY)").option("--provider <id>", "Provider id (openai, anthropic, openrouter, google, deepseek, together, fireworks, groq, ollama, lmstudio, vllm, custom).").option("--list-providers", "List built-in providers and their API-key env vars, then exit").option("--no-repo-map", "Disable automatic repository mapping").option("--no-markdown", "Disable markdown rendering in output").option("--max-iterations <n>", "Max tool-call rounds per turn", "25").option("-y, --yes", "Auto-confirm destructive actions").option("--trust", "Trust mode \u2014 auto-allow all tool calls").option("--theme <theme>", `UI color theme (${THEME_IDS.join(", ")})`).option("--resume", "Resume the last session for this project").option("--session <id>", "Resume a specific session by ID").option("--cwd <dir>", "Set working directory").option("--json", "Emit JSONL event stream on stdout (headless/CI mode)").option("--output-schema <file>", "Path to JSON Schema constraining the final structured output").option("--output-last-message <file>", "Write the final assistant message to this file on exit").option("--guardian", "Enable Guardian risk scoring when a provider-backed guardian model is configured").option("--coordinator", "Coordinator mode: top-level agent can only spawn/continue/stop workers (plus read/grep/glob). All real work is delegated.").option("--no-auto-dream", "Disable the background memory-consolidation daemon (default: enabled in REPL)").option("--no-update", "Disable the background update check on launch (equivalent to NOTCH_AUTO_UPDATE=0)").option("--update-channel <name>", "npm dist-tag to follow for updates (latest | next | beta)").option(
   "--image <path>",
   "Attach an image (file path, URL, or data URL). Repeatable.",
   (val, prev) => prev ? [...prev, val] : [val],
@@ -8377,22 +8756,29 @@ var opts = program.opts();
 var promptArgs = program.args;
 function printByokProviderList() {
   const t = theme();
-  console.log(t.dim("\n  BYOK providers (point --provider at any of these):\n"));
-  const header = `  ${"id".padEnd(12)} ${"label".padEnd(20)} ${"env var".padEnd(22)} default model`;
+  console.log(t.dim("\n  Model providers (point --provider at any of these):\n"));
+  const providers = listByokProviders();
+  const idWidth = Math.max(12, ...providers.map((p) => p.id.length));
+  const labelWidth = Math.max(20, ...providers.map((p) => p.label.length));
+  const envWidth = Math.max(22, ...providers.map((p) => (p.apiKeyEnv || "(none)").length));
+  const keyWidth = 7;
+  const header = `  ${"id".padEnd(idWidth)} ${"label".padEnd(labelWidth)} ${"env var".padEnd(envWidth)} ${"key".padEnd(keyWidth)} default model`;
   console.log(t.dim(header));
   console.log(t.dim(`  ${"-".repeat(header.length - 2)}`));
-  for (const p of listByokProviders()) {
-    const keyPresent = p.apiKeyEnv ? process.env[p.apiKeyEnv] ? t.success("\u2713") : t.dim("\u2717") : t.dim("\u2013");
+  for (const p of providers) {
+    const keyLabel = p.apiKeyEnv ? process.env[p.apiKeyEnv] ? "set" : p.fallbackApiKey ? "local" : "missing" : "none";
+    const paddedKey = keyLabel.padEnd(keyWidth);
+    const keyPresent = keyLabel === "set" ? t.success(paddedKey) : keyLabel === "missing" ? t.dim(paddedKey) : t.dim(paddedKey);
     const envDisplay = p.apiKeyEnv || "(none)";
     console.log(
-      `  ${t.brand(p.id.padEnd(12))} ${p.label.padEnd(20)} ${envDisplay.padEnd(20)} ${keyPresent}  ${t.dim(p.defaultModel || "\u2014")}`
+      `  ${t.brand(p.id.padEnd(idWidth))} ${p.label.padEnd(labelWidth)} ${envDisplay.padEnd(envWidth)} ${keyPresent} ${t.dim(p.defaultModel || "\u2014")}`
     );
   }
   console.log("");
   console.log(t.dim("  Use it like:"));
   console.log(t.dim("    export OPENROUTER_API_KEY=sk-or-..."));
-  console.log(t.dim("    notch --provider openrouter --model anthropic/claude-opus-4-7"));
-  console.log(t.dim("    notch --model openrouter:anthropic/claude-opus-4-7   # equivalent"));
+  console.log(t.dim("    notch --provider openrouter --model anthropic/claude-sonnet-4-6"));
+  console.log(t.dim("    notch --model openrouter/anthropic/claude-sonnet-4-6   # equivalent"));
   console.log(t.dim("    notch --provider custom --base-url http://localhost:8000/v1 --model my-model"));
   console.log("");
 }
@@ -8414,7 +8800,8 @@ async function persistByokChoice(projectRoot, providerId, defaultModel) {
     } catch {
     }
     const idToPersist = providerId === "__custom__" ? "custom" : providerId;
-    current.byok = { provider: idToPersist, model: defaultModel };
+    current.provider = { provider: idToPersist, model: defaultModel };
+    delete current.byok;
     await fs20.writeFile(p, JSON.stringify(current, null, 2) + "\n", "utf-8");
   } catch {
   }
@@ -8428,14 +8815,12 @@ function interactiveModelPicker(activeModel) {
   return new Promise((resolve3) => {
     const t = theme();
     const rows = [];
-    rows.push({ kind: "notch-header" });
-    for (const id of MODEL_IDS) rows.push({ kind: "notch", id });
     rows.push({ kind: "byok-header" });
     for (const p of listByokProviders()) rows.push({ kind: "byok", provider: p });
-    const selectableIndexes = rows.map((r, i) => r.kind === "notch" || r.kind === "byok" ? i : -1).filter((i) => i >= 0);
+    const selectableIndexes = rows.map((r, i) => r.kind === "byok" ? i : -1).filter((i) => i >= 0);
     let cursor = selectableIndexes.find((i) => {
       const r = rows[i];
-      return r && r.kind === "notch" && r.id === activeModel;
+      return r && r.kind === "byok" && typeof activeModel === "string" && isByokRef(activeModel) && parseByokRef(activeModel).provider === r.provider.id;
     }) ?? selectableIndexes[0] ?? 0;
     const rowCount = rows.length;
     const headerLines = 2;
@@ -8452,46 +8837,31 @@ function interactiveModelPicker(activeModel) {
       console.log(t.dim("  Select a model (\u2191\u2193 to move, Enter to select, Esc to cancel)\n"));
       for (let i = 0; i < rows.length; i++) {
         const row = rows[i];
-        if (row.kind === "notch-header") {
-          console.log(`  ${t.dim("\u2500\u2500\u2500 Notch models (default) \u2500\u2500\u2500")}`);
-          continue;
-        }
         if (row.kind === "byok-header") {
-          console.log(`  ${t.dim("\u2500\u2500\u2500 BYOK providers (your own API key) \u2500\u2500\u2500")}`);
+          console.log(`  ${t.dim("--- Providers ---")}`);
           continue;
         }
         const isSelected = i === cursor;
         const pointer = isSelected ? t.brand("\u276F") : " ";
-        if (row.kind === "notch") {
-          const info = MODEL_CATALOG[row.id];
-          const isCurrent = row.id === activeModel;
-          const dot = isCurrent ? t.success("\u25CF") : " ";
-          const label = isSelected ? t.bold(info.label) : t.dim(info.label);
-          const size = t.dim(info.size);
-          const gpu = t.dim(info.gpu);
-          const ctx = t.dim(`${(info.contextWindow / 1024).toFixed(0)}K`);
-          console.log(`  ${pointer} ${dot} ${t.brand(row.id.replace("notch-", "").padEnd(12))} ${label.padEnd(20)} ${size.padEnd(6)} ${gpu.padEnd(12)} ${ctx}`);
-        } else {
-          const p = row.provider;
-          const isCurrent = typeof activeModel === "string" && isByokRef(activeModel) ? parseByokRef(activeModel).provider === p.id : false;
-          const dot = isCurrent ? t.success("\u25CF") : " ";
-          const envHit = p.apiKeyEnv ? Boolean(process.env[p.apiKeyEnv]) : false;
-          let syncHit = false;
-          if (!envHit) {
-            try {
-              const { loadSyncedByokKeysSync } = (init_auth(), __toCommonJS(auth_exports));
-              const synced = loadSyncedByokKeysSync();
-              const fromSync = synced?.keys[p.id];
-              syncHit = typeof fromSync === "string" && fromSync.length > 0;
-            } catch {
-            }
+        const p = row.provider;
+        const isCurrent = typeof activeModel === "string" && isByokRef(activeModel) ? parseByokRef(activeModel).provider === p.id : false;
+        const dot = isCurrent ? t.success("\u25CF") : " ";
+        const envHit = p.apiKeyEnv ? Boolean(process.env[p.apiKeyEnv]) : false;
+        let syncHit = false;
+        if (!envHit) {
+          try {
+            const { loadSyncedByokKeysSync } = (init_auth(), __toCommonJS(auth_exports));
+            const synced = loadSyncedByokKeysSync();
+            const fromSync = synced?.keys[p.id];
+            syncHit = typeof fromSync === "string" && fromSync.length > 0;
+          } catch {
           }
-          const keyPresent = p.apiKeyEnv ? envHit ? t.success("\u2713") : syncHit ? t.brand("\u2713") : t.dim("\u2717") : t.dim("\u2013");
-          const label = isSelected ? t.bold(p.label) : t.dim(p.label);
-          const envDisplay = t.dim((p.apiKeyEnv || "local").padEnd(22));
-          const defModel = t.dim(p.defaultModel ? p.defaultModel.slice(0, 34) : "\u2014");
-          console.log(`  ${pointer} ${dot} ${t.brand(p.id.padEnd(12))} ${label.padEnd(20)} ${envDisplay} ${keyPresent}  ${defModel}`);
         }
+        const keyPresent = p.apiKeyEnv ? envHit ? t.success("\u2713") : syncHit ? t.brand("\u2713") : t.dim("\u2717") : t.dim("\u2013");
+        const label = isSelected ? t.bold(p.label) : t.dim(p.label);
+        const envDisplay = t.dim((p.apiKeyEnv || "local").padEnd(22));
+        const defModel = t.dim(p.defaultModel ? p.defaultModel.slice(0, 34) : "\u2014");
+        console.log(`  ${pointer} ${dot} ${t.brand(p.id.padEnd(12))} ${label.padEnd(20)} ${envDisplay} ${keyPresent}  ${defModel}`);
       }
     };
     const render = (first) => {
@@ -8517,16 +8887,12 @@ function interactiveModelPicker(activeModel) {
       } else if (s === "\r" || s === "\n") {
         const row = rows[cursor];
         cleanup();
-        if (!row || row.kind !== "notch" && row.kind !== "byok") {
+        if (!row || row.kind !== "byok") {
           resolve3(null);
           return;
         }
-        if (row.kind === "notch") {
-          resolve3({ kind: "notch", id: row.id });
-        } else {
-          const modelRef = `${row.provider.id}:${row.provider.defaultModel}`;
-          resolve3({ kind: "byok", provider: row.provider, modelRef });
-        }
+        const modelRef = `${row.provider.id}/${row.provider.defaultModel}`;
+        resolve3({ kind: "byok", provider: row.provider, modelRef });
       } else if (s === "\x1B" || s === "") {
         cleanup();
         resolve3(null);
@@ -8543,13 +8909,11 @@ function interactiveModelPicker(activeModel) {
 function printHelp() {
   console.log(chalk30.gray(`
   Commands:
-    /model              \u2014 Show available models (Notch + BYOK)
-    /model <name>       \u2014 Switch model: /model pyre  OR  /model openrouter:anthropic/claude-sonnet-4-6
-    /model download <name> \u2014 Pull a Notch model's local weights in the background (HF Hub)
-    /downloads          \u2014 Show local download progress for this session
-    /sync-keys          \u2014 Pull BYOK keys you added on freesyntax.dev
-    /providers          \u2014 List built-in BYOK providers and whether their keys are set
-    /status             \u2014 Check backend health (Notch API or BYOK endpoint)
+    /model              \u2014 Pick a provider and default model
+    /model <ref>        \u2014 Switch model: /model openrouter/anthropic/claude-sonnet-4-6
+    /sync-keys          \u2014 Pull provider keys you added on freesyntax.dev
+    /providers          \u2014 List built-in providers and whether their keys are set
+    /status             \u2014 Check active provider health
     /undo               \u2014 Undo last file changes
     /usage              \u2014 Show token usage + context meter
     /cost               \u2014 Show estimated session cost
@@ -8656,16 +9020,16 @@ async function main() {
         const providerCount = Object.keys(synced.keys).length;
         if (providerCount > 0) {
           console.log(chalk30.gray(
-            `  Synced ${providerCount} BYOK provider key(s) from freesyntax.dev \u2192 ${(await import("./auth-UAMMP5IJ.js")).getByokSyncPath()}`
+            `  Synced ${providerCount} provider key(s) from freesyntax.dev \u2192 ${(await import("./auth-UAMMP5IJ.js")).getByokSyncPath()}`
           ));
         } else {
           console.log(chalk30.gray(
-            `  No BYOK keys on your profile yet \u2014 add them at freesyntax.dev/settings/keys then run ${chalk30.white("notch sync-keys")}.`
+            `  No provider keys on your profile yet \u2014 add them at freesyntax.dev/settings/keys then run ${chalk30.white("notch sync-keys")}.`
           ));
         }
       } catch (syncErr) {
         console.log(chalk30.yellow(
-          `  (BYOK sync skipped: ${syncErr.message.slice(0, 120)})`
+          `  (provider key sync skipped: ${syncErr.message.slice(0, 120)})`
         ));
       }
       console.log("");
@@ -8684,7 +9048,7 @@ async function main() {
       console.log(chalk30.gray("\n  Not signed in. Run: notch login\n"));
       return;
     }
-    const spinner = ora7("Pulling BYOK keys from freesyntax.dev...").start();
+    const spinner = ora7("Pulling provider keys from freesyntax.dev...").start();
     try {
       const { syncByokKeys, getByokSyncPath } = await import("./auth-UAMMP5IJ.js");
       const synced = await syncByokKeys(creds.token);
@@ -8692,7 +9056,7 @@ async function main() {
       const providers = Object.keys(synced.keys);
       if (providers.length === 0) {
         console.log(chalk30.gray(`
-  No BYOK keys on your profile yet.`));
+  No provider keys on your profile yet.`));
         console.log(chalk30.gray(`  Add them at ${chalk30.white("https://freesyntax.dev/settings/keys")} then rerun.
 `));
       } else {
@@ -8737,7 +9101,7 @@ async function main() {
     return;
   }
   if (promptArgs[0] === "mcp-serve" || promptArgs[0] === "mcp-server") {
-    const { runMcpServer } = await import("./server-IGOZHW52.js");
+    const { runMcpServer } = await import("./server-GMF4WV67.js");
     await runMcpServer({
       cwd: opts.cwd ?? process.cwd(),
       version: VERSION,
@@ -8776,7 +9140,7 @@ async function main() {
     const providerId = opts.provider === "custom" ? "__custom__" : opts.provider;
     const byokInfo = findByokProvider(providerId);
     if (!byokInfo) {
-      console.error(chalk30.red(`  Unknown BYOK provider: ${opts.provider}`));
+      console.error(chalk30.red(`  Unknown provider: ${opts.provider}`));
       console.error(chalk30.gray(`  Run  notch --list-providers  to see built-ins, or pass --provider custom --base-url <url>.`));
       process.exit(1);
     }
@@ -8793,7 +9157,7 @@ async function main() {
     } else if (isByokRef(opts.model)) {
       const { provider } = parseByokRef(opts.model);
       if (!findByokProvider(provider)) {
-        console.error(chalk30.red(`  Unknown BYOK provider in ref: ${provider}`));
+        console.error(chalk30.red(`  Unknown provider in model ref: ${provider}`));
         console.error(chalk30.gray(`  Run  notch --list-providers  to see built-ins.`));
         process.exit(1);
       }
@@ -8801,8 +9165,8 @@ async function main() {
       config.models.chat.byokProvider = void 0;
     } else {
       console.error(chalk30.red(`  Unknown model: ${opts.model}`));
-      console.error(chalk30.gray(`  Notch models: ${modelChoices}`));
-      console.error(chalk30.gray(`  For BYOK use "<provider>:<model>" (e.g. openrouter:anthropic/claude-sonnet-4-6), or run  notch --list-providers .`));
+      console.error(chalk30.gray("  Use provider/model, for example openrouter/anthropic/claude-sonnet-4-6."));
+      console.error(chalk30.gray("  Run  notch --list-providers  to see built-ins."));
       process.exit(1);
     }
   }
@@ -8856,7 +9220,7 @@ async function main() {
     if (err instanceof ByokMissingApiKeyError) {
       printWordmark(VERSION);
       const p = err.provider;
-      console.log(`  To use ${p.label} (BYOK) you need an API key.`);
+      console.log(`  To use ${p.label} you need an API key.`);
       console.log("");
       if (p.apiKeyEnv) {
         console.log("  \x1B[1mOption 1:\x1B[0m Set the env var");
@@ -8883,9 +9247,9 @@ async function main() {
       console.log("  \x1B[1mOption 3:\x1B[0m Pass it inline");
       console.log("    \x1B[33m$ notch --api-key your-key-here\x1B[0m");
       console.log("");
-      console.log("  \x1B[1mOr:\x1B[0m Bring your own key (OpenAI, Anthropic, OpenRouter, ...)");
+      console.log("  \x1B[1mOr:\x1B[0m Use any provider key (OpenAI, Anthropic, OpenRouter, ...)");
       console.log("    \x1B[33m$ notch --list-providers\x1B[0m");
-      console.log("    \x1B[33m$ notch --provider openrouter --model anthropic/claude-opus-4-7\x1B[0m");
+      console.log("    \x1B[33m$ notch --model openrouter/anthropic/claude-sonnet-4-6\x1B[0m");
       console.log("");
       console.log("  Get your Notch key at: \x1B[4mhttps://freesyntax.dev/settings\x1B[0m");
       console.log("");
@@ -8896,7 +9260,7 @@ async function main() {
   const info = activeByok ? {
     id: activeModelId,
     label: `${activeByok.label}`,
-    size: "BYOK",
+    size: "provider",
     gpu: activeByok.id,
     contextWindow: 128e3,
     maxOutputTokens: 8192,
@@ -8976,7 +9340,6 @@ async function main() {
       spinner.warn("Could not build repo map");
     }
   }
-  const baseSystemPrompt = await buildSystemPrompt(config.projectRoot, activeModelId);
   let outputSchema = null;
   if (opts.outputSchema) {
     try {
@@ -8991,19 +9354,6 @@ async function main() {
     }
   }
   const coordinatorMode = !!opts.coordinator || isCoordinatorModeEnv();
-  const systemPrompt = coordinatorMode ? [
-    COORDINATOR_SYSTEM_PROMPT,
-    repoMapStr ? `
-## Repository Map
-${repoMapStr}` : "",
-    outputSchema ? schemaInstructions(outputSchema) : ""
-  ].join("") : [
-    baseSystemPrompt,
-    repoMapStr ? `
-## Repository Map
-${repoMapStr}` : "",
-    outputSchema ? schemaInstructions(outputSchema) : ""
-  ].join("");
   if (coordinatorMode && !jsonMode) {
     console.log(
       chalk30.green(
@@ -9029,19 +9379,11 @@ ${repoMapStr}` : "",
   const branches = /* @__PURE__ */ new Map();
   let currentBranch = "main";
   const costTracker = new CostTracker();
-  const mcpClients = [];
   try {
     const configRaw = await fs20.readFile(nodePath2.resolve(config.projectRoot, ".notch.json"), "utf-8").catch(() => "{}");
-    const mcpConfigs = parseMCPConfig(JSON.parse(configRaw));
-    for (const [name, mcpConfig] of Object.entries(mcpConfigs)) {
-      try {
-        const client = new MCPClient(mcpConfig, name);
-        await client.connect();
-        mcpClients.push(client);
-        console.log(chalk30.green(`  MCP: Connected to ${name} (${client.tools.length} tools)`));
-      } catch (err) {
-        console.log(chalk30.yellow(`  MCP: Could not connect to ${name}: ${err.message}`));
-      }
+    const toolCount = await initMCPServers(JSON.parse(configRaw));
+    if (toolCount > 0) {
+      console.log(chalk30.green(`  MCP: Connected (${toolCount} tools)`));
     }
   } catch {
   }
@@ -9059,13 +9401,16 @@ ${repoMapStr}` : "",
   if (opts.guardian) {
     try {
       guardianModel = resolveModel({
-        model: "notch-solace-lite",
+        model: process.env.NOTCH_GUARDIAN_MODEL ?? config.models.chat.model,
         apiKey: config.models.chat.apiKey,
         baseUrl: process.env.NOTCH_GUARDIAN_BASE_URL,
-        headers: config.models.chat.headers
+        headers: config.models.chat.headers,
+        byokProvider: config.models.chat.byokProvider,
+        byokHeaders: config.models.chat.byokHeaders,
+        byokApiShape: config.models.chat.byokApiShape
       });
       if (!jsonMode) {
-        console.log(chalk30.green("  Guardian: enabled (notch-solace-lite)"));
+        console.log(chalk30.green(`  Guardian: enabled (${process.env.NOTCH_GUARDIAN_MODEL ?? config.models.chat.model})`));
       }
     } catch (err) {
       if (!jsonMode) {
@@ -9110,6 +9455,28 @@ ${repoMapStr}` : "",
       }
     }
   };
+  const baseSystemPrompt = await buildSystemPrompt(config.projectRoot, activeModelId, toolCtx);
+  const systemPrompt = coordinatorMode ? [
+    COORDINATOR_SYSTEM_PROMPT,
+    "\n\n",
+    baseSystemPrompt,
+    repoMapStr ? `
+
+## Repository Map
+${repoMapStr}` : "",
+    outputSchema ? `
+
+${schemaInstructions(outputSchema)}` : ""
+  ].join("") : [
+    baseSystemPrompt,
+    repoMapStr ? `
+
+## Repository Map
+${repoMapStr}` : "",
+    outputSchema ? `
+
+${schemaInstructions(outputSchema)}` : ""
+  ].join("");
   await toolCtx.runHook?.("session-start", {});
   const stopFileWatcher = startFileWatcher(config.projectRoot, hookConfig, (event, results) => {
     for (const r of results) {
@@ -9181,9 +9548,10 @@ Analyze the above input.`;
       attachments.push(res);
     }
     if (attachments.length > 0 && !modelSupportsImages(activeModelId)) {
-      const msg = `Selected model "${activeModelId}" may not support images. Switch with --model or --provider.`;
-      if (jsonMode) events.emit({ type: "warning", message: msg });
-      else console.warn(chalk30.yellow(`  ${msg}`));
+      const msg = `Selected model "${activeModelId}" does not support image attachments. Switch to a vision-capable model.`;
+      if (jsonMode) events.emit({ type: "error", message: msg });
+      else console.error(chalk30.red(`  ${msg}`));
+      process.exit(1);
     }
     if (attachments.length > 0) {
       const imageBlocks = attachments.map(imageToContentBlock);
@@ -9391,11 +9759,9 @@ Analyze the above input.`;
         } catch {
         }
       }
-      for (const client of mcpClients) {
-        try {
-          await client.disconnect();
-        } catch {
-        }
+      try {
+        disconnectMCPServers();
+      } catch {
       }
       slashMenu.cleanup();
       stopActiveLoop();
@@ -9422,35 +9788,6 @@ Analyze the above input.`;
       const picked = await interactiveModelPicker(activeModelId);
       if (!picked) {
         console.log(chalk30.gray("  Cancelled\n"));
-      } else if (picked.kind === "notch") {
-        if (picked.id === activeModelId) {
-          console.log(chalk30.gray(`  Already using ${MODEL_CATALOG[picked.id].label}
-`));
-        } else {
-          activeModelId = picked.id;
-          config.models.chat.model = activeModelId;
-          config.models.chat.byokProvider = void 0;
-          try {
-            model = resolveModel(config.models.chat);
-            const switchedInfo = MODEL_CATALOG[picked.id];
-            console.log(chalk30.green(`  \u2713 Switched to ${switchedInfo.label} (${switchedInfo.id})`));
-            const shortName = picked.id.replace("notch-", "");
-            const hw = switchedInfo.hardware;
-            console.log(chalk30.gray(
-              `    Hosted inference is live. To run locally you need ~${hw.vramGb}GB VRAM (${switchedInfo.hardware.recommendedGpu}) + ${hw.diskGb}GB disk for Q4 weights.`
-            ));
-            if (switchedInfo.hfRepo) {
-              console.log(chalk30.gray(`    Run ${chalk30.white(`/model download ${shortName}`)} to pull them in the background.
-`));
-            } else {
-              console.log(chalk30.gray(`    Local weights aren't published yet \u2014 sticking with the hosted API.
-`));
-            }
-          } catch (e) {
-            console.log(chalk30.red(`  Failed to switch: ${e.message}
-`));
-          }
-        }
       } else {
         activeModelId = picked.modelRef;
         config.models.chat.model = picked.modelRef;
@@ -9474,12 +9811,12 @@ Analyze the above input.`;
       rl.prompt();
       return;
     }
-    if (input.startsWith("/model ")) {
+    if (input.startsWith("/model ") && !input.startsWith("/model download")) {
       const arg = input.replace("/model ", "").trim();
       if (isByokRef(arg)) {
         const { provider } = parseByokRef(arg);
         if (!findByokProvider(provider)) {
-          console.log(chalk30.red(`  Unknown BYOK provider: ${provider}`));
+          console.log(chalk30.red(`  Unknown provider: ${provider}`));
           console.log(chalk30.gray(`  Run /providers to list built-ins.`));
           rl.prompt();
           return;
@@ -9498,6 +9835,29 @@ Analyze the above input.`;
 `));
           } else {
             console.log(chalk30.red(`  Failed to switch: ${e.message}
+`));
+          }
+        }
+        rl.prompt();
+        return;
+      }
+      const currentProvider = activeByokProvider(config.models.chat);
+      if (currentProvider) {
+        const providerId = currentProvider.id === "__custom__" ? "custom" : currentProvider.id;
+        const modelRef = `${providerId}/${arg}`;
+        activeModelId = modelRef;
+        config.models.chat.model = modelRef;
+        config.models.chat.byokProvider = void 0;
+        try {
+          model = resolveModel(config.models.chat);
+          console.log(chalk30.green(`  Switched to ${currentProvider.label} (${modelRef})
+`));
+        } catch (e) {
+          if (e instanceof ByokMissingApiKeyError) {
+            console.log(chalk30.yellow(`  \u26A0  ${e.message}
+`));
+          } else {
+            console.log(chalk30.red(`  Failed to switch: ${e.message}
 `));
           }
         }
@@ -9510,8 +9870,8 @@ Analyze the above input.`;
       }
       if (!isValidModel(newModel)) {
         console.log(chalk30.red(`  Unknown model: ${arg}`));
-        console.log(chalk30.gray(`  Notch models: ${modelChoices}`));
-        console.log(chalk30.gray(`  For BYOK use "<provider>:<model>" (e.g. openrouter:anthropic/claude-sonnet-4-6). Run /providers to list built-ins.`));
+        console.log(chalk30.gray("  Use provider/model, for example openrouter/anthropic/claude-sonnet-4-6."));
+        console.log(chalk30.gray("  Run /providers to list built-ins."));
         rl.prompt();
         return;
       }
@@ -9526,52 +9886,16 @@ Analyze the above input.`;
       return;
     }
     if (input.startsWith("/model download")) {
-      const arg = input.replace("/model download", "").trim();
-      if (!arg) {
-        console.log(chalk30.red("  Usage: /model download <pyre|ignis|solace|solace-lite>"));
-        rl.prompt();
-        return;
-      }
-      const normalised = arg.startsWith("notch-") ? arg : `notch-${arg}`;
-      if (!isValidModel(normalised)) {
-        console.log(chalk30.red(`  Unknown model: ${arg}`));
-        console.log(chalk30.gray(`  Notch models: ${modelChoices}`));
-        rl.prompt();
-        return;
-      }
-      const info2 = MODEL_CATALOG[normalised];
-      const { probeSystemCapabilities, evaluateHardware, renderVerdict, startModelDownload } = await import("./model-download-3NDKS3VM.js");
-      const caps = await probeSystemCapabilities();
-      const verdict = evaluateHardware(info2, caps);
-      console.log(`
-${renderVerdict(info2, verdict)}
-`);
-      if (!info2.hfRepo) {
-        console.log(chalk30.yellow(
-          `  ${info2.label} isn't published to Hugging Face yet \u2014 the Notch team is still preparing merged weights for public download. Use the hosted API in the meantime (run ${chalk30.white("notch login")} if you haven't already).
-`
-        ));
-        rl.prompt();
-        return;
-      }
-      try {
-        const handle = await startModelDownload(normalised);
-        console.log(chalk30.green(`  \u2193 Downloading ${info2.label} in background (pid ${handle.pid}).`));
-        console.log(chalk30.gray(`    Cache: ${handle.cacheDir}`));
-        console.log(chalk30.gray(`    Poll status with ${chalk30.white("/downloads")} \u2014 the pull keeps running even if you exit the REPL (${info2.hardware.diskGb} GB transfer).
-`));
-      } catch (err) {
-        console.log(chalk30.red(`  Download failed to start: ${err.message}
-`));
-      }
+      console.log(chalk30.gray("  Direct weight downloads are no longer part of the model picker."));
+      console.log(chalk30.gray("  Run a local provider instead, then select it with /model ollama/<model>, /model lmstudio/<model>, or /model custom/<model>.\n"));
       rl.prompt();
       return;
     }
     if (input === "/downloads") {
-      const { listDownloads } = await import("./model-download-3NDKS3VM.js");
+      const { listDownloads } = await import("./model-download-KCQJCEPW.js");
       const active = listDownloads();
       if (active.length === 0) {
-        console.log(chalk30.gray("  No downloads started this session. Try: /model download pyre\n"));
+        console.log(chalk30.gray("  No local downloads started this session.\n"));
       } else {
         console.log(chalk30.gray("\n  Model              State       Last line"));
         for (const h of active) {
@@ -9594,13 +9918,13 @@ ${renderVerdict(info2, verdict)}
         rl.prompt();
         return;
       }
-      const spinner2 = ora7("Pulling BYOK keys from freesyntax.dev...").start();
+      const spinner2 = ora7("Pulling provider keys from freesyntax.dev...").start();
       try {
         const synced = await syncByokKeys(creds.token);
         spinner2.stop();
         const providers = Object.keys(synced.keys);
         if (providers.length === 0) {
-          console.log(chalk30.gray(`  No BYOK keys on your profile yet. Add them at https://freesyntax.dev/settings/keys.
+          console.log(chalk30.gray(`  No provider keys on your profile yet. Add them at https://freesyntax.dev/settings/keys.
 `));
         } else {
           console.log(chalk30.green(`  \u2713 Synced ${providers.length} key(s): ${providers.join(", ")}`));
@@ -10204,9 +10528,10 @@ ${renderVerdict(info2, verdict)}
     if (replAttachments.length > 0 && !modelSupportsImages(activeModelId)) {
       console.warn(
         chalk30.yellow(
-          `  \u26A0 ${activeModelId} may not support images. Switch with /model.`
+          `  \u26A0 ${activeModelId} does not support image attachments. Switch with /model; skipping image(s).`
         )
       );
+      replAttachments.length = 0;
     }
     const textForRefs = imgRefs.cleanedText || (replAttachments.length > 0 ? "Describe the attached image(s)." : "");
     const { cleanInput, references } = await resolveReferences(textForRefs, config.projectRoot);
@@ -10215,6 +10540,10 @@ ${renderVerdict(info2, verdict)}
     if (references.length > 0) {
       console.log(chalk30.gray(`  Injected ${references.length} reference(s)`));
     }
+    if (!finalPrompt.trim() && replAttachments.length === 0) {
+      console.warn(chalk30.yellow("  Add a message, or switch to a vision-capable model before sending only images."));
+      return;
+    }
     if (replAttachments.length > 0) {
       const imageBlocks = replAttachments.map(imageToContentBlock);
       messages.push({
@@ -10390,7 +10719,6 @@ async function handleRalphSubcommand(args, cliOpts) {
   const config = await loadConfig(cliOpts.cwd ? { projectRoot: cliOpts.cwd } : {});
   if (cliOpts.model) config.models.chat.model = cliOpts.model;
   const model = resolveModel(config.models.chat);
-  const systemPrompt = await buildSystemPrompt(config.projectRoot, config.models.chat.model);
   const toolCtx = {
     cwd: config.projectRoot,
     requireConfirm: false,