codemaxxing 0.4.0 → 0.4.1

package/dist/config.d.ts

@@ -17,6 +17,7 @@ export interface CodemaxxingConfig {
         contextCompressionThreshold?: number;
         architectModel?: string;
         autoLint?: boolean;
+        stopOllamaOnExit?: boolean;
     };
 }
 export interface CLIArgs {
@@ -30,6 +31,8 @@ export interface CLIArgs {
  */
 export declare function parseCLIArgs(): CLIArgs;
 export declare function loadConfig(): CodemaxxingConfig;
+/** Save config to disk (merges with existing) */
+export declare function saveConfig(updates: Partial<CodemaxxingConfig>): void;
 /**
  * Apply CLI overrides to a provider config
  */

package/dist/config.js

@@ -102,6 +102,19 @@ export function loadConfig() {
         return DEFAULT_CONFIG;
     }
 }
+/** Save config to disk (merges with existing) */
+export function saveConfig(updates) {
+    const current = loadConfig();
+    const merged = {
+        ...current,
+        ...updates,
+        defaults: { ...current.defaults, ...(updates.defaults ?? {}) },
+    };
+    if (!existsSync(CONFIG_DIR)) {
+        mkdirSync(CONFIG_DIR, { recursive: true });
+    }
+    writeFileSync(CONFIG_FILE, JSON.stringify(merged, null, 2));
+}
 /**
  * Apply CLI overrides to a provider config
  */

package/dist/index.js

@@ -5,7 +5,7 @@ import { render, Box, Text, useInput, useApp, useStdout } from "ink";
 import { EventEmitter } from "events";
 import TextInput from "ink-text-input";
 import { CodingAgent } from "./agent.js";
-import { loadConfig, detectLocalProvider, parseCLIArgs, applyOverrides, listModels } from "./config.js";
+import { loadConfig, saveConfig, detectLocalProvider, parseCLIArgs, applyOverrides, listModels } from "./config.js";
 import { listSessions, getSession, loadMessages, deleteSession } from "./utils/sessions.js";
 import { execSync } from "child_process";
 import { isGitRepo, getBranch, getStatus, getDiff, undoLastCommit } from "./utils/git.js";
@@ -14,8 +14,8 @@ import { PROVIDERS, getCredentials, openRouterOAuth, anthropicSetupToken, import
 import { listInstalledSkills, installSkill, removeSkill, getRegistrySkills, searchRegistry, createSkillScaffold, getActiveSkills, getActiveSkillCount } from "./utils/skills.js";
 import { listServers, addServer, removeServer, getConnectedServers } from "./utils/mcp.js";
 import { detectHardware, formatBytes } from "./utils/hardware.js";
-import { getRecommendations, getFitIcon } from "./utils/models.js";
-import { isOllamaInstalled, isOllamaRunning, getOllamaInstallCommand, startOllama, pullModel } from "./utils/ollama.js";
+import { getRecommendationsWithLlmfit, getFitIcon, isLlmfitAvailable } from "./utils/models.js";
+import { isOllamaInstalled, isOllamaRunning, getOllamaInstallCommand, startOllama, stopOllama, pullModel, listInstalledModelsDetailed, deleteModel, getGPUMemoryUsage } from "./utils/ollama.js";
 const VERSION = "0.1.9";
 // ── Helpers ──
 function formatTimeAgo(date) {
@@ -67,6 +67,12 @@ const SLASH_COMMANDS = [
     { cmd: "/mcp add", desc: "add MCP server" },
     { cmd: "/mcp remove", desc: "remove MCP server" },
     { cmd: "/mcp reconnect", desc: "reconnect MCP servers" },
+    { cmd: "/ollama", desc: "Ollama status & models" },
+    { cmd: "/ollama list", desc: "list installed models" },
+    { cmd: "/ollama start", desc: "start Ollama server" },
+    { cmd: "/ollama stop", desc: "stop Ollama server" },
+    { cmd: "/ollama pull", desc: "download a model" },
+    { cmd: "/ollama delete", desc: "delete a model" },
     { cmd: "/quit", desc: "exit" },
 ];
 const SPINNER_FRAMES = ["⣾", "⣽", "⣻", "⢿", "⡿", "⣟", "⣯", "⣷"];
@@ -140,6 +146,10 @@ function App() {
     const [skillsPickerIndex, setSkillsPickerIndex] = useState(0);
     const [sessionDisabledSkills, setSessionDisabledSkills] = useState(new Set());
     const [approval, setApproval] = useState(null);
+    // ── Ollama Management State ──
+    const [ollamaDeleteConfirm, setOllamaDeleteConfirm] = useState(null);
+    const [ollamaPulling, setOllamaPulling] = useState(null);
+    const [ollamaExitPrompt, setOllamaExitPrompt] = useState(false);
     const [wizardScreen, setWizardScreen] = useState(null);
     const [wizardIndex, setWizardIndex] = useState(0);
     const [wizardHardware, setWizardHardware] = useState(null);
@@ -329,7 +339,8 @@ function App() {
                 // Commands that need args (like /commit, /model) — fill input instead of executing
                 if (selected.cmd === "/commit" || selected.cmd === "/model" || selected.cmd === "/session delete" ||
                     selected.cmd === "/skills install" || selected.cmd === "/skills remove" || selected.cmd === "/skills search" ||
-                    selected.cmd === "/skills on" || selected.cmd === "/skills off" || selected.cmd === "/architect") {
+                    selected.cmd === "/skills on" || selected.cmd === "/skills off" || selected.cmd === "/architect" ||
+                    selected.cmd === "/ollama pull" || selected.cmd === "/ollama delete") {
                     setInput(selected.cmd + " ");
                     setCmdIndex(0);
                     setInputKey((k) => k + 1);
@@ -355,7 +366,22 @@ function App() {
             return;
         addMsg("user", trimmed);
         if (trimmed === "/quit" || trimmed === "/exit") {
-            exit();
+            // Check if Ollama is running and offer to stop it
+            const ollamaUp = await isOllamaRunning();
+            if (ollamaUp) {
+                const config = loadConfig();
+                if (config.defaults.stopOllamaOnExit) {
+                    addMsg("info", "Stopping Ollama...");
+                    await stopOllama();
+                    exit();
+                }
+                else {
+                    setOllamaExitPrompt(true);
+                }
+            }
+            else {
+                exit();
+            }
             return;
         }
         if (trimmed === "/login" || trimmed === "/auth") {
@@ -398,6 +424,12 @@ function App() {
                 "  /mcp add   — add MCP server to global config",
                 "  /mcp remove — remove MCP server",
                 "  /mcp reconnect — reconnect all MCP servers",
+                "  /ollama    — Ollama status, models & GPU usage",
+                "  /ollama list — list installed models with sizes",
+                "  /ollama start — start Ollama server",
+                "  /ollama stop — stop Ollama server (frees GPU RAM)",
+                "  /ollama pull <model> — download a model",
+                "  /ollama delete <model> — delete a model from disk",
                 "  /quit      — exit",
             ].join("\n"));
             return;
@@ -565,6 +597,142 @@ function App() {
             addMsg("info", "🔍 Auto-lint OFF");
             return;
         }
+        // ── Ollama commands (work without agent) ──
+        if (trimmed === "/ollama" || trimmed === "/ollama status") {
+            const running = await isOllamaRunning();
+            const lines = [`Ollama: ${running ? "running" : "stopped"}`];
+            if (running) {
+                const models = await listInstalledModelsDetailed();
+                if (models.length > 0) {
+                    lines.push(`Installed models (${models.length}):`);
+                    for (const m of models) {
+                        const sizeGB = (m.size / (1024 * 1024 * 1024)).toFixed(1);
+                        lines.push(`  ${m.name}  (${sizeGB} GB)`);
+                    }
+                }
+                else {
+                    lines.push("No models installed.");
+                }
+                const gpuMem = getGPUMemoryUsage();
+                if (gpuMem)
+                    lines.push(`GPU: ${gpuMem}`);
+            }
+            else {
+                lines.push("Start with: /ollama start");
+            }
+            addMsg("info", lines.join("\n"));
+            return;
+        }
+        if (trimmed === "/ollama list") {
+            const running = await isOllamaRunning();
+            if (!running) {
+                addMsg("info", "Ollama is not running. Start with /ollama start");
+                return;
+            }
+            const models = await listInstalledModelsDetailed();
+            if (models.length === 0) {
+                addMsg("info", "No models installed. Pull one with /ollama pull <model>");
+            }
+            else {
+                const lines = models.map((m) => {
+                    const sizeGB = (m.size / (1024 * 1024 * 1024)).toFixed(1);
+                    return `  ${m.name}  (${sizeGB} GB)`;
+                });
+                addMsg("info", `Installed models:\n${lines.join("\n")}`);
+            }
+            return;
+        }
+        if (trimmed === "/ollama start") {
+            const running = await isOllamaRunning();
+            if (running) {
+                addMsg("info", "Ollama is already running.");
+                return;
+            }
+            if (!isOllamaInstalled()) {
+                addMsg("error", `Ollama is not installed. Install with: ${getOllamaInstallCommand(process.platform === "darwin" ? "macos" : process.platform === "win32" ? "windows" : "linux")}`);
+                return;
+            }
+            startOllama();
+            addMsg("info", "Starting Ollama server...");
+            // Wait for it to come up
+            for (let i = 0; i < 10; i++) {
+                await new Promise(r => setTimeout(r, 1000));
+                if (await isOllamaRunning()) {
+                    addMsg("info", "Ollama is running.");
+                    return;
+                }
+            }
+            addMsg("error", "Ollama did not start in time. Try running 'ollama serve' manually.");
+            return;
+        }
+        if (trimmed === "/ollama stop") {
+            const running = await isOllamaRunning();
+            if (!running) {
+                addMsg("info", "Ollama is not running.");
+                return;
+            }
+            addMsg("info", "Stopping Ollama...");
+            const result = await stopOllama();
+            addMsg(result.ok ? "info" : "error", result.ok ? `\u2705 ${result.message}` : `\u274C ${result.message}`);
+            return;
+        }
+        if (trimmed.startsWith("/ollama pull ")) {
+            const modelId = trimmed.replace("/ollama pull ", "").trim();
+            if (!modelId) {
+                addMsg("info", "Usage: /ollama pull <model>\n  Example: /ollama pull qwen2.5-coder:7b");
+                return;
+            }
+            if (!isOllamaInstalled()) {
+                addMsg("error", "Ollama is not installed.");
+                return;
+            }
+            // Ensure ollama is running
+            let running = await isOllamaRunning();
+            if (!running) {
+                startOllama();
+                addMsg("info", "Starting Ollama server...");
+                for (let i = 0; i < 10; i++) {
+                    await new Promise(r => setTimeout(r, 1000));
+                    if (await isOllamaRunning()) {
+                        running = true;
+                        break;
+                    }
+                }
+                if (!running) {
+                    addMsg("error", "Could not start Ollama. Run 'ollama serve' manually.");
+                    return;
+                }
+            }
+            setOllamaPulling({ model: modelId, progress: { status: "starting", percent: 0 } });
+            try {
+                await pullModel(modelId, (p) => {
+                    setOllamaPulling({ model: modelId, progress: p });
+                });
+                setOllamaPulling(null);
+                addMsg("info", `\u2705 Downloaded ${modelId}`);
+            }
+            catch (err) {
+                setOllamaPulling(null);
+                addMsg("error", `Failed to pull ${modelId}: ${err.message}`);
+            }
+            return;
+        }
+        if (trimmed.startsWith("/ollama delete ")) {
+            const modelId = trimmed.replace("/ollama delete ", "").trim();
+            if (!modelId) {
+                addMsg("info", "Usage: /ollama delete <model>");
+                return;
+            }
+            // Look up size for confirmation
+            const models = await listInstalledModelsDetailed();
+            const found = models.find((m) => m.name === modelId || m.name.startsWith(modelId));
+            if (!found) {
+                addMsg("error", `Model "${modelId}" not found. Use /ollama list to see installed models.`);
+                return;
+            }
+            setOllamaDeleteConfirm({ model: found.name, size: found.size });
+            return;
+        }
         // ── MCP commands (partially work without agent) ──
         if (trimmed === "/mcp" || trimmed === "/mcp list") {
             const servers = listServers(process.cwd());
@@ -1125,6 +1293,47 @@ function App() {
             }
             return;
         }
+        // ── Ollama delete confirmation ──
+        if (ollamaDeleteConfirm) {
+            if (inputChar === "y" || inputChar === "Y") {
+                const model = ollamaDeleteConfirm.model;
+                setOllamaDeleteConfirm(null);
+                const result = deleteModel(model);
+                addMsg(result.ok ? "info" : "error", result.ok ? `\u2705 ${result.message}` : `\u274C ${result.message}`);
+                return;
+            }
+            if (inputChar === "n" || inputChar === "N" || key.escape) {
+                setOllamaDeleteConfirm(null);
+                addMsg("info", "Delete cancelled.");
+                return;
+            }
+            return;
+        }
+        // ── Ollama exit prompt ──
+        if (ollamaExitPrompt) {
+            if (inputChar === "y" || inputChar === "Y") {
+                setOllamaExitPrompt(false);
+                stopOllama().then(() => exit());
+                return;
+            }
+            if (inputChar === "n" || inputChar === "N") {
+                setOllamaExitPrompt(false);
+                exit();
+                return;
+            }
+            if (inputChar === "a" || inputChar === "A") {
+                setOllamaExitPrompt(false);
+                saveConfig({ defaults: { ...loadConfig().defaults, stopOllamaOnExit: true } });
+                addMsg("info", "Saved preference: always stop Ollama on exit.");
+                stopOllama().then(() => exit());
+                return;
+            }
+            if (key.escape) {
+                setOllamaExitPrompt(false);
+                return;
+            }
+            return;
+        }
         // ── Setup Wizard Navigation ──
         if (wizardScreen) {
             if (wizardScreen === "connection") {
@@ -1144,11 +1353,11 @@ function App() {
                 if (key.return) {
                     const selected = items[wizardIndex];
                     if (selected === "local") {
-                        // Scan hardware and show model picker
+                        // Scan hardware and show model picker (use llmfit if available)
                         const hw = detectHardware();
                         setWizardHardware(hw);
-                        const recs = getRecommendations(hw).filter(m => m.fit !== "skip");
-                        setWizardModels(recs);
+                        const { models: recs } = getRecommendationsWithLlmfit(hw);
+                        setWizardModels(recs.filter(m => m.fit !== "skip"));
                         setWizardScreen("models");
                         setWizardIndex(0);
                     }
@@ -1494,7 +1703,14 @@ function App() {
         }
         if (key.ctrl && inputChar === "c") {
             if (ctrlCPressed) {
-                exit();
+                // Force quit on second Ctrl+C — don't block
+                const config = loadConfig();
+                if (config.defaults.stopOllamaOnExit) {
+                    stopOllama().finally(() => exit());
+                }
+                else {
+                    exit();
+                }
             }
             else {
                 setCtrlCPressed(true);
@@ -1572,12 +1788,12 @@ function App() {
             })(), skillsPicker === "remove" && (() => {
                 const installed = listInstalledSkills();
                 return (_jsxs(Box, { flexDirection: "column", borderStyle: "single", borderColor: theme.colors.error, paddingX: 1, marginBottom: 0, children: [_jsx(Text, { bold: true, color: theme.colors.error, children: "Remove a skill:" }), installed.map((s, i) => (_jsxs(Text, { children: [i === skillsPickerIndex ? _jsx(Text, { color: theme.colors.suggestion, bold: true, children: "▸ " }) : _jsx(Text, { children: "  " }), _jsxs(Text, { color: i === skillsPickerIndex ? theme.colors.suggestion : theme.colors.muted, children: [s.name, " \u2014 ", s.description] })] }, s.name))), _jsx(Text, { dimColor: true, children: "  ↑↓ navigate · Enter remove · Esc back" })] }));
-            })(), themePicker && (_jsxs(Box, { flexDirection: "column", borderStyle: "single", borderColor: theme.colors.border, paddingX: 1, marginBottom: 0, children: [_jsx(Text, { bold: true, color: theme.colors.secondary, children: "Choose a theme:" }), listThemes().map((key, i) => (_jsxs(Text, { children: [i === themePickerIndex ? _jsx(Text, { color: theme.colors.suggestion, bold: true, children: "▸ " }) : _jsx(Text, { children: "  " }), _jsx(Text, { color: i === themePickerIndex ? theme.colors.suggestion : theme.colors.primary, bold: true, children: key }), _jsxs(Text, { color: theme.colors.muted, children: [" — ", THEMES[key].description] }), key === theme.name.toLowerCase() ? _jsx(Text, { color: theme.colors.muted, children: " (current)" }) : null] }, key))), _jsx(Text, { dimColor: true, children: "  ↑↓ navigate · Enter select · Esc cancel" })] })), sessionPicker && (_jsxs(Box, { flexDirection: "column", borderStyle: "single", borderColor: theme.colors.secondary, paddingX: 1, marginBottom: 0, children: [_jsx(Text, { bold: true, color: theme.colors.secondary, children: "Resume a session:" }), sessionPicker.map((s, i) => (_jsxs(Text, { children: [i === sessionPickerIndex ? _jsx(Text, { color: theme.colors.suggestion, bold: true, children: "▸ " }) : _jsx(Text, { children: "  " }), _jsx(Text, { color: i === sessionPickerIndex ? theme.colors.suggestion : theme.colors.muted, children: s.display })] }, s.id))), _jsx(Text, { dimColor: true, children: "  ↑↓ navigate · Enter select · Esc cancel" })] })), deleteSessionPicker && (_jsxs(Box, { flexDirection: "column", borderStyle: "single", borderColor: theme.colors.error, paddingX: 1, marginBottom: 0, children: [_jsx(Text, { bold: true, color: theme.colors.error, children: "Delete a session:" }), deleteSessionPicker.map((s, i) => (_jsxs(Text, { children: [i === deleteSessionPickerIndex ? _jsx(Text, { color: theme.colors.suggestion, bold: true, children: "▸ " }) : _jsx(Text, { children: "  " }), _jsx(Text, { color: i === deleteSessionPickerIndex ? theme.colors.suggestion : theme.colors.muted, children: s.display })] }, s.id))), _jsx(Text, { dimColor: true, children: "  ↑↓ navigate · Enter select · Esc cancel" })] })), deleteSessionConfirm && (_jsxs(Box, { flexDirection: "column", borderStyle: "single", borderColor: theme.colors.warning, paddingX: 1, marginBottom: 0, children: [_jsxs(Text, { bold: true, color: theme.colors.warning, children: ["Delete session ", deleteSessionConfirm.id, "?"] }), _jsxs(Text, { color: theme.colors.muted, children: ["  ", deleteSessionConfirm.display] }), _jsxs(Text, { children: [_jsx(Text, { color: theme.colors.error, bold: true, children: " [y]" }), _jsx(Text, { children: "es  " }), _jsx(Text, { color: theme.colors.success, bold: true, children: "[n]" }), _jsx(Text, { children: "o" })] })] })), wizardScreen === "connection" && (_jsxs(Box, { flexDirection: "column", borderStyle: "single", borderColor: theme.colors.border, paddingX: 1, marginBottom: 0, children: [_jsx(Text, { bold: true, color: theme.colors.secondary, children: "No LLM detected. How do you want to connect?" }), _jsx(Text, { children: "" }), [
+            })(), themePicker && (_jsxs(Box, { flexDirection: "column", borderStyle: "single", borderColor: theme.colors.border, paddingX: 1, marginBottom: 0, children: [_jsx(Text, { bold: true, color: theme.colors.secondary, children: "Choose a theme:" }), listThemes().map((key, i) => (_jsxs(Text, { children: [i === themePickerIndex ? _jsx(Text, { color: theme.colors.suggestion, bold: true, children: "▸ " }) : _jsx(Text, { children: "  " }), _jsx(Text, { color: i === themePickerIndex ? theme.colors.suggestion : theme.colors.primary, bold: true, children: key }), _jsxs(Text, { color: theme.colors.muted, children: [" — ", THEMES[key].description] }), key === theme.name.toLowerCase() ? _jsx(Text, { color: theme.colors.muted, children: " (current)" }) : null] }, key))), _jsx(Text, { dimColor: true, children: "  ↑↓ navigate · Enter select · Esc cancel" })] })), sessionPicker && (_jsxs(Box, { flexDirection: "column", borderStyle: "single", borderColor: theme.colors.secondary, paddingX: 1, marginBottom: 0, children: [_jsx(Text, { bold: true, color: theme.colors.secondary, children: "Resume a session:" }), sessionPicker.map((s, i) => (_jsxs(Text, { children: [i === sessionPickerIndex ? _jsx(Text, { color: theme.colors.suggestion, bold: true, children: "▸ " }) : _jsx(Text, { children: "  " }), _jsx(Text, { color: i === sessionPickerIndex ? theme.colors.suggestion : theme.colors.muted, children: s.display })] }, s.id))), _jsx(Text, { dimColor: true, children: "  ↑↓ navigate · Enter select · Esc cancel" })] })), deleteSessionPicker && (_jsxs(Box, { flexDirection: "column", borderStyle: "single", borderColor: theme.colors.error, paddingX: 1, marginBottom: 0, children: [_jsx(Text, { bold: true, color: theme.colors.error, children: "Delete a session:" }), deleteSessionPicker.map((s, i) => (_jsxs(Text, { children: [i === deleteSessionPickerIndex ? _jsx(Text, { color: theme.colors.suggestion, bold: true, children: "▸ " }) : _jsx(Text, { children: "  " }), _jsx(Text, { color: i === deleteSessionPickerIndex ? theme.colors.suggestion : theme.colors.muted, children: s.display })] }, s.id))), _jsx(Text, { dimColor: true, children: "  ↑↓ navigate · Enter select · Esc cancel" })] })), deleteSessionConfirm && (_jsxs(Box, { flexDirection: "column", borderStyle: "single", borderColor: theme.colors.warning, paddingX: 1, marginBottom: 0, children: [_jsxs(Text, { bold: true, color: theme.colors.warning, children: ["Delete session ", deleteSessionConfirm.id, "?"] }), _jsxs(Text, { color: theme.colors.muted, children: ["  ", deleteSessionConfirm.display] }), _jsxs(Text, { children: [_jsx(Text, { color: theme.colors.error, bold: true, children: " [y]" }), _jsx(Text, { children: "es  " }), _jsx(Text, { color: theme.colors.success, bold: true, children: "[n]" }), _jsx(Text, { children: "o" })] })] })), ollamaDeleteConfirm && (_jsxs(Box, { flexDirection: "column", borderStyle: "single", borderColor: theme.colors.warning, paddingX: 1, marginBottom: 0, children: [_jsxs(Text, { bold: true, color: theme.colors.warning, children: ["Delete ", ollamaDeleteConfirm.model, " (", (ollamaDeleteConfirm.size / (1024 * 1024 * 1024)).toFixed(1), " GB)?"] }), _jsxs(Text, { children: [_jsx(Text, { color: theme.colors.error, bold: true, children: " [y]" }), _jsx(Text, { children: "es  " }), _jsx(Text, { color: theme.colors.success, bold: true, children: "[n]" }), _jsx(Text, { children: "o" })] })] })), ollamaPulling && (_jsxs(Box, { flexDirection: "column", borderStyle: "single", borderColor: theme.colors.border, paddingX: 1, marginBottom: 0, children: [_jsxs(Text, { bold: true, color: theme.colors.secondary, children: ["  Downloading ", ollamaPulling.model, "..."] }), ollamaPulling.progress.status === "downloading" || ollamaPulling.progress.percent > 0 ? (_jsxs(Text, { children: ["  ", _jsxs(Text, { color: theme.colors.primary, children: ["\u2588".repeat(Math.floor(ollamaPulling.progress.percent / 5)), "\u2591".repeat(20 - Math.floor(ollamaPulling.progress.percent / 5))] }), "  ", _jsxs(Text, { bold: true, children: [ollamaPulling.progress.percent, "%"] }), ollamaPulling.progress.completed != null && ollamaPulling.progress.total != null ? (_jsxs(Text, { color: theme.colors.muted, children: [" \u00B7 ", formatBytes(ollamaPulling.progress.completed), " / ", formatBytes(ollamaPulling.progress.total)] })) : null] })) : (_jsxs(Text, { color: theme.colors.muted, children: ["  ", ollamaPulling.progress.status, "..."] }))] })), ollamaExitPrompt && (_jsxs(Box, { flexDirection: "column", borderStyle: "single", borderColor: theme.colors.warning, paddingX: 1, marginBottom: 0, children: [_jsx(Text, { bold: true, color: theme.colors.warning, children: "Ollama is still running." }), _jsx(Text, { color: theme.colors.muted, children: "Stop it to free GPU memory?" }), _jsxs(Text, { children: [_jsx(Text, { color: theme.colors.success, bold: true, children: " [y]" }), _jsx(Text, { children: "es  " }), _jsx(Text, { color: theme.colors.error, bold: true, children: "[n]" }), _jsx(Text, { children: "o  " }), _jsx(Text, { color: theme.colors.primary, bold: true, children: "[a]" }), _jsx(Text, { children: "lways" })] })] })), wizardScreen === "connection" && (_jsxs(Box, { flexDirection: "column", borderStyle: "single", borderColor: theme.colors.border, paddingX: 1, marginBottom: 0, children: [_jsx(Text, { bold: true, color: theme.colors.secondary, children: "No LLM detected. How do you want to connect?" }), _jsx(Text, { children: "" }), [
                         { key: "local", icon: "\uD83D\uDDA5\uFE0F", label: "Set up a local model", desc: "free, runs on your machine" },
                         { key: "openrouter", icon: "\uD83C\uDF10", label: "OpenRouter", desc: "200+ cloud models, browser login" },
                         { key: "apikey", icon: "\uD83D\uDD11", label: "Enter API key manually", desc: "" },
                         { key: "existing", icon: "\u2699\uFE0F", label: "I already have a server running", desc: "" },
-                    ].map((item, i) => (_jsxs(Text, { children: [i === wizardIndex ? _jsx(Text, { color: theme.colors.suggestion, bold: true, children: "  \u25B8 " }) : _jsx(Text, { children: "    " }), _jsxs(Text, { color: i === wizardIndex ? theme.colors.suggestion : theme.colors.primary, bold: true, children: [item.icon, "  ", item.label] }), item.desc ? _jsxs(Text, { color: theme.colors.muted, children: [" (", item.desc, ")"] }) : null] }, item.key))), _jsx(Text, { children: "" }), _jsx(Text, { dimColor: true, children: "  \u2191\u2193 navigate \u00B7 Enter to select" })] })), wizardScreen === "models" && wizardHardware && (_jsxs(Box, { flexDirection: "column", borderStyle: "single", borderColor: theme.colors.border, paddingX: 1, marginBottom: 0, children: [_jsx(Text, { bold: true, color: theme.colors.secondary, children: "Your hardware:" }), _jsxs(Text, { color: theme.colors.muted, children: ["  CPU: ", wizardHardware.cpu.name, " (", wizardHardware.cpu.cores, " cores)"] }), _jsxs(Text, { color: theme.colors.muted, children: ["  RAM: ", formatBytes(wizardHardware.ram)] }), wizardHardware.gpu ? (_jsxs(Text, { color: theme.colors.muted, children: ["  GPU: ", wizardHardware.gpu.name, wizardHardware.gpu.vram > 0 ? ` (${formatBytes(wizardHardware.gpu.vram)})` : ""] })) : (_jsx(Text, { color: theme.colors.muted, children: "  GPU: none detected" })), _jsx(Text, { children: "" }), _jsx(Text, { bold: true, color: theme.colors.secondary, children: "Recommended models:" }), _jsx(Text, { children: "" }), wizardModels.map((m, i) => (_jsxs(Text, { children: [i === wizardIndex ? _jsx(Text, { color: theme.colors.suggestion, bold: true, children: "  \u25B8 " }) : _jsx(Text, { children: "    " }), _jsxs(Text, { children: [getFitIcon(m.fit), " "] }), _jsx(Text, { color: i === wizardIndex ? theme.colors.suggestion : theme.colors.primary, bold: true, children: m.name }), _jsxs(Text, { color: theme.colors.muted, children: ["     ~", m.size, " GB \u00B7 ", m.quality === "best" ? "Best" : m.quality === "great" ? "Great" : "Good", " quality \u00B7 ", m.speed] })] }, m.ollamaId))), wizardModels.length === 0 && (_jsx(Text, { color: theme.colors.error, children: "  No suitable models found for your hardware." })), _jsx(Text, { children: "" }), _jsx(Text, { dimColor: true, children: "  \u2191\u2193 navigate \u00B7 Enter to install \u00B7 Esc back" })] })), wizardScreen === "install-ollama" && (_jsxs(Box, { flexDirection: "column", borderStyle: "single", borderColor: theme.colors.warning, paddingX: 1, marginBottom: 0, children: [_jsx(Text, { bold: true, color: theme.colors.warning, children: "Ollama is required for local models." }), _jsx(Text, { children: "" }), _jsxs(Text, { color: theme.colors.primary, children: ["  Install with: ", _jsx(Text, { bold: true, children: getOllamaInstallCommand(wizardHardware?.os ?? "linux") })] }), _jsx(Text, { children: "" }), _jsx(Text, { dimColor: true, children: "  Run the command above, then press Enter to continue..." }), _jsx(Text, { dimColor: true, children: "  Esc to go back" })] })), wizardScreen === "pulling" && wizardSelectedModel && (_jsx(Box, { flexDirection: "column", borderStyle: "single", borderColor: theme.colors.border, paddingX: 1, marginBottom: 0, children: wizardPullError ? (_jsxs(_Fragment, { children: [_jsxs(Text, { color: theme.colors.error, bold: true, children: ["  \u274C Error: ", wizardPullError] }), _jsx(Text, { children: "" }), _jsx(Text, { dimColor: true, children: "  Press Enter to retry \u00B7 Esc to go back" })] })) : wizardPullProgress ? (_jsxs(_Fragment, { children: [_jsxs(Text, { bold: true, color: theme.colors.secondary, children: ["  Downloading ", wizardSelectedModel.name, "..."] }), wizardPullProgress.status === "downloading" || wizardPullProgress.percent > 0 ? (_jsx(_Fragment, { children: _jsxs(Text, { children: ["  ", _jsxs(Text, { color: theme.colors.primary, children: ["\u2588".repeat(Math.floor(wizardPullProgress.percent / 5)), "\u2591".repeat(20 - Math.floor(wizardPullProgress.percent / 5))] }), "  ", _jsxs(Text, { bold: true, children: [wizardPullProgress.percent, "%"] }), wizardPullProgress.completed != null && wizardPullProgress.total != null ? (_jsxs(Text, { color: theme.colors.muted, children: [" \u00B7 ", formatBytes(wizardPullProgress.completed), " / ", formatBytes(wizardPullProgress.total)] })) : null] }) })) : (_jsxs(Text, { color: theme.colors.muted, children: ["  ", wizardPullProgress.status, "..."] }))] })) : null })), showSuggestions && (_jsxs(Box, { flexDirection: "column", borderStyle: "single", borderColor: theme.colors.muted, paddingX: 1, marginBottom: 0, children: [cmdMatches.slice(0, 6).map((c, i) => (_jsxs(Text, { children: [i === cmdIndex ? _jsx(Text, { color: theme.colors.suggestion, bold: true, children: "▸ " }) : _jsx(Text, { children: "  " }), _jsx(Text, { color: i === cmdIndex ? theme.colors.suggestion : theme.colors.primary, bold: true, children: c.cmd }), _jsxs(Text, { color: theme.colors.muted, children: [" — ", c.desc] })] }, i))), _jsx(Text, { dimColor: true, children: "  ↑↓ navigate · Tab select" })] })), _jsxs(Box, { borderStyle: "single", borderColor: approval ? theme.colors.warning : theme.colors.border, paddingX: 1, children: [_jsx(Text, { color: theme.colors.secondary, bold: true, children: "> " }), approval ? (_jsx(Text, { color: theme.colors.warning, children: "waiting for approval..." })) : ready && !loading ? (_jsxs(Box, { children: [pastedChunks.map((p) => (_jsxs(Text, { color: theme.colors.muted, children: ["[Pasted text #", p.id, " +", p.lines, " lines]"] }, p.id))), _jsx(TextInput, { value: input, onChange: (v) => { setInput(v); setCmdIndex(0); }, onSubmit: handleSubmit }, inputKey)] })) : (_jsx(Text, { dimColor: true, children: loading ? "waiting for response..." : "initializing..." }))] }), agent && (_jsx(Box, { paddingX: 2, children: _jsxs(Text, { dimColor: true, children: ["💬 ", agent.getContextLength(), " messages · ~", (() => {
+                    ].map((item, i) => (_jsxs(Text, { children: [i === wizardIndex ? _jsx(Text, { color: theme.colors.suggestion, bold: true, children: "  \u25B8 " }) : _jsx(Text, { children: "    " }), _jsxs(Text, { color: i === wizardIndex ? theme.colors.suggestion : theme.colors.primary, bold: true, children: [item.icon, "  ", item.label] }), item.desc ? _jsxs(Text, { color: theme.colors.muted, children: [" (", item.desc, ")"] }) : null] }, item.key))), _jsx(Text, { children: "" }), _jsx(Text, { dimColor: true, children: "  \u2191\u2193 navigate \u00B7 Enter to select" })] })), wizardScreen === "models" && wizardHardware && (_jsxs(Box, { flexDirection: "column", borderStyle: "single", borderColor: theme.colors.border, paddingX: 1, marginBottom: 0, children: [_jsx(Text, { bold: true, color: theme.colors.secondary, children: "Your hardware:" }), _jsxs(Text, { color: theme.colors.muted, children: ["  CPU: ", wizardHardware.cpu.name, " (", wizardHardware.cpu.cores, " cores)"] }), _jsxs(Text, { color: theme.colors.muted, children: ["  RAM: ", formatBytes(wizardHardware.ram)] }), wizardHardware.gpu ? (_jsxs(Text, { color: theme.colors.muted, children: ["  GPU: ", wizardHardware.gpu.name, wizardHardware.gpu.vram > 0 ? ` (${formatBytes(wizardHardware.gpu.vram)})` : ""] })) : (_jsx(Text, { color: theme.colors.muted, children: "  GPU: none detected" })), !isLlmfitAvailable() && (_jsx(Text, { dimColor: true, children: "  Tip: Install llmfit for smarter recommendations: brew install llmfit" })), _jsx(Text, { children: "" }), _jsx(Text, { bold: true, color: theme.colors.secondary, children: "Recommended models:" }), _jsx(Text, { children: "" }), wizardModels.map((m, i) => (_jsxs(Text, { children: [i === wizardIndex ? _jsx(Text, { color: theme.colors.suggestion, bold: true, children: "  \u25B8 " }) : _jsx(Text, { children: "    " }), _jsxs(Text, { children: [getFitIcon(m.fit), " "] }), _jsx(Text, { color: i === wizardIndex ? theme.colors.suggestion : theme.colors.primary, bold: true, children: m.name }), _jsxs(Text, { color: theme.colors.muted, children: ["     ~", m.size, " GB \u00B7 ", m.quality === "best" ? "Best" : m.quality === "great" ? "Great" : "Good", " quality \u00B7 ", m.speed] })] }, m.ollamaId))), wizardModels.length === 0 && (_jsx(Text, { color: theme.colors.error, children: "  No suitable models found for your hardware." })), _jsx(Text, { children: "" }), _jsx(Text, { dimColor: true, children: "  \u2191\u2193 navigate \u00B7 Enter to install \u00B7 Esc back" })] })), wizardScreen === "install-ollama" && (_jsxs(Box, { flexDirection: "column", borderStyle: "single", borderColor: theme.colors.warning, paddingX: 1, marginBottom: 0, children: [_jsx(Text, { bold: true, color: theme.colors.warning, children: "Ollama is required for local models." }), _jsx(Text, { children: "" }), _jsxs(Text, { color: theme.colors.primary, children: ["  Install with: ", _jsx(Text, { bold: true, children: getOllamaInstallCommand(wizardHardware?.os ?? "linux") })] }), _jsx(Text, { children: "" }), _jsx(Text, { dimColor: true, children: "  Run the command above, then press Enter to continue..." }), _jsx(Text, { dimColor: true, children: "  Esc to go back" })] })), wizardScreen === "pulling" && wizardSelectedModel && (_jsx(Box, { flexDirection: "column", borderStyle: "single", borderColor: theme.colors.border, paddingX: 1, marginBottom: 0, children: wizardPullError ? (_jsxs(_Fragment, { children: [_jsxs(Text, { color: theme.colors.error, bold: true, children: ["  \u274C Error: ", wizardPullError] }), _jsx(Text, { children: "" }), _jsx(Text, { dimColor: true, children: "  Press Enter to retry \u00B7 Esc to go back" })] })) : wizardPullProgress ? (_jsxs(_Fragment, { children: [_jsxs(Text, { bold: true, color: theme.colors.secondary, children: ["  Downloading ", wizardSelectedModel.name, "..."] }), wizardPullProgress.status === "downloading" || wizardPullProgress.percent > 0 ? (_jsx(_Fragment, { children: _jsxs(Text, { children: ["  ", _jsxs(Text, { color: theme.colors.primary, children: ["\u2588".repeat(Math.floor(wizardPullProgress.percent / 5)), "\u2591".repeat(20 - Math.floor(wizardPullProgress.percent / 5))] }), "  ", _jsxs(Text, { bold: true, children: [wizardPullProgress.percent, "%"] }), wizardPullProgress.completed != null && wizardPullProgress.total != null ? (_jsxs(Text, { color: theme.colors.muted, children: [" \u00B7 ", formatBytes(wizardPullProgress.completed), " / ", formatBytes(wizardPullProgress.total)] })) : null] }) })) : (_jsxs(Text, { color: theme.colors.muted, children: ["  ", wizardPullProgress.status, "..."] }))] })) : null })), showSuggestions && (_jsxs(Box, { flexDirection: "column", borderStyle: "single", borderColor: theme.colors.muted, paddingX: 1, marginBottom: 0, children: [cmdMatches.slice(0, 6).map((c, i) => (_jsxs(Text, { children: [i === cmdIndex ? _jsx(Text, { color: theme.colors.suggestion, bold: true, children: "▸ " }) : _jsx(Text, { children: "  " }), _jsx(Text, { color: i === cmdIndex ? theme.colors.suggestion : theme.colors.primary, bold: true, children: c.cmd }), _jsxs(Text, { color: theme.colors.muted, children: [" — ", c.desc] })] }, i))), _jsx(Text, { dimColor: true, children: "  ↑↓ navigate · Tab select" })] })), _jsxs(Box, { borderStyle: "single", borderColor: approval ? theme.colors.warning : theme.colors.border, paddingX: 1, children: [_jsx(Text, { color: theme.colors.secondary, bold: true, children: "> " }), approval ? (_jsx(Text, { color: theme.colors.warning, children: "waiting for approval..." })) : ready && !loading ? (_jsxs(Box, { children: [pastedChunks.map((p) => (_jsxs(Text, { color: theme.colors.muted, children: ["[Pasted text #", p.id, " +", p.lines, " lines]"] }, p.id))), _jsx(TextInput, { value: input, onChange: (v) => { setInput(v); setCmdIndex(0); }, onSubmit: handleSubmit }, inputKey)] })) : (_jsx(Text, { dimColor: true, children: loading ? "waiting for response..." : "initializing..." }))] }), agent && (_jsx(Box, { paddingX: 2, children: _jsxs(Text, { dimColor: true, children: ["💬 ", agent.getContextLength(), " messages · ~", (() => {
                             const tokens = agent.estimateTokens();
                             return tokens >= 1000 ? `${(tokens / 1000).toFixed(1)}k` : String(tokens);
                         })(), " tokens", (() => {

package/dist/utils/models.d.ts

@@ -15,3 +15,10 @@ export interface ScoredModel extends RecommendedModel {
 }
 export declare function getRecommendations(hardware: HardwareInfo): ScoredModel[];
 export declare function getFitIcon(fit: ModelFit): string;
+/** Check if llmfit binary is available */
+export declare function isLlmfitAvailable(): boolean;
+/** Get recommendations using llmfit if available, otherwise fall back to hardcoded list */
+export declare function getRecommendationsWithLlmfit(hardware: HardwareInfo): {
+    models: ScoredModel[];
+    usedLlmfit: boolean;
+};

package/dist/utils/models.js

@@ -1,3 +1,4 @@
+import { execSync } from "child_process";
 const MODELS = [
     {
         name: "Qwen 2.5 Coder 3B",
@@ -111,3 +112,66 @@ export function getFitIcon(fit) {
         case "skip": return "\u274C"; // ❌
     }
 }
+/** Check if llmfit binary is available */
+export function isLlmfitAvailable() {
+    try {
+        const cmd = process.platform === "win32" ? "where llmfit" : "which llmfit";
+        execSync(cmd, { stdio: ["pipe", "pipe", "pipe"], timeout: 3000 });
+        return true;
+    }
+    catch {
+        return false;
+    }
+}
+function mapLlmfitFit(fit) {
+    switch (fit) {
+        case "perfect": return "perfect";
+        case "good": return "good";
+        case "marginal": return "tight";
+        default: return "skip";
+    }
+}
+function mapLlmfitQuality(params_b) {
+    if (params_b >= 14)
+        return "best";
+    if (params_b >= 7)
+        return "great";
+    return "good";
+}
+/** Get recommendations using llmfit if available, otherwise fall back to hardcoded list */
+export function getRecommendationsWithLlmfit(hardware) {
+    if (!isLlmfitAvailable()) {
+        return { models: getRecommendations(hardware), usedLlmfit: false };
+    }
+    try {
+        const raw = execSync("llmfit recommend --use-case coding --format json --limit 10", {
+            encoding: "utf-8",
+            timeout: 15000,
+            stdio: ["pipe", "pipe", "pipe"],
+        });
+        const llmfitModels = JSON.parse(raw.trim());
+        if (!Array.isArray(llmfitModels) || llmfitModels.length === 0) {
+            return { models: getRecommendations(hardware), usedLlmfit: false };
+        }
+        const scored = llmfitModels
+            .filter((m) => m.provider === "ollama")
+            .map((m) => ({
+            name: m.name.split(":")[0]?.replace(/-/g, " ").replace(/\b\w/g, c => c.toUpperCase()) + ` ${m.params_b}B`,
+            ollamaId: m.name,
+            size: Math.ceil(m.vram_gb),
+            ramRequired: Math.ceil(m.ram_gb),
+            vramOptimal: Math.ceil(m.vram_gb),
+            description: `${m.quant} · ~${m.estimated_tps.toFixed(0)} tok/s`,
+            speed: `~${m.estimated_tps.toFixed(0)} tok/s`,
+            quality: mapLlmfitQuality(m.params_b),
+            fit: mapLlmfitFit(m.fit),
+        }));
+        if (scored.length === 0) {
+            return { models: getRecommendations(hardware), usedLlmfit: false };
+        }
+        return { models: scored, usedLlmfit: true };
+    }
+    catch {
+        return { models: getRecommendations(hardware), usedLlmfit: false };
+    }
+}

package/dist/utils/ollama.d.ts

@@ -18,5 +18,25 @@ export interface PullProgress {
  * Returns a promise that resolves when complete.
  */
 export declare function pullModel(modelId: string, onProgress?: (progress: PullProgress) => void): Promise<void>;
+export interface OllamaModelInfo {
+    name: string;
+    size: number;
+    modified_at: string;
+    digest: string;
+}
+/** List models installed in Ollama with detailed info */
+export declare function listInstalledModelsDetailed(): Promise<OllamaModelInfo[]>;
 /** List models installed in Ollama */
 export declare function listInstalledModels(): Promise<string[]>;
+/** Stop all loaded models (frees VRAM) and kill the Ollama server process */
+export declare function stopOllama(): Promise<{
+    ok: boolean;
+    message: string;
+}>;
+/** Delete a model from disk */
+export declare function deleteModel(modelId: string): {
+    ok: boolean;
+    message: string;
+};
+/** Get GPU memory usage info (best-effort) */
+export declare function getGPUMemoryUsage(): string | null;

package/dist/utils/ollama.js

@@ -104,8 +104,8 @@ export function pullModel(modelId, onProgress) {
         });
     });
 }
-/** List models installed in Ollama */
-export async function listInstalledModels() {
+/** List models installed in Ollama with detailed info */
+export async function listInstalledModelsDetailed() {
     try {
         const controller = new AbortController();
         const timeout = setTimeout(() => controller.abort(), 3000);
@@ -113,9 +113,101 @@ export async function listInstalledModels() {
         clearTimeout(timeout);
         if (res.ok) {
             const data = (await res.json());
-            return (data.models ?? []).map((m) => m.name);
+            return (data.models ?? []).map((m) => ({
+                name: m.name,
+                size: m.size,
+                modified_at: m.modified_at,
+                digest: m.digest,
+            }));
         }
     }
     catch { /* not running */ }
     return [];
 }
+/** List models installed in Ollama */
+export async function listInstalledModels() {
+    const models = await listInstalledModelsDetailed();
+    return models.map((m) => m.name);
+}
+/** Stop all loaded models (frees VRAM) and kill the Ollama server process */
+export async function stopOllama() {
+    try {
+        // First unload all models from memory
+        try {
+            execSync("ollama stop", { stdio: ["pipe", "pipe", "pipe"], timeout: 5000 });
+        }
+        catch { /* may fail if no models loaded */ }
+        // Kill the server process
+        if (process.platform === "win32") {
+            execSync("taskkill /f /im ollama.exe", { stdio: ["pipe", "pipe", "pipe"], timeout: 5000 });
+        }
+        else if (process.platform === "darwin") {
+            // Try launchctl first (Ollama app), then pkill
+            try {
+                execSync("launchctl stop com.ollama.ollama", { stdio: ["pipe", "pipe", "pipe"], timeout: 3000 });
+            }
+            catch {
+                try {
+                    execSync("pkill ollama", { stdio: ["pipe", "pipe", "pipe"], timeout: 3000 });
+                }
+                catch { /* already stopped */ }
+            }
+        }
+        else {
+            // Linux
+            try {
+                execSync("systemctl stop ollama", { stdio: ["pipe", "pipe", "pipe"], timeout: 3000 });
+            }
+            catch {
+                try {
+                    execSync("pkill ollama", { stdio: ["pipe", "pipe", "pipe"], timeout: 3000 });
+                }
+                catch { /* already stopped */ }
+            }
+        }
+        // Verify it stopped
+        await new Promise(r => setTimeout(r, 500));
+        const stillRunning = await isOllamaRunning();
+        if (stillRunning) {
+            return { ok: false, message: "Ollama is still running. Try killing it manually." };
+        }
+        return { ok: true, message: "Ollama stopped." };
+    }
+    catch (err) {
+        return { ok: false, message: `Failed to stop Ollama: ${err.message}` };
+    }
+}
+/** Delete a model from disk */
+export function deleteModel(modelId) {
+    try {
+        execSync(`ollama rm ${modelId}`, { stdio: ["pipe", "pipe", "pipe"], timeout: 30000 });
+        return { ok: true, message: `Deleted ${modelId}` };
+    }
+    catch (err) {
+        return { ok: false, message: `Failed to delete ${modelId}: ${err.stderr?.toString().trim() || err.message}` };
+    }
+}
+/** Get GPU memory usage info (best-effort) */
+export function getGPUMemoryUsage() {
+    try {
+        if (process.platform === "darwin") {
+            // Apple Silicon — check memory pressure
+            const raw = execSync("memory_pressure", { encoding: "utf-8", timeout: 3000, stdio: ["pipe", "pipe", "pipe"] });
+            const match = raw.match(/System-wide memory free percentage:\s*(\d+)%/);
+            if (match) {
+                return `${100 - parseInt(match[1])}% system memory in use`;
+            }
+            return null;
+        }
+        // NVIDIA GPU
+        const raw = execSync("nvidia-smi --query-gpu=memory.used,memory.total --format=csv,noheader,nounits", {
+            encoding: "utf-8", timeout: 3000, stdio: ["pipe", "pipe", "pipe"],
+        });
+        const parts = raw.trim().split(",").map(s => s.trim());
+        if (parts.length === 2) {
+            return `${parts[0]} MiB / ${parts[1]} MiB GPU memory`;
+        }
+    }
+    catch { /* no GPU info available */ }
+    return null;
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "codemaxxing",
-  "version": "0.4.0",
+  "version": "0.4.1",
   "description": "Open-source terminal coding agent. Connect any LLM. Max your code.",
   "main": "dist/index.js",
   "bin": {

package/src/config.ts

@@ -24,6 +24,7 @@ export interface CodemaxxingConfig {
     contextCompressionThreshold?: number;
     architectModel?: string;
     autoLint?: boolean;
+    stopOllamaOnExit?: boolean;
   };
 }
 
@@ -138,6 +139,20 @@ export function loadConfig(): CodemaxxingConfig {
   }
 }
 
+/** Save config to disk (merges with existing) */
+export function saveConfig(updates: Partial<CodemaxxingConfig>): void {
+  const current = loadConfig();
+  const merged = {
+    ...current,
+    ...updates,
+    defaults: { ...current.defaults, ...(updates.defaults ?? {}) },
+  };
+  if (!existsSync(CONFIG_DIR)) {
+    mkdirSync(CONFIG_DIR, { recursive: true });
+  }
+  writeFileSync(CONFIG_FILE, JSON.stringify(merged, null, 2));
+}
+
 /**
  * Apply CLI overrides to a provider config
  */

package/src/index.tsx

@@ -5,7 +5,7 @@ import { render, Box, Text, useInput, useApp, useStdout } from "ink";
 import { EventEmitter } from "events";
 import TextInput from "ink-text-input";
 import { CodingAgent } from "./agent.js";
-import { loadConfig, detectLocalProvider, parseCLIArgs, applyOverrides, listModels } from "./config.js";
+import { loadConfig, saveConfig, detectLocalProvider, parseCLIArgs, applyOverrides, listModels } from "./config.js";
 import { listSessions, getSession, loadMessages, deleteSession } from "./utils/sessions.js";
 import { execSync } from "child_process";
 import { isGitRepo, getBranch, getStatus, getDiff, undoLastCommit } from "./utils/git.js";
@@ -14,8 +14,8 @@ import { PROVIDERS, getCredentials, openRouterOAuth, anthropicSetupToken, import
 import { listInstalledSkills, installSkill, removeSkill, getRegistrySkills, searchRegistry, createSkillScaffold, getActiveSkills, getActiveSkillCount } from "./utils/skills.js";
 import { listServers, addServer, removeServer, getAllMCPTools, getConnectedServers } from "./utils/mcp.js";
 import { detectHardware, formatBytes, type HardwareInfo } from "./utils/hardware.js";
-import { getRecommendations, getFitIcon, type ScoredModel } from "./utils/models.js";
-import { isOllamaInstalled, isOllamaRunning, getOllamaInstallCommand, startOllama, pullModel, type PullProgress } from "./utils/ollama.js";
+import { getRecommendations, getRecommendationsWithLlmfit, getFitIcon, isLlmfitAvailable, type ScoredModel } from "./utils/models.js";
+import { isOllamaInstalled, isOllamaRunning, getOllamaInstallCommand, startOllama, stopOllama, pullModel, listInstalledModelsDetailed, deleteModel, getGPUMemoryUsage, type PullProgress } from "./utils/ollama.js";
 
 const VERSION = "0.1.9";
 
@@ -67,6 +67,12 @@ const SLASH_COMMANDS = [
   { cmd: "/mcp add", desc: "add MCP server" },
   { cmd: "/mcp remove", desc: "remove MCP server" },
   { cmd: "/mcp reconnect", desc: "reconnect MCP servers" },
+  { cmd: "/ollama", desc: "Ollama status & models" },
+  { cmd: "/ollama list", desc: "list installed models" },
+  { cmd: "/ollama start", desc: "start Ollama server" },
+  { cmd: "/ollama stop", desc: "stop Ollama server" },
+  { cmd: "/ollama pull", desc: "download a model" },
+  { cmd: "/ollama delete", desc: "delete a model" },
   { cmd: "/quit", desc: "exit" },
 ];
 
@@ -175,6 +181,11 @@ function App() {
     resolve: (decision: "yes" | "no" | "always") => void;
   } | null>(null);
 
+  // ── Ollama Management State ──
+  const [ollamaDeleteConfirm, setOllamaDeleteConfirm] = useState<{ model: string; size: number } | null>(null);
+  const [ollamaPulling, setOllamaPulling] = useState<{ model: string; progress: PullProgress } | null>(null);
+  const [ollamaExitPrompt, setOllamaExitPrompt] = useState(false);
+
   // ── Setup Wizard State ──
   type WizardScreen = "connection" | "models" | "install-ollama" | "pulling" | null;
   const [wizardScreen, setWizardScreen] = useState<WizardScreen>(null);
@@ -383,7 +394,8 @@ function App() {
         // Commands that need args (like /commit, /model) — fill input instead of executing
         if (selected.cmd === "/commit" || selected.cmd === "/model" || selected.cmd === "/session delete" ||
             selected.cmd === "/skills install" || selected.cmd === "/skills remove" || selected.cmd === "/skills search" ||
-            selected.cmd === "/skills on" || selected.cmd === "/skills off" || selected.cmd === "/architect") {
+            selected.cmd === "/skills on" || selected.cmd === "/skills off" || selected.cmd === "/architect" ||
+            selected.cmd === "/ollama pull" || selected.cmd === "/ollama delete") {
           setInput(selected.cmd + " ");
           setCmdIndex(0);
           setInputKey((k) => k + 1);
@@ -411,7 +423,20 @@ function App() {
     addMsg("user", trimmed);
 
     if (trimmed === "/quit" || trimmed === "/exit") {
-      exit();
+      // Check if Ollama is running and offer to stop it
+      const ollamaUp = await isOllamaRunning();
+      if (ollamaUp) {
+        const config = loadConfig();
+        if (config.defaults.stopOllamaOnExit) {
+          addMsg("info", "Stopping Ollama...");
+          await stopOllama();
+          exit();
+        } else {
+          setOllamaExitPrompt(true);
+        }
+      } else {
+        exit();
+      }
       return;
     }
     if (trimmed === "/login" || trimmed === "/auth") {
@@ -454,6 +479,12 @@ function App() {
         "  /mcp add   — add MCP server to global config",
         "  /mcp remove — remove MCP server",
         "  /mcp reconnect — reconnect all MCP servers",
+        "  /ollama    — Ollama status, models & GPU usage",
+        "  /ollama list — list installed models with sizes",
+        "  /ollama start — start Ollama server",
+        "  /ollama stop — stop Ollama server (frees GPU RAM)",
+        "  /ollama pull <model> — download a model",
+        "  /ollama delete <model> — delete a model from disk",
         "  /quit      — exit",
       ].join("\n"));
       return;
@@ -614,6 +645,135 @@ function App() {
       return;
     }
 
+    // ── Ollama commands (work without agent) ──
+    if (trimmed === "/ollama" || trimmed === "/ollama status") {
+      const running = await isOllamaRunning();
+      const lines: string[] = [`Ollama: ${running ? "running" : "stopped"}`];
+      if (running) {
+        const models = await listInstalledModelsDetailed();
+        if (models.length > 0) {
+          lines.push(`Installed models (${models.length}):`);
+          for (const m of models) {
+            const sizeGB = (m.size / (1024 * 1024 * 1024)).toFixed(1);
+            lines.push(`  ${m.name}  (${sizeGB} GB)`);
+          }
+        } else {
+          lines.push("No models installed.");
+        }
+        const gpuMem = getGPUMemoryUsage();
+        if (gpuMem) lines.push(`GPU: ${gpuMem}`);
+      } else {
+        lines.push("Start with: /ollama start");
+      }
+      addMsg("info", lines.join("\n"));
+      return;
+    }
+    if (trimmed === "/ollama list") {
+      const running = await isOllamaRunning();
+      if (!running) {
+        addMsg("info", "Ollama is not running. Start with /ollama start");
+        return;
+      }
+      const models = await listInstalledModelsDetailed();
+      if (models.length === 0) {
+        addMsg("info", "No models installed. Pull one with /ollama pull <model>");
+      } else {
+        const lines = models.map((m) => {
+          const sizeGB = (m.size / (1024 * 1024 * 1024)).toFixed(1);
+          return `  ${m.name}  (${sizeGB} GB)`;
+        });
+        addMsg("info", `Installed models:\n${lines.join("\n")}`);
+      }
+      return;
+    }
+    if (trimmed === "/ollama start") {
+      const running = await isOllamaRunning();
+      if (running) {
+        addMsg("info", "Ollama is already running.");
+        return;
+      }
+      if (!isOllamaInstalled()) {
+        addMsg("error", `Ollama is not installed. Install with: ${getOllamaInstallCommand(process.platform === "darwin" ? "macos" : process.platform === "win32" ? "windows" : "linux")}`);
+        return;
+      }
+      startOllama();
+      addMsg("info", "Starting Ollama server...");
+      // Wait for it to come up
+      for (let i = 0; i < 10; i++) {
+        await new Promise(r => setTimeout(r, 1000));
+        if (await isOllamaRunning()) {
+          addMsg("info", "Ollama is running.");
+          return;
+        }
+      }
+      addMsg("error", "Ollama did not start in time. Try running 'ollama serve' manually.");
+      return;
+    }
+    if (trimmed === "/ollama stop") {
+      const running = await isOllamaRunning();
+      if (!running) {
+        addMsg("info", "Ollama is not running.");
+        return;
+      }
+      addMsg("info", "Stopping Ollama...");
+      const result = await stopOllama();
+      addMsg(result.ok ? "info" : "error", result.ok ? `\u2705 ${result.message}` : `\u274C ${result.message}`);
+      return;
+    }
+    if (trimmed.startsWith("/ollama pull ")) {
+      const modelId = trimmed.replace("/ollama pull ", "").trim();
+      if (!modelId) {
+        addMsg("info", "Usage: /ollama pull <model>\n  Example: /ollama pull qwen2.5-coder:7b");
+        return;
+      }
+      if (!isOllamaInstalled()) {
+        addMsg("error", "Ollama is not installed.");
+        return;
+      }
+      // Ensure ollama is running
+      let running = await isOllamaRunning();
+      if (!running) {
+        startOllama();
+        addMsg("info", "Starting Ollama server...");
+        for (let i = 0; i < 10; i++) {
+          await new Promise(r => setTimeout(r, 1000));
+          if (await isOllamaRunning()) { running = true; break; }
+        }
+        if (!running) {
+          addMsg("error", "Could not start Ollama. Run 'ollama serve' manually.");
+          return;
+        }
+      }
+      setOllamaPulling({ model: modelId, progress: { status: "starting", percent: 0 } });
+      try {
+        await pullModel(modelId, (p) => {
+          setOllamaPulling({ model: modelId, progress: p });
+        });
+        setOllamaPulling(null);
+        addMsg("info", `\u2705 Downloaded ${modelId}`);
+      } catch (err: any) {
+        setOllamaPulling(null);
+        addMsg("error", `Failed to pull ${modelId}: ${err.message}`);
+      }
+      return;
+    }
+    if (trimmed.startsWith("/ollama delete ")) {
+      const modelId = trimmed.replace("/ollama delete ", "").trim();
+      if (!modelId) {
+        addMsg("info", "Usage: /ollama delete <model>");
+        return;
+      }
+      // Look up size for confirmation
+      const models = await listInstalledModelsDetailed();
+      const found = models.find((m) => m.name === modelId || m.name.startsWith(modelId));
+      if (!found) {
+        addMsg("error", `Model "${modelId}" not found. Use /ollama list to see installed models.`);
+        return;
+      }
+      setOllamaDeleteConfirm({ model: found.name, size: found.size });
+      return;
+    }
+
     // ── MCP commands (partially work without agent) ──
     if (trimmed === "/mcp" || trimmed === "/mcp list") {
       const servers = listServers(process.cwd());
@@ -1155,6 +1315,49 @@ function App() {
       return;
     }
 
+    // ── Ollama delete confirmation ──
+    if (ollamaDeleteConfirm) {
+      if (inputChar === "y" || inputChar === "Y") {
+        const model = ollamaDeleteConfirm.model;
+        setOllamaDeleteConfirm(null);
+        const result = deleteModel(model);
+        addMsg(result.ok ? "info" : "error", result.ok ? `\u2705 ${result.message}` : `\u274C ${result.message}`);
+        return;
+      }
+      if (inputChar === "n" || inputChar === "N" || key.escape) {
+        setOllamaDeleteConfirm(null);
+        addMsg("info", "Delete cancelled.");
+        return;
+      }
+      return;
+    }
+
+    // ── Ollama exit prompt ──
+    if (ollamaExitPrompt) {
+      if (inputChar === "y" || inputChar === "Y") {
+        setOllamaExitPrompt(false);
+        stopOllama().then(() => exit());
+        return;
+      }
+      if (inputChar === "n" || inputChar === "N") {
+        setOllamaExitPrompt(false);
+        exit();
+        return;
+      }
+      if (inputChar === "a" || inputChar === "A") {
+        setOllamaExitPrompt(false);
+        saveConfig({ defaults: { ...loadConfig().defaults, stopOllamaOnExit: true } });
+        addMsg("info", "Saved preference: always stop Ollama on exit.");
+        stopOllama().then(() => exit());
+        return;
+      }
+      if (key.escape) {
+        setOllamaExitPrompt(false);
+        return;
+      }
+      return;
+    }
+
     // ── Setup Wizard Navigation ──
     if (wizardScreen) {
       if (wizardScreen === "connection") {
@@ -1174,11 +1377,11 @@ function App() {
         if (key.return) {
           const selected = items[wizardIndex];
           if (selected === "local") {
-            // Scan hardware and show model picker
+            // Scan hardware and show model picker (use llmfit if available)
             const hw = detectHardware();
             setWizardHardware(hw);
-            const recs = getRecommendations(hw).filter(m => m.fit !== "skip");
-            setWizardModels(recs);
+            const { models: recs } = getRecommendationsWithLlmfit(hw);
+            setWizardModels(recs.filter(m => m.fit !== "skip"));
             setWizardScreen("models");
             setWizardIndex(0);
           } else if (selected === "openrouter") {
@@ -1527,7 +1730,13 @@ function App() {
 
     if (key.ctrl && inputChar === "c") {
       if (ctrlCPressed) {
-        exit();
+        // Force quit on second Ctrl+C — don't block
+        const config = loadConfig();
+        if (config.defaults.stopOllamaOnExit) {
+          stopOllama().finally(() => exit());
+        } else {
+          exit();
+        }
       } else {
         setCtrlCPressed(true);
         addMsg("info", "Press Ctrl+C again to exit.");
@@ -1832,6 +2041,52 @@ function App() {
         </Box>
       )}
 
+      {/* ═══ OLLAMA DELETE CONFIRM ═══ */}
+      {ollamaDeleteConfirm && (
+        <Box flexDirection="column" borderStyle="single" borderColor={theme.colors.warning} paddingX={1} marginBottom={0}>
+          <Text bold color={theme.colors.warning}>Delete {ollamaDeleteConfirm.model} ({(ollamaDeleteConfirm.size / (1024 * 1024 * 1024)).toFixed(1)} GB)?</Text>
+          <Text>
+            <Text color={theme.colors.error} bold> [y]</Text><Text>es  </Text>
+            <Text color={theme.colors.success} bold>[n]</Text><Text>o</Text>
+          </Text>
+        </Box>
+      )}
+
+      {/* ═══ OLLAMA PULL PROGRESS ═══ */}
+      {ollamaPulling && (
+        <Box flexDirection="column" borderStyle="single" borderColor={theme.colors.border} paddingX={1} marginBottom={0}>
+          <Text bold color={theme.colors.secondary}>{"  Downloading "}{ollamaPulling.model}{"..."}</Text>
+          {ollamaPulling.progress.status === "downloading" || ollamaPulling.progress.percent > 0 ? (
+            <Text>
+              {"  "}
+              <Text color={theme.colors.primary}>
+                {"\u2588".repeat(Math.floor(ollamaPulling.progress.percent / 5))}
+                {"\u2591".repeat(20 - Math.floor(ollamaPulling.progress.percent / 5))}
+              </Text>
+              {"  "}<Text bold>{ollamaPulling.progress.percent}%</Text>
+              {ollamaPulling.progress.completed != null && ollamaPulling.progress.total != null ? (
+                <Text color={theme.colors.muted}>{" \u00B7 "}{formatBytes(ollamaPulling.progress.completed)}{" / "}{formatBytes(ollamaPulling.progress.total)}</Text>
+              ) : null}
+            </Text>
+          ) : (
+            <Text color={theme.colors.muted}>{"  "}{ollamaPulling.progress.status}...</Text>
+          )}
+        </Box>
+      )}
+
+      {/* ═══ OLLAMA EXIT PROMPT ═══ */}
+      {ollamaExitPrompt && (
+        <Box flexDirection="column" borderStyle="single" borderColor={theme.colors.warning} paddingX={1} marginBottom={0}>
+          <Text bold color={theme.colors.warning}>Ollama is still running.</Text>
+          <Text color={theme.colors.muted}>Stop it to free GPU memory?</Text>
+          <Text>
+            <Text color={theme.colors.success} bold> [y]</Text><Text>es  </Text>
+            <Text color={theme.colors.error} bold>[n]</Text><Text>o  </Text>
+            <Text color={theme.colors.primary} bold>[a]</Text><Text>lways</Text>
+          </Text>
+        </Box>
+      )}
+
       {/* ═══ SETUP WIZARD ═══ */}
       {wizardScreen === "connection" && (
         <Box flexDirection="column" borderStyle="single" borderColor={theme.colors.border} paddingX={1} marginBottom={0}>
@@ -1864,6 +2119,9 @@ function App() {
           ) : (
             <Text color={theme.colors.muted}>{"  GPU: none detected"}</Text>
           )}
+          {!isLlmfitAvailable() && (
+            <Text dimColor>{"  Tip: Install llmfit for smarter recommendations: brew install llmfit"}</Text>
+          )}
           <Text>{""}</Text>
           <Text bold color={theme.colors.secondary}>Recommended models:</Text>
           <Text>{""}</Text>

package/src/utils/models.ts

@@ -1,3 +1,4 @@
+import { execSync } from "child_process";
 import type { HardwareInfo } from "./hardware.js";
 
 export interface RecommendedModel {
@@ -135,3 +136,82 @@ export function getFitIcon(fit: ModelFit): string {
     case "skip": return "\u274C";    // ❌
   }
 }
+
+/** Check if llmfit binary is available */
+export function isLlmfitAvailable(): boolean {
+  try {
+    const cmd = process.platform === "win32" ? "where llmfit" : "which llmfit";
+    execSync(cmd, { stdio: ["pipe", "pipe", "pipe"], timeout: 3000 });
+    return true;
+  } catch {
+    return false;
+  }
+}
+
+interface LlmfitModel {
+  name: string;
+  provider: string;
+  params_b: number;
+  quant: string;
+  fit: string;
+  estimated_tps: number;
+  vram_gb: number;
+  ram_gb: number;
+}
+
+function mapLlmfitFit(fit: string): ModelFit {
+  switch (fit) {
+    case "perfect": return "perfect";
+    case "good": return "good";
+    case "marginal": return "tight";
+    default: return "skip";
+  }
+}
+
+function mapLlmfitQuality(params_b: number): "good" | "great" | "best" {
+  if (params_b >= 14) return "best";
+  if (params_b >= 7) return "great";
+  return "good";
+}
+
+/** Get recommendations using llmfit if available, otherwise fall back to hardcoded list */
+export function getRecommendationsWithLlmfit(hardware: HardwareInfo): { models: ScoredModel[]; usedLlmfit: boolean } {
+  if (!isLlmfitAvailable()) {
+    return { models: getRecommendations(hardware), usedLlmfit: false };
+  }
+
+  try {
+    const raw = execSync("llmfit recommend --use-case coding --format json --limit 10", {
+      encoding: "utf-8",
+      timeout: 15000,
+      stdio: ["pipe", "pipe", "pipe"],
+    });
+
+    const llmfitModels: LlmfitModel[] = JSON.parse(raw.trim());
+    if (!Array.isArray(llmfitModels) || llmfitModels.length === 0) {
+      return { models: getRecommendations(hardware), usedLlmfit: false };
+    }
+
+    const scored: ScoredModel[] = llmfitModels
+      .filter((m) => m.provider === "ollama")
+      .map((m) => ({
+        name: m.name.split(":")[0]?.replace(/-/g, " ").replace(/\b\w/g, c => c.toUpperCase()) + ` ${m.params_b}B`,
+        ollamaId: m.name,
+        size: Math.ceil(m.vram_gb),
+        ramRequired: Math.ceil(m.ram_gb),
+        vramOptimal: Math.ceil(m.vram_gb),
+        description: `${m.quant} · ~${m.estimated_tps.toFixed(0)} tok/s`,
+        speed: `~${m.estimated_tps.toFixed(0)} tok/s`,
+        quality: mapLlmfitQuality(m.params_b),
+        fit: mapLlmfitFit(m.fit),
+      }));
+
+    if (scored.length === 0) {
+      return { models: getRecommendations(hardware), usedLlmfit: false };
+    }
+
+    return { models: scored, usedLlmfit: true };
+  } catch {
+    return { models: getRecommendations(hardware), usedLlmfit: false };
+  }
+}

package/src/utils/ollama.ts

@@ -121,17 +121,112 @@ export function pullModel(
   });
 }
 
-/** List models installed in Ollama */
-export async function listInstalledModels(): Promise<string[]> {
+export interface OllamaModelInfo {
+  name: string;
+  size: number; // bytes
+  modified_at: string;
+  digest: string;
+}
+
+/** List models installed in Ollama with detailed info */
+export async function listInstalledModelsDetailed(): Promise<OllamaModelInfo[]> {
   try {
     const controller = new AbortController();
     const timeout = setTimeout(() => controller.abort(), 3000);
     const res = await fetch("http://localhost:11434/api/tags", { signal: controller.signal });
     clearTimeout(timeout);
     if (res.ok) {
-      const data = (await res.json()) as { models?: Array<{ name: string }> };
-      return (data.models ?? []).map((m) => m.name);
+      const data = (await res.json()) as { models?: Array<{ name: string; size: number; modified_at: string; digest: string }> };
+      return (data.models ?? []).map((m) => ({
+        name: m.name,
+        size: m.size,
+        modified_at: m.modified_at,
+        digest: m.digest,
+      }));
     }
   } catch { /* not running */ }
   return [];
 }
+
+/** List models installed in Ollama */
+export async function listInstalledModels(): Promise<string[]> {
+  const models = await listInstalledModelsDetailed();
+  return models.map((m) => m.name);
+}
+
+/** Stop all loaded models (frees VRAM) and kill the Ollama server process */
+export async function stopOllama(): Promise<{ ok: boolean; message: string }> {
+  try {
+    // First unload all models from memory
+    try {
+      execSync("ollama stop", { stdio: ["pipe", "pipe", "pipe"], timeout: 5000 });
+    } catch { /* may fail if no models loaded */ }
+
+    // Kill the server process
+    if (process.platform === "win32") {
+      execSync("taskkill /f /im ollama.exe", { stdio: ["pipe", "pipe", "pipe"], timeout: 5000 });
+    } else if (process.platform === "darwin") {
+      // Try launchctl first (Ollama app), then pkill
+      try {
+        execSync("launchctl stop com.ollama.ollama", { stdio: ["pipe", "pipe", "pipe"], timeout: 3000 });
+      } catch {
+        try {
+          execSync("pkill ollama", { stdio: ["pipe", "pipe", "pipe"], timeout: 3000 });
+        } catch { /* already stopped */ }
+      }
+    } else {
+      // Linux
+      try {
+        execSync("systemctl stop ollama", { stdio: ["pipe", "pipe", "pipe"], timeout: 3000 });
+      } catch {
+        try {
+          execSync("pkill ollama", { stdio: ["pipe", "pipe", "pipe"], timeout: 3000 });
+        } catch { /* already stopped */ }
+      }
+    }
+
+    // Verify it stopped
+    await new Promise(r => setTimeout(r, 500));
+    const stillRunning = await isOllamaRunning();
+    if (stillRunning) {
+      return { ok: false, message: "Ollama is still running. Try killing it manually." };
+    }
+    return { ok: true, message: "Ollama stopped." };
+  } catch (err: any) {
+    return { ok: false, message: `Failed to stop Ollama: ${err.message}` };
+  }
+}
+
+/** Delete a model from disk */
+export function deleteModel(modelId: string): { ok: boolean; message: string } {
+  try {
+    execSync(`ollama rm ${modelId}`, { stdio: ["pipe", "pipe", "pipe"], timeout: 30000 });
+    return { ok: true, message: `Deleted ${modelId}` };
+  } catch (err: any) {
+    return { ok: false, message: `Failed to delete ${modelId}: ${err.stderr?.toString().trim() || err.message}` };
+  }
+}
+
+/** Get GPU memory usage info (best-effort) */
+export function getGPUMemoryUsage(): string | null {
+  try {
+    if (process.platform === "darwin") {
+      // Apple Silicon — check memory pressure
+      const raw = execSync("memory_pressure", { encoding: "utf-8", timeout: 3000, stdio: ["pipe", "pipe", "pipe"] });
+      const match = raw.match(/System-wide memory free percentage:\s*(\d+)%/);
+      if (match) {
+        return `${100 - parseInt(match[1])}% system memory in use`;
+      }
+      return null;
+    }
+    // NVIDIA GPU
+    const raw = execSync("nvidia-smi --query-gpu=memory.used,memory.total --format=csv,noheader,nounits", {
+      encoding: "utf-8", timeout: 3000, stdio: ["pipe", "pipe", "pipe"],
+    });
+    const parts = raw.trim().split(",").map(s => s.trim());
+    if (parts.length === 2) {
+      return `${parts[0]} MiB / ${parts[1]} MiB GPU memory`;
+    }
+  } catch { /* no GPU info available */ }
+  return null;
+}