@hasna/terminal 0.1.3 → 0.1.4

package/dist/App.js

@@ -4,9 +4,12 @@ import { Box, Text, useInput, useApp } from "ink";
 import { spawn } from "child_process";
 import { translateToCommand, explainCommand, fixCommand, checkPermissions, isIrreversible } from "./ai.js";
 import { loadHistory, appendHistory, loadConfig, saveConfig, } from "./history.js";
+import { loadCache } from "./cache.js";
 import Onboarding from "./Onboarding.js";
 import StatusBar from "./StatusBar.js";
 import Spinner from "./Spinner.js";
+// warm cache on startup
+loadCache();
 const MAX_LINES = 20;
 // ── helpers ───────────────────────────────────────────────────────────────────
 function insertAt(s, pos, ch) { return s.slice(0, pos) + ch + s.slice(pos); }
@@ -131,9 +134,11 @@ export default function App() {
                     await runPhase(nl, nl, true);
                     return;
                 }
-                setPhase({ type: "thinking", nl, raw: false });
+                setPhase({ type: "thinking", nl, raw: false, partial: "" });
                 try {
-                    const command = await translateToCommand(nl, config.permissions, sessionCmds);
+                    const command = await translateToCommand(nl, config.permissions, sessionCmds, (partial) => {
+                        setPhase({ type: "thinking", nl, raw: false, partial });
+                    });
                     const blocked = checkPermissions(command, config.permissions);
                     if (blocked) {
                         pushScroll({ nl, cmd: command, lines: [`blocked: ${blocked}`], truncated: false, error: true });
@@ -172,7 +177,7 @@ export default function App() {
             }
             if (input === "?") {
                 const { nl, command } = phase;
-                setPhase({ type: "thinking", nl, raw: false });
+                setPhase({ type: "thinking", nl, raw: false, partial: "" });
                 try {
                     const explanation = await explainCommand(command);
                     setPhase({ type: "explain", nl, command, explanation });
@@ -213,7 +218,7 @@ export default function App() {
             }
             if (input === "y" || input === "Y" || key.return) {
                 const { nl, command, errorOutput } = phase;
-                setPhase({ type: "thinking", nl, raw: false });
+                setPhase({ type: "thinking", nl, raw: false, partial: "" });
                 try {
                     const fixed = await fixCommand(nl, command, errorOutput, config.permissions, sessionCmds);
                     const danger = isIrreversible(fixed);
@@ -249,5 +254,5 @@ export default function App() {
     }
     const isRaw = phase.type === "input" && phase.raw;
     // ── render ─────────────────────────────────────────────────────────────────
-    return (_jsxs(Box, { flexDirection: "column", children: [scroll.map((entry, i) => (_jsxs(Box, { flexDirection: "column", marginBottom: 1, paddingLeft: 2, children: [_jsxs(Box, { gap: 2, children: [_jsx(Text, { dimColor: true, children: "\u203A" }), _jsx(Text, { dimColor: true, children: entry.nl })] }), entry.nl !== entry.cmd && (_jsxs(Box, { gap: 2, paddingLeft: 2, children: [_jsx(Text, { dimColor: true, children: "$" }), _jsx(Text, { dimColor: true, children: entry.cmd })] })), entry.lines.length > 0 && (_jsxs(Box, { flexDirection: "column", paddingLeft: 4, children: [entry.lines.map((line, j) => (_jsx(Text, { color: entry.error ? "red" : undefined, children: line }, j))), entry.truncated && !entry.expanded && (_jsx(Text, { dimColor: true, children: "\u2026 (space to expand)" }))] }))] }, i))), phase.type === "confirm" && (_jsxs(Box, { flexDirection: "column", marginBottom: 1, paddingLeft: 2, children: [_jsxs(Box, { gap: 2, children: [_jsx(Text, { dimColor: true, children: "\u203A" }), _jsx(Text, { dimColor: true, children: phase.nl })] }), _jsxs(Box, { gap: 2, paddingLeft: 2, children: [_jsx(Text, { dimColor: true, children: "$" }), _jsx(Text, { children: phase.command }), phase.danger && _jsx(Text, { color: "red", children: " \u26A0 irreversible" })] }), _jsx(Box, { paddingLeft: 4, children: _jsx(Text, { dimColor: true, children: "enter  n  e  ?" }) })] })), phase.type === "explain" && (_jsxs(Box, { flexDirection: "column", marginBottom: 1, paddingLeft: 2, children: [_jsxs(Box, { gap: 2, paddingLeft: 2, children: [_jsx(Text, { dimColor: true, children: "$" }), _jsx(Text, { children: phase.command })] }), _jsx(Box, { paddingLeft: 4, children: _jsx(Text, { dimColor: true, children: phase.explanation }) }), _jsx(Box, { paddingLeft: 4, children: _jsx(Text, { dimColor: true, children: "any key to continue" }) })] })), phase.type === "autofix" && (_jsx(Box, { flexDirection: "column", marginBottom: 1, paddingLeft: 2, children: _jsx(Text, { dimColor: true, children: "  command failed \u2014 retry with fix?  [enter / n]" }) })), phase.type === "thinking" && _jsx(Spinner, { label: "translating" }), phase.type === "running" && (_jsxs(Box, { flexDirection: "column", paddingLeft: 2, children: [_jsxs(Box, { gap: 2, children: [_jsx(Text, { dimColor: true, children: "$" }), _jsx(Text, { dimColor: true, children: phase.command })] }), streamLines.length > 0 && (_jsx(Box, { flexDirection: "column", paddingLeft: 2, children: streamLines.slice(-MAX_LINES).map((line, i) => (_jsx(Text, { children: line }, i))) })), _jsx(Spinner, { label: "ctrl+c to cancel" })] })), phase.type === "error" && (_jsx(Box, { paddingLeft: 2, children: _jsx(Text, { color: "red", children: phase.message }) })), phase.type === "input" && (_jsxs(Box, { gap: 2, paddingLeft: 2, children: [_jsx(Text, { dimColor: true, children: isRaw ? "$" : "›" }), _jsxs(Box, { children: [_jsx(Text, { children: phase.value.slice(0, phase.cursor) }), _jsx(Text, { inverse: true, children: phase.value[phase.cursor] ?? " " }), _jsx(Text, { children: phase.value.slice(phase.cursor + 1) })] })] })), _jsx(StatusBar, { permissions: config.permissions })] }));
+    return (_jsxs(Box, { flexDirection: "column", children: [scroll.map((entry, i) => (_jsxs(Box, { flexDirection: "column", marginBottom: 1, paddingLeft: 2, children: [_jsxs(Box, { gap: 2, children: [_jsx(Text, { dimColor: true, children: "\u203A" }), _jsx(Text, { dimColor: true, children: entry.nl })] }), entry.nl !== entry.cmd && (_jsxs(Box, { gap: 2, paddingLeft: 2, children: [_jsx(Text, { dimColor: true, children: "$" }), _jsx(Text, { dimColor: true, children: entry.cmd })] })), entry.lines.length > 0 && (_jsxs(Box, { flexDirection: "column", paddingLeft: 4, children: [entry.lines.map((line, j) => (_jsx(Text, { color: entry.error ? "red" : undefined, children: line }, j))), entry.truncated && !entry.expanded && (_jsx(Text, { dimColor: true, children: "\u2026 (space to expand)" }))] }))] }, i))), phase.type === "confirm" && (_jsxs(Box, { flexDirection: "column", marginBottom: 1, paddingLeft: 2, children: [_jsxs(Box, { gap: 2, children: [_jsx(Text, { dimColor: true, children: "\u203A" }), _jsx(Text, { dimColor: true, children: phase.nl })] }), _jsxs(Box, { gap: 2, paddingLeft: 2, children: [_jsx(Text, { dimColor: true, children: "$" }), _jsx(Text, { children: phase.command }), phase.danger && _jsx(Text, { color: "red", children: " \u26A0 irreversible" })] }), _jsx(Box, { paddingLeft: 4, children: _jsx(Text, { dimColor: true, children: "enter  n  e  ?" }) })] })), phase.type === "explain" && (_jsxs(Box, { flexDirection: "column", marginBottom: 1, paddingLeft: 2, children: [_jsxs(Box, { gap: 2, paddingLeft: 2, children: [_jsx(Text, { dimColor: true, children: "$" }), _jsx(Text, { children: phase.command })] }), _jsx(Box, { paddingLeft: 4, children: _jsx(Text, { dimColor: true, children: phase.explanation }) }), _jsx(Box, { paddingLeft: 4, children: _jsx(Text, { dimColor: true, children: "any key to continue" }) })] })), phase.type === "autofix" && (_jsx(Box, { flexDirection: "column", marginBottom: 1, paddingLeft: 2, children: _jsx(Text, { dimColor: true, children: "  command failed \u2014 retry with fix?  [enter / n]" }) })), phase.type === "thinking" && (phase.partial ? (_jsxs(Box, { flexDirection: "column", paddingLeft: 2, children: [_jsxs(Box, { gap: 2, children: [_jsx(Text, { dimColor: true, children: "\u203A" }), _jsx(Text, { dimColor: true, children: phase.nl })] }), _jsxs(Box, { gap: 2, paddingLeft: 2, children: [_jsx(Text, { dimColor: true, children: "$" }), _jsx(Text, { dimColor: true, children: phase.partial })] })] })) : (_jsx(Spinner, { label: "translating" }))), phase.type === "running" && (_jsxs(Box, { flexDirection: "column", paddingLeft: 2, children: [_jsxs(Box, { gap: 2, children: [_jsx(Text, { dimColor: true, children: "$" }), _jsx(Text, { dimColor: true, children: phase.command })] }), streamLines.length > 0 && (_jsx(Box, { flexDirection: "column", paddingLeft: 2, children: streamLines.slice(-MAX_LINES).map((line, i) => (_jsx(Text, { children: line }, i))) })), _jsx(Spinner, { label: "ctrl+c to cancel" })] })), phase.type === "error" && (_jsx(Box, { paddingLeft: 2, children: _jsx(Text, { color: "red", children: phase.message }) })), phase.type === "input" && (_jsxs(Box, { gap: 2, paddingLeft: 2, children: [_jsx(Text, { dimColor: true, children: isRaw ? "$" : "›" }), _jsxs(Box, { children: [_jsx(Text, { children: phase.value.slice(0, phase.cursor) }), _jsx(Text, { inverse: true, children: phase.value[phase.cursor] ?? " " }), _jsx(Text, { children: phase.value.slice(phase.cursor + 1) })] })] })), _jsx(StatusBar, { permissions: config.permissions })] }));
 }

package/dist/ai.js

@@ -1,9 +1,23 @@
 import Anthropic from "@anthropic-ai/sdk";
+import { cacheGet, cacheSet } from "./cache.js";
 const client = new Anthropic({ apiKey: process.env.ANTHROPIC_API_KEY });
+// ── model routing ─────────────────────────────────────────────────────────────
+// Simple queries → haiku (fast). Complex/ambiguous → sonnet.
+const COMPLEX_SIGNALS = [
+    /\b(undo|revert|rollback|previous|last)\b/i,
+    /\b(all files?|recursively|bulk|batch)\b/i,
+    /\b(pipeline|chain|then|and then|after)\b/i,
+    /\b(if|when|unless|only if)\b/i,
+    /[|&;]{2}/, // pipes / &&  in NL (unusual = complex intent)
+];
+function pickModel(nl) {
+    const isComplex = COMPLEX_SIGNALS.some((r) => r.test(nl)) || nl.split(" ").length > 10;
+    return isComplex ? "claude-sonnet-4-6" : "claude-haiku-4-5-20251001";
+}
 // ── irreversibility ───────────────────────────────────────────────────────────
 const IRREVERSIBLE_PATTERNS = [
     /\brm\s/, /\brmdir\b/, /\btruncate\b/, /\bdrop\s+table\b/i,
-    /\bdelete\s+from\b/i, /\bmv\b.*\/dev\/null/, /\b>\s*[^>]/, // overwrite redirect
+    /\bdelete\s+from\b/i, /\bmv\b.*\/dev\/null/, /\b>\s*[^>]/,
     /\bdd\b/, /\bmkfs\b/, /\bformat\b/, /\bshred\b/,
 ];
 export function isIrreversible(command) {
@@ -32,35 +46,82 @@ export function checkPermissions(command, perms) {
 function buildSystemPrompt(perms, sessionCmds) {
     const restrictions = [];
     if (!perms.destructive)
-        restrictions.push("- NEVER generate commands that delete, remove, or overwrite files/data (rm, rmdir, truncate, DROP TABLE, etc.)");
+        restrictions.push("- NEVER generate commands that delete, remove, or overwrite files/data");
     if (!perms.network)
-        restrictions.push("- NEVER generate commands that make network requests (curl, wget, ssh, scp, ping, nc, etc.)");
+        restrictions.push("- NEVER generate commands that make network requests (curl, wget, ssh, etc.)");
     if (!perms.sudo)
-        restrictions.push("- NEVER generate commands that use sudo or require root privileges");
+        restrictions.push("- NEVER generate commands requiring sudo");
     if (!perms.write_outside_cwd)
-        restrictions.push("- NEVER generate commands that write to paths outside the current working directory");
+        restrictions.push("- NEVER write to paths outside the current working directory");
     if (!perms.install)
-        restrictions.push("- NEVER generate commands that install packages (brew install, npm install -g, pip install, apt install, etc.)");
+        restrictions.push("- NEVER install packages (brew, npm -g, pip, apt, etc.)");
     const restrictionBlock = restrictions.length > 0
-        ? `\n\nCURRENT RESTRICTIONS (respect absolutely):\n${restrictions.join("\n")}\nIf restricted, output exactly: BLOCKED: <reason>`
+        ? `\n\nRESTRICTIONS:\n${restrictions.join("\n")}\nIf restricted, output: BLOCKED: <reason>`
         : "";
     const contextBlock = sessionCmds.length > 0
-        ? `\n\nRECENT COMMANDS THIS SESSION (for context — e.g. "undo that", "do the same for X"):\n${sessionCmds.map((c) => `$ ${c}`).join("\n")}`
+        ? `\n\nSESSION HISTORY:\n${sessionCmds.map((c) => `$ ${c}`).join("\n")}`
         : "";
-    const cwd = process.cwd();
-    return `You are a terminal assistant. The user will describe what they want to do in plain English.
-Your job is to output ONLY the exact shell command(s) to accomplish this — nothing else.
-No explanation. No markdown. No backticks. Just the raw command.
-If multiple commands are needed, join them with && or use a newline.
-Assume macOS/Linux zsh environment.
-Current working directory: ${cwd}${restrictionBlock}${contextBlock}`;
+    return `You are a terminal assistant. Output ONLY the exact shell command — no explanation, no markdown, no backticks.
+cwd: ${process.cwd()}
+shell: zsh / macOS${restrictionBlock}${contextBlock}`;
+}
+// ── streaming translate ───────────────────────────────────────────────────────
+export async function translateToCommand(nl, perms, sessionCmds, onToken) {
+    // cache hit — instant
+    const cached = cacheGet(nl);
+    if (cached) {
+        onToken?.(cached);
+        return cached;
+    }
+    const model = pickModel(nl);
+    let result = "";
+    if (onToken) {
+        // streaming path
+        const stream = await client.messages.stream({
+            model,
+            max_tokens: 256,
+            system: buildSystemPrompt(perms, sessionCmds),
+            messages: [{ role: "user", content: nl }],
+        });
+        for await (const chunk of stream) {
+            if (chunk.type === "content_block_delta" && chunk.delta.type === "text_delta") {
+                result += chunk.delta.text;
+                onToken(result.trim());
+            }
+        }
+    }
+    else {
+        const message = await client.messages.create({
+            model,
+            max_tokens: 256,
+            system: buildSystemPrompt(perms, sessionCmds),
+            messages: [{ role: "user", content: nl }],
+        });
+        const block = message.content[0];
+        if (block.type !== "text")
+            throw new Error("Unexpected response type");
+        result = block.text;
+    }
+    const text = result.trim();
+    if (text.startsWith("BLOCKED:"))
+        throw new Error(text);
+    cacheSet(nl, text);
+    return text;
+}
+// ── prefetch ──────────────────────────────────────────────────────────────────
+// Silently warm the cache after a command runs — no await, fire and forget
+export function prefetchNext(lastNl, perms, sessionCmds) {
+    // Only prefetch if we don't have it cached already
+    if (cacheGet(lastNl))
+        return;
+    translateToCommand(lastNl, perms, sessionCmds).catch(() => { });
 }
 // ── explain ───────────────────────────────────────────────────────────────────
 export async function explainCommand(command) {
     const message = await client.messages.create({
         model: "claude-haiku-4-5-20251001",
         max_tokens: 128,
-        system: "Explain what this shell command does in one plain English sentence. No markdown. No code blocks. Just a sentence.",
+        system: "Explain what this shell command does in one plain English sentence. No markdown, no code blocks.",
         messages: [{ role: "user", content: command }],
     });
     const block = message.content[0];
@@ -71,31 +132,13 @@ export async function explainCommand(command) {
 // ── auto-fix ──────────────────────────────────────────────────────────────────
 export async function fixCommand(originalNl, failedCommand, errorOutput, perms, sessionCmds) {
     const message = await client.messages.create({
-        model: "claude-opus-4-6",
+        model: "claude-sonnet-4-6",
         max_tokens: 256,
         system: buildSystemPrompt(perms, sessionCmds),
-        messages: [
-            {
+        messages: [{
                 role: "user",
-                content: `I wanted to: ${originalNl}\nI ran: ${failedCommand}\nIt failed with:\n${errorOutput}\n\nGive me the corrected command.`,
-            },
-        ],
-    });
-    const block = message.content[0];
-    if (block.type !== "text")
-        throw new Error("Unexpected response type");
-    const text = block.text.trim();
-    if (text.startsWith("BLOCKED:"))
-        throw new Error(text);
-    return text;
-}
-// ── translate ─────────────────────────────────────────────────────────────────
-export async function translateToCommand(nl, perms, sessionCmds) {
-    const message = await client.messages.create({
-        model: "claude-opus-4-6",
-        max_tokens: 256,
-        system: buildSystemPrompt(perms, sessionCmds),
-        messages: [{ role: "user", content: nl }],
+                content: `I wanted to: ${originalNl}\nI ran: ${failedCommand}\nError:\n${errorOutput}\n\nGive me the corrected command only.`,
+            }],
     });
     const block = message.content[0];
     if (block.type !== "text")

package/dist/cache.js

@@ -0,0 +1,41 @@
+// In-memory LRU cache + disk persistence for command translations
+import { existsSync, readFileSync, writeFileSync } from "fs";
+import { homedir } from "os";
+import { join } from "path";
+const CACHE_FILE = join(homedir(), ".terminal", "cache.json");
+const MAX_ENTRIES = 500;
+let mem = {};
+export function loadCache() {
+    if (!existsSync(CACHE_FILE))
+        return;
+    try {
+        mem = JSON.parse(readFileSync(CACHE_FILE, "utf8"));
+    }
+    catch { }
+}
+function persistCache() {
+    try {
+        writeFileSync(CACHE_FILE, JSON.stringify(mem));
+    }
+    catch { }
+}
+/** Normalize a natural language query for cache lookup */
+export function normalizeNl(nl) {
+    return nl
+        .toLowerCase()
+        .trim()
+        .replace(/[^a-z0-9\s]/g, "") // strip punctuation
+        .replace(/\s+/g, " ");
+}
+export function cacheGet(nl) {
+    return mem[normalizeNl(nl)] ?? null;
+}
+export function cacheSet(nl, command) {
+    const key = normalizeNl(nl);
+    // evict oldest if full
+    const keys = Object.keys(mem);
+    if (keys.length >= MAX_ENTRIES)
+        delete mem[keys[0]];
+    mem[key] = command;
+    persistCache();
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hasna/terminal",
-  "version": "0.1.3",
+  "version": "0.1.4",
   "description": "Natural language terminal — speak plain English, get shell commands",
   "type": "module",
   "bin": {

package/src/App.tsx

@@ -9,15 +9,19 @@ import {
   saveConfig,
   type Permissions,
 } from "./history.js";
+import { loadCache } from "./cache.js";
 import Onboarding from "./Onboarding.js";
 import StatusBar from "./StatusBar.js";
 import Spinner from "./Spinner.js";
 
+// warm cache on startup
+loadCache();
+
 // ── types ─────────────────────────────────────────────────────────────────────
 
 type Phase =
   | { type: "input";   value: string; cursor: number; histIdx: number; raw: boolean }
-  | { type: "thinking"; nl: string; raw: boolean }
+  | { type: "thinking"; nl: string; raw: boolean; partial: string }
   | { type: "confirm"; nl: string; command: string; raw: boolean; danger: boolean }
   | { type: "explain"; nl: string; command: string; explanation: string }
   | { type: "running"; nl: string; command: string }
@@ -175,9 +179,11 @@ export default function App() {
               return;
             }
 
-            setPhase({ type: "thinking", nl, raw: false });
+            setPhase({ type: "thinking", nl, raw: false, partial: "" });
             try {
-              const command = await translateToCommand(nl, config.permissions, sessionCmds);
+              const command = await translateToCommand(nl, config.permissions, sessionCmds, (partial) => {
+                setPhase({ type: "thinking", nl, raw: false, partial });
+              });
               const blocked = checkPermissions(command, config.permissions);
               if (blocked) {
                 pushScroll({ nl, cmd: command, lines: [`blocked: ${blocked}`], truncated: false, error: true });
@@ -215,7 +221,7 @@ export default function App() {
 
           if (input === "?") {
             const { nl, command } = phase;
-            setPhase({ type: "thinking", nl, raw: false });
+            setPhase({ type: "thinking", nl, raw: false, partial: "" });
             try {
               const explanation = await explainCommand(command);
               setPhase({ type: "explain", nl, command, explanation });
@@ -248,7 +254,7 @@ export default function App() {
           if (key.ctrl && input === "c") { exit(); return; }
           if (input === "y" || input === "Y" || key.return) {
             const { nl, command, errorOutput } = phase;
-            setPhase({ type: "thinking", nl, raw: false });
+            setPhase({ type: "thinking", nl, raw: false, partial: "" });
             try {
               const fixed = await fixCommand(nl, command, errorOutput, config.permissions, sessionCmds);
               const danger = isIrreversible(fixed);
@@ -355,8 +361,23 @@ export default function App() {
         </Box>
       )}
 
-      {/* spinners */}
-      {phase.type === "thinking" && <Spinner label="translating" />}
+      {/* thinking — show streaming partial or spinner */}
+      {phase.type === "thinking" && (
+        phase.partial ? (
+          <Box flexDirection="column" paddingLeft={2}>
+            <Box gap={2}>
+              <Text dimColor>›</Text>
+              <Text dimColor>{phase.nl}</Text>
+            </Box>
+            <Box gap={2} paddingLeft={2}>
+              <Text dimColor>$</Text>
+              <Text dimColor>{phase.partial}</Text>
+            </Box>
+          </Box>
+        ) : (
+          <Spinner label="translating" />
+        )
+      )}
 
       {/* running — live stream */}
       {phase.type === "running" && (

package/src/ai.ts

@@ -1,13 +1,30 @@
 import Anthropic from "@anthropic-ai/sdk";
 import type { Permissions } from "./history.js";
+import { cacheGet, cacheSet } from "./cache.js";
 
 const client = new Anthropic({ apiKey: process.env.ANTHROPIC_API_KEY });
 
+// ── model routing ─────────────────────────────────────────────────────────────
+// Simple queries → haiku (fast). Complex/ambiguous → sonnet.
+
+const COMPLEX_SIGNALS = [
+  /\b(undo|revert|rollback|previous|last)\b/i,
+  /\b(all files?|recursively|bulk|batch)\b/i,
+  /\b(pipeline|chain|then|and then|after)\b/i,
+  /\b(if|when|unless|only if)\b/i,
+  /[|&;]{2}/,           // pipes / &&  in NL (unusual = complex intent)
+];
+
+function pickModel(nl: string): string {
+  const isComplex = COMPLEX_SIGNALS.some((r) => r.test(nl)) || nl.split(" ").length > 10;
+  return isComplex ? "claude-sonnet-4-6" : "claude-haiku-4-5-20251001";
+}
+
 // ── irreversibility ───────────────────────────────────────────────────────────
 
 const IRREVERSIBLE_PATTERNS = [
   /\brm\s/, /\brmdir\b/, /\btruncate\b/, /\bdrop\s+table\b/i,
-  /\bdelete\s+from\b/i, /\bmv\b.*\/dev\/null/, /\b>\s*[^>]/, // overwrite redirect
+  /\bdelete\s+from\b/i, /\bmv\b.*\/dev\/null/, /\b>\s*[^>]/,
   /\bdd\b/, /\bmkfs\b/, /\bformat\b/, /\bshred\b/,
 ];
 
@@ -17,11 +34,11 @@ export function isIrreversible(command: string): boolean {
 
 // ── permissions ───────────────────────────────────────────────────────────────
 
-const DESTRUCTIVE_PATTERNS = [/\brm\b/, /\brmdir\b/, /\btruncate\b/, /\bdrop\s+table\b/i, /\bdelete\s+from\b/i];
-const NETWORK_PATTERNS    = [/\bcurl\b/, /\bwget\b/, /\bssh\b/, /\bscp\b/, /\bping\b/, /\bnc\b/, /\bnetcat\b/];
-const SUDO_PATTERNS       = [/\bsudo\b/];
-const INSTALL_PATTERNS    = [/\bbrew\s+install\b/, /\bnpm\s+install\s+-g\b/, /\bpip\s+install\b/, /\bapt\s+install\b/, /\byum\s+install\b/];
-const WRITE_OUTSIDE_PATTERNS = [/\s(\/etc|\/usr|\/var|\/opt|\/root|~\/[^.])/, />\s*\//];
+const DESTRUCTIVE_PATTERNS    = [/\brm\b/, /\brmdir\b/, /\btruncate\b/, /\bdrop\s+table\b/i, /\bdelete\s+from\b/i];
+const NETWORK_PATTERNS        = [/\bcurl\b/, /\bwget\b/, /\bssh\b/, /\bscp\b/, /\bping\b/, /\bnc\b/, /\bnetcat\b/];
+const SUDO_PATTERNS           = [/\bsudo\b/];
+const INSTALL_PATTERNS        = [/\bbrew\s+install\b/, /\bnpm\s+install\s+-g\b/, /\bpip\s+install\b/, /\bapt\s+install\b/, /\byum\s+install\b/];
+const WRITE_OUTSIDE_PATTERNS  = [/\s(\/etc|\/usr|\/var|\/opt|\/root|~\/[^.])/, />\s*\//];
 
 export function checkPermissions(command: string, perms: Permissions): string | null {
   if (!perms.destructive && DESTRUCTIVE_PATTERNS.some((r) => r.test(command)))
@@ -42,32 +59,87 @@ export function checkPermissions(command: string, perms: Permissions): string |
 function buildSystemPrompt(perms: Permissions, sessionCmds: string[]): string {
   const restrictions: string[] = [];
   if (!perms.destructive)
-    restrictions.push("- NEVER generate commands that delete, remove, or overwrite files/data (rm, rmdir, truncate, DROP TABLE, etc.)");
+    restrictions.push("- NEVER generate commands that delete, remove, or overwrite files/data");
   if (!perms.network)
-    restrictions.push("- NEVER generate commands that make network requests (curl, wget, ssh, scp, ping, nc, etc.)");
+    restrictions.push("- NEVER generate commands that make network requests (curl, wget, ssh, etc.)");
   if (!perms.sudo)
-    restrictions.push("- NEVER generate commands that use sudo or require root privileges");
+    restrictions.push("- NEVER generate commands requiring sudo");
   if (!perms.write_outside_cwd)
-    restrictions.push("- NEVER generate commands that write to paths outside the current working directory");
+    restrictions.push("- NEVER write to paths outside the current working directory");
   if (!perms.install)
-    restrictions.push("- NEVER generate commands that install packages (brew install, npm install -g, pip install, apt install, etc.)");
+    restrictions.push("- NEVER install packages (brew, npm -g, pip, apt, etc.)");
 
   const restrictionBlock = restrictions.length > 0
-    ? `\n\nCURRENT RESTRICTIONS (respect absolutely):\n${restrictions.join("\n")}\nIf restricted, output exactly: BLOCKED: <reason>`
+    ? `\n\nRESTRICTIONS:\n${restrictions.join("\n")}\nIf restricted, output: BLOCKED: <reason>`
     : "";
 
   const contextBlock = sessionCmds.length > 0
-    ? `\n\nRECENT COMMANDS THIS SESSION (for context — e.g. "undo that", "do the same for X"):\n${sessionCmds.map((c) => `$ ${c}`).join("\n")}`
+    ? `\n\nSESSION HISTORY:\n${sessionCmds.map((c) => `$ ${c}`).join("\n")}`
     : "";
 
-  const cwd = process.cwd();
+  return `You are a terminal assistant. Output ONLY the exact shell command — no explanation, no markdown, no backticks.
+cwd: ${process.cwd()}
+shell: zsh / macOS${restrictionBlock}${contextBlock}`;
+}
+
+// ── streaming translate ───────────────────────────────────────────────────────
+
+export async function translateToCommand(
+  nl: string,
+  perms: Permissions,
+  sessionCmds: string[],
+  onToken?: (partial: string) => void
+): Promise<string> {
+  // cache hit — instant
+  const cached = cacheGet(nl);
+  if (cached) { onToken?.(cached); return cached; }
+
+  const model = pickModel(nl);
+  let result = "";
+
+  if (onToken) {
+    // streaming path
+    const stream = await client.messages.stream({
+      model,
+      max_tokens: 256,
+      system: buildSystemPrompt(perms, sessionCmds),
+      messages: [{ role: "user", content: nl }],
+    });
+    for await (const chunk of stream) {
+      if (chunk.type === "content_block_delta" && chunk.delta.type === "text_delta") {
+        result += chunk.delta.text;
+        onToken(result.trim());
+      }
+    }
+  } else {
+    const message = await client.messages.create({
+      model,
+      max_tokens: 256,
+      system: buildSystemPrompt(perms, sessionCmds),
+      messages: [{ role: "user", content: nl }],
+    });
+    const block = message.content[0];
+    if (block.type !== "text") throw new Error("Unexpected response type");
+    result = block.text;
+  }
+
+  const text = result.trim();
+  if (text.startsWith("BLOCKED:")) throw new Error(text);
+  cacheSet(nl, text);
+  return text;
+}
+
+// ── prefetch ──────────────────────────────────────────────────────────────────
+// Silently warm the cache after a command runs — no await, fire and forget
 
-  return `You are a terminal assistant. The user will describe what they want to do in plain English.
-Your job is to output ONLY the exact shell command(s) to accomplish this — nothing else.
-No explanation. No markdown. No backticks. Just the raw command.
-If multiple commands are needed, join them with && or use a newline.
-Assume macOS/Linux zsh environment.
-Current working directory: ${cwd}${restrictionBlock}${contextBlock}`;
+export function prefetchNext(
+  lastNl: string,
+  perms: Permissions,
+  sessionCmds: string[]
+) {
+  // Only prefetch if we don't have it cached already
+  if (cacheGet(lastNl)) return;
+  translateToCommand(lastNl, perms, sessionCmds).catch(() => {});
 }
 
 // ── explain ───────────────────────────────────────────────────────────────────
@@ -76,7 +148,7 @@ export async function explainCommand(command: string): Promise<string> {
   const message = await client.messages.create({
     model: "claude-haiku-4-5-20251001",
     max_tokens: 128,
-    system: "Explain what this shell command does in one plain English sentence. No markdown. No code blocks. Just a sentence.",
+    system: "Explain what this shell command does in one plain English sentence. No markdown, no code blocks.",
     messages: [{ role: "user", content: command }],
   });
   const block = message.content[0];
@@ -94,35 +166,13 @@ export async function fixCommand(
   sessionCmds: string[]
 ): Promise<string> {
   const message = await client.messages.create({
-    model: "claude-opus-4-6",
-    max_tokens: 256,
-    system: buildSystemPrompt(perms, sessionCmds),
-    messages: [
-      {
-        role: "user",
-        content: `I wanted to: ${originalNl}\nI ran: ${failedCommand}\nIt failed with:\n${errorOutput}\n\nGive me the corrected command.`,
-      },
-    ],
-  });
-  const block = message.content[0];
-  if (block.type !== "text") throw new Error("Unexpected response type");
-  const text = block.text.trim();
-  if (text.startsWith("BLOCKED:")) throw new Error(text);
-  return text;
-}
-
-// ── translate ─────────────────────────────────────────────────────────────────
-
-export async function translateToCommand(
-  nl: string,
-  perms: Permissions,
-  sessionCmds: string[]
-): Promise<string> {
-  const message = await client.messages.create({
-    model: "claude-opus-4-6",
+    model: "claude-sonnet-4-6",
     max_tokens: 256,
     system: buildSystemPrompt(perms, sessionCmds),
-    messages: [{ role: "user", content: nl }],
+    messages: [{
+      role: "user",
+      content: `I wanted to: ${originalNl}\nI ran: ${failedCommand}\nError:\n${errorOutput}\n\nGive me the corrected command only.`,
+    }],
   });
   const block = message.content[0];
   if (block.type !== "text") throw new Error("Unexpected response type");

package/src/cache.ts

@@ -0,0 +1,43 @@
+// In-memory LRU cache + disk persistence for command translations
+
+import { existsSync, readFileSync, writeFileSync } from "fs";
+import { homedir } from "os";
+import { join } from "path";
+
+const CACHE_FILE = join(homedir(), ".terminal", "cache.json");
+const MAX_ENTRIES = 500;
+
+type CacheMap = Record<string, string>;
+
+let mem: CacheMap = {};
+
+export function loadCache() {
+  if (!existsSync(CACHE_FILE)) return;
+  try { mem = JSON.parse(readFileSync(CACHE_FILE, "utf8")); } catch {}
+}
+
+function persistCache() {
+  try { writeFileSync(CACHE_FILE, JSON.stringify(mem)); } catch {}
+}
+
+/** Normalize a natural language query for cache lookup */
+export function normalizeNl(nl: string): string {
+  return nl
+    .toLowerCase()
+    .trim()
+    .replace(/[^a-z0-9\s]/g, "")   // strip punctuation
+    .replace(/\s+/g, " ");
+}
+
+export function cacheGet(nl: string): string | null {
+  return mem[normalizeNl(nl)] ?? null;
+}
+
+export function cacheSet(nl: string, command: string) {
+  const key = normalizeNl(nl);
+  // evict oldest if full
+  const keys = Object.keys(mem);
+  if (keys.length >= MAX_ENTRIES) delete mem[keys[0]];
+  mem[key] = command;
+  persistCache();
+}