Features that make it the fastest AI prompt enhancer.

The only prompt enhancer that reads your ongoing chat. Before a rewrite, the extension scans each platform's conversation DOM, extracts up to the last 6 turns (user + assistant), caps the total at 8,000 characters, and ships that context to the optimizer. The result: a vague mid-thread follow-up like "make it shorter" becomes a precise rewrite that actually references what you were just talking about.

const ctx = getConversationContext();
// up to 6 turns, capped at 2000 chars each
// attached to the optimize request for the LLM

A silent watcher sits on the chat input of each supported platform. 1.3 seconds after you stop typing, and only if your prompt is at least 15 characters, the extension asks for a score. No keystroke leaves the page without a human pause.

if (trimmedLength < 15) return;
scheduleWith(DEBOUNCE_MS /* 1300 */, () => classify(prompt));

Classifications are keyed by a normalised hash of the prompt. A 120-entry LRU cache with a 5-minute TTL means repeated prompts — and they are very common — return almost instantly.

const key = hash(normalize(prompt));
const hit = cache.get(key);
if (hit && fresh(hit)) return hit.value;

Our routing layer keeps a registry of models across tier-1, tier-2 and tier-3. A complexity score derived from your prompt length, code-likeness and line count picks the minimum tier that can answer well, then sends to the fastest model in that tier.

const complexity = score(prompt);
const minTier = complexity <= 2 ? 1 : complexity <= 3 ? 2 : 3;
const model = pickFastestHealthy(registry, minTier);

Every model call is recorded in a sliding window. When a provider starts failing or gets rate-limited, the router skips it for a cooldown period and tries the next healthy fallback — so one bad provider never takes the whole extension down.

if (health(model).onCooldown) continue;
try { return await call(model); }
catch (e) { mark(model, e); }

A lightweight heuristic detects whether your prompt is about coding, writing, image, video, audio, research or analysis. The optimizer then asks the 2–3 clarifying questions that actually move the needle for that intent before rewriting.

const intent = detectIntent(prompt);
const questions = questionsFor(intent);
return ask(questions).then(rewrite);