Short answer: open Manus, click into the task field, press ⌃+⌥+R (Mac) or Ctrl+Alt+R (Windows/Linux), speak the full specification, press again. AICHE inserts cleaned-up instructions in 2-3 seconds. Start the agent.
Manus is autonomous. Once you start it, there's no chat to clarify, no back-and-forth to correct course mid-flight. The agent reads your instructions and runs with them. Vague spec, vague output. Detailed spec, useful output. The single most valuable thing you can do for an autonomous agent is invest in the prompt up front.
That investment is hard to make when typing it costs 10-15 minutes. Voice makes the detailed version a 60-second task.
How It Works
- Open Manus in your browser.
- Start a new task or open an existing workflow.
- Click into the task description.
- Press ⌃+⌥+R (Mac) or Ctrl+Alt+R (Windows/Linux).
- Speak the spec: scope, constraints, structure, output format. No length cap.
- Press the hotkey again. AICHE transcribes, applies AI cleanup, inserts.
- Review the prompt, start the agent.
Where Voice Pays Off in Manus
The Detail Premium
A vague prompt: "Find good project management tools." A useful prompt: "Research the top 5 PM tools used by remote engineering teams of 10-50 people. For each: pricing for a team of 25, async-comm features, GitHub and Linear integrations, real user reviews from the last 6 months. Output a Markdown comparison table with a recommendation, prioritizing async over real-time features."
The detailed version takes 60 seconds to speak and saves you from throwing the result away.
Constraints That Prevent Wrong Turns
Autonomous agents wander. Boundaries pulled forward into the prompt save you the redo. Speak them: "Use only official pricing pages and G2 reviews. Skip anything primarily targeting enterprise teams above 200 people. Output a Markdown table with columns for name, $/seat/month, async features, GitHub integration quality, and your assessment. Recommendation under 200 words." Twenty seconds spoken, ten minutes saved.
Multi-Step Structure
Numbered steps make Manus execute in sequence instead of trying everything at once: "Step 1: find Python HTTP client libs with >5,000 GitHub stars. Step 2: for each, get latest release date, open issue count, async support. Step 3: write a simple GET example for the three most popular. Step 4: compare error handling. Step 5: recommendation memo for a team that needs async and good error messages." Thirty seconds spoken, structured execution.
Output Format Up Front
Specify the format you want: Markdown table, one-page memo, numbered list, JSON. Otherwise Manus guesses, and you'll spend ten minutes on cleanup. One sentence at the end of the prompt: "Output as a Markdown table with these columns. Recommendation under 200 words."
Refinement Without Redoing
After the agent completes, dictate the follow-up: "The table is good but pricing for tool #3 looks outdated. Check the pricing page directly and update. Add a column for free-tier availability." Targeted, specific, ten seconds. Manus adjusts without redoing the whole job.
What You Get
- Unlimited voice notes with AI cleanup - filler words removed, punctuation and paragraph breaks added.
- Auto-categorization - if you keep voice notes outside Manus, AICHE auto-buckets them by topic.
- Custom vocabulary - drop in product names, internal tool names, brand terms.
- System-wide dictation - same hotkey works in Manus, ChatGPT, Claude, your IDE, anywhere.
- Multilingual voice input - speak in your language, transcribe in that language or auto-translate to English.
- Zero-retention audio - audio purged immediately after processing, within 1 second.
Plans start at $3.99/mo (annual) with a 7-day free trial, no credit card. See pricing.
Common Questions
Q: Can I dictate while Manus is mid-run?
A: Yes, in any other app. Manus itself runs autonomously and won't accept input mid-execution; that's the whole point. Use the time to dictate notes or the next agent task.
Q: Will AICHE preserve my numbered-step structure?
A: Yes. Speak "step one... step two..." or "first... then... finally..." and AICHE keeps the structure with auto-paragraph breaks.
Q: My agent prompts reference internal tools and codenames. Spelling?
A: Add them to AICHE's Custom Vocabulary. They'll be spelled correctly across every dictation.
Q: Can I dictate JSON or function-calling specs?
A: Mixed approach works best: dictate the natural-language description, then add JSON syntax. Voice for the prose, keystrokes for the punctuation-heavy parts.
Q: Does the audio get sent to Manus?
A: No. AICHE transcribes the audio (and purges it immediately after processing, within 1 second) before any text reaches Manus. Manus only ever sees the text prompt you submit.
Result: detailed Manus specs in 60 seconds instead of 10-15 minutes. Constraints and output format land in the prompt instead of being remembered after the fact. Agent output you can actually use the first time.
Try it now: open Manus, press your hotkey, and dictate one research task you've been deferring. Include scope, constraints, sources to prioritize, and the output format. Run it.