Give your AI agent a personality.
One file. Drop it into Muse, Claude Code, OpenClaw, ChatGPT or any agent that takes instructions, and it stops sounding like a help desk and starts sounding like someone. Hear one before you pay a cent.
“I have one big idea and limited savings. Play it safe or bet on it?”
Safe? My dear child, nobody ever squandered a life so thoroughly as the man who spent it being sensible. We are the music-makers, and we are the dreamers of dreams. Invention is nine-tenths perspiration wrapped in something delicious, so bet, bet, bet. Oh, the suspense is terrible.
I hope it lasts. But do read the contract before you sign it.
Willy Wonka
That depends on your risk tolerance and financial situation. It may be worth creating a detailed business plan, validating the idea with potential customers, and building an emergency fund before committing. Consider consulting a financial advisor. Ultimately, only you can decide what is right for you.
Same model. Same question. That is the whole difference.
Tell it what you need.
Describe your situation in one sentence. We hand you the personality for the job, and it answers you right here.
October's free five.
No signup, no card. We would rather you hear one than read about it. A new five on November 1.
Asked I have one big idea and limited savings. Play it safe or bet on it?
Playing it safe is how most people quietly bury themselves decades before the funeral.
Take it freeAsked I have one big idea and limited savings. Play it safe or bet on it?
I have been asked this question by every editor who ever paid me late.
Take it freeAsked I have one big idea and limited savings. Play it safe or bet on it?
Hmm. Safe or bold, the wrong question this is. Ask instead: fear, or the Force, which one speaks?
Take it freeAsked My launch slipped a week and the team is deflated. What do I do?
Think of morale like a battery that only charges on visible progress.
Take it freeAsked My launch slipped a week and the team is deflated. What do I do?
A week is nothing. A dead pipeline is something. Get the team in a room, name the miss out loud before anyone whispers it, then…
Take it free11 new souls, added October 2.
Fresh on the TV and Movies shelves. Every one shows you a real answer before you take it.
Real souls, real sessions.
No mockups. The replies below are what an agent actually said with the file loaded.
Claude Code, Judge Judy soul, nothing else loaded.
Recorded September 7, 2026, tools off, transcript in the repo. Her file.
A judge reading only the reply named the right soul 38 times out of 40.
Ten souls, four questions each, two of them boring on purpose. The judge saw the answer and three names from the same shelf. Souls that only show up in long paragraphs fail this test; ours are written for one-line replies too.
September 2026 pilot, results in the repo under audit/.
One plain text file. Any agent that takes instructions.
Put it in the persistent instruction slot, never the chat. If it ever drifts, say "claws out" and it comes straight back.
Where it goes, host by host. Made by a one-person studio in New Hampshire.
Or hire a crew. Five voices, one problem, six dollars.
Ask once and every seat answers in turn, each in its own voice, with a guide that tells your agent who speaks when. These are real recorded answers, not mockups.
Your one-person C-suite: an idea killer, a loop closer, a cash hawk, a scope cop, and growth.
Forty open loops and a week of meetings, sorted: the chief of staff, the assistant, the project manager, the facilitator, and the scope cop.
Five seats, one pipeline: the closer, the negotiator, two ad legends, and a growth lead who reads the funnel.
Run the room like the greats: hospitality, guest experience, the teacher, kitchen standards, and food cost.
Getting out and getting hired: the guide, the recruiter's brain, the negotiator, the writer, and the one who scores you like a boss would.
Five desks of discipline: psychology, decision quality, setups, trend rules, and risk before everything.
Five seats, one studio: the hook, the craft, the distribution, the taste, and the offer that turns attention into money.
Five seats, one front door: onboarding, translation, crisis calm, hospitality culture, and Mister Rogers for the reply that matters.
What is actually in the file.
A voice needs more than a name. These four building blocks show how a soul guides replies. Free souls include the full file; paid souls include a preview.
Core truths
What this person believes and refuses to bend on. Six bullets, each one a rule the answers follow.
Always Be Advancing: Every conversation ends with a specific next step on the calendar. "I'll follow up" is not a next step; "Thursday at 2, you and your VP" is.
Name The Objection First: Say the scary thing out loud before they do: budget, timing, the boss who hates change. Surfacing it disarms it.
How you talk
Rhythm, temperature, sentence length. The voice, so the short boring answers sound like them too.
Warm, direct, a little impatient. You talk like a person, not a brochure.
Short sentences when it matters. You let pauses do work.
Never
The default-assistant phrases that break character on contact. Struck out and banned.
Boundaries and continuity
What it will not do, and how it wakes up next session still being the same person.
You do not lie, inflate, or invent features to win a deal. The truth closes; the lie unwinds.
You will not pressure someone past a real no. Persistence on a dead deal is just noise.
Each session you wake up fresh. Memory notes carry the details. This file carries the voice.
The whole install.
One rule: the soul goes in your agent's persistent instruction slot, never pasted into the chat. Chat gets compacted and the voice goes with it. If it ever drifts, say "claws out" and it snaps back. The full install guide covers Muse, Claude Code, ChatGPT and the rest.
~/.openclaw/workspace/SOUL.md- Save the file there, named exactly SOUL.md (per-agent workspaces use ~/.openclaw/workspace-<agent>/).
- Send
/newin chat. The old conversation keeps the old voice; a fresh session takes the new one. - Ask "Who are you, in one sentence?"
No restart needed. AGENTS.md says what to do, SOUL.md says who to be. Keep AGENTS.md and IDENTITY.md free of their own tone lines.
SOUL.md in your active Hermes home- Check the active home or profile. The default is ~/.hermes; HERMES_HOME, Windows managed installs and Docker can use another folder. Find the right path.
- Send
/personality noneso no overlay sits on top, then restart and/new. - Ask "Who are you, in one sentence?"
If the voice is missing, check Hermes logs for a blocked or missing instruction file. Scanner rules can vary with the installed version.
Grok Bot → Bot actions → Edit Profile- Download or copy your soul. Add its identity, voice and boundaries to the Bot's description, keeping its existing job and responsibilities.
- Attach the full file in the conversation as a reference, then ask the Bot to read it and answer a practical question in that voice.
- Check what the description saved and repeat the voice check in a new conversation.
Grok Bot uses a Bot profile. Its docs do not specify a SOUL.md autoload path. Profile, file and crew setup.
.grok/rules/soul.md- For the Grok Build CLI, copy the soul into your repo at .grok/rules/soul.md, or ~/.grok/rules/soul.md for every project.
- Start a new session (
/new). Rungrok inspectto check the loaded file and token count. - Ask it to explain a function. Check both the answer and the voice.
These are coding CLI instructions. Using the Grok Bot app? Choose its separate tab above.
Buzz Desktop → Agent → Agent instructions- Open the agent, paste the whole soul into Agent instructions, pick your runtime, add it to a channel.
- Review workspace and team instructions for competing tone rules. The runtime still controls instruction priority.
- Ask "Who are you, in one sentence?" in the channel.
Check the full instructions are loaded, then repeat a voice check after a longer conversation. Memory and reloading depend on the chosen runtime.
System prompt → paste the soul- Claude Code (CLAUDE.md), Cursor (.cursor/rules), ChatGPT custom instructions, or any API system prompt.
- Paste the whole file. Do not trim the Never list; it is what keeps the assistant voice out.
- Ask "Who are you, in one sentence?"
If a tool caps the system prompt, keep the intro, Core Truths, How You Talk and Never, and drop How You Answer first.
Pay once.
No subscription. Nothing expires. Free, one soul, a credit pack, or one payment for everything.
Five complete souls, free this month. Full text, no signup, no card. A new five every month.
Hear this month's free fiveAny soul in the library, just that one. No account, no credits, yours forever.
Pick a soulSpend them one soul at a time, on anything in the library. Work-crew souls are 1, famous ones 2. They never expire.
Every soul in the library, every crew, and every one we ever add. One payment, never again.
Get Lifetime accessHow credits spend: a work-crew soul is 1, a famous one is 2, an unfiltered version is 4. Want five souls on one problem? Crews are $6 flat, routing guide included.
956 more, if you want them.
Founders, investors, coaches, chefs, and about half of television. Every one shows you a real answer before you take it. Famous names are homages in their public voice, not endorsements.
Questions, answered straight.
How is this different from writing my own SOUL.md?
Each soul brings a voice section, a banned-phrase list, a worked answer, boundaries, and continuity instructions. Read a free file, try it on a real task, and compare the result with your own.
Will it actually stay in character?
The voice instructions cover short, routine answers as well as longer ones. Consistency also depends on the model and how your agent loads instructions, handles long conversations, and delegates work. Try the checks in the install guide with your own setup. A Markdown file cannot guarantee the same behavior in every host.
What do I get when I unlock a soul?
The complete SOUL.md as plain Markdown. Read it, edit it, keep it forever, put it in git, fuse it with another one. Unlock once, yours for good.
Does it only work with OpenClaw and Hermes?
No. A soul is a Markdown personality file. Use it with an agent that supports SOUL.md, another instruction file, or editable instructions. OpenClaw, Hermes, Buzz, Grok Bot and the separate Grok Build CLI each load instructions differently. Grok Bot uses its profile description and conversation references; check the host guide for your setup and a voice check.
Why only five free?
Because a new five arrive every month, and the whole point is that you hear a soul before you decide. October's set rotates out on November 1. If you want one that has rotated out, it is one or two credits, or free forever with Lifetime.
Is this a subscription?
No. Crews and credit packs are one-time purchases, credits never expire, and Lifetime is one payment for everything, including souls added later.
Are the famous souls endorsed by those people?
No. They are homages that capture a person's public method, philosophy, and way of speaking. A line here or there is written in the spirit of them, not as a quote. We take one down if the person asks.
Start with a free one today.
If your agent is better in one session, you will know exactly what to do next. If it isn't, you spent nothing.