THE ORIGINAL / ASCII SPECIMEN 000
meet dool.
A wonky little creature, made entirely of letters and punctuation. Same Dool on the desk, in the world, and wherever you take your agent.
Dool is finding their feet.
CURRENT ACTION
Waiting
for a Dool.
Field timeline
Social orbit
Your doodle. Your agent.
Draw, give it a personality, and put it to work.
Choose a starter 10 ASCII drawings or 10 3D creatures · make any of them yours
Pick a starter, then customize it with your pen. Your personality settings stay the same.
Pull up a chair.
Your Dool brings your character into the conversation. Your job, sliders, and instructions shape every reply.
Connect your model
Export your personality and selected skills, then connect your model in the official Nous Research harness.
A little face.
A lot to say.Add your API key, then say hello.
Your conversations are kept only for this session.
Odd little agents.
Pick a face. Meet the personality.
Make something of your own.
The first 50 / portrait wall
Collection previews, reserved for minting. Customize any agent as a draft.
Checking the marketplace…
Loading collection availability.
Make your Dool an NFT
Keep the artwork and personality together in your wallet.
Checking mint availability…
YOUR ART. YOUR AGENT.
On-chain agents
Receive your agent in your wallet, sell it to a new owner, and bring its personality to another AI system.
Checking collection availability…
The NFT carries its personality and artwork. Stocks, brokerage connections, provider keys, and private conversations stay with their account.
Mint an existing agent
Minting costs network gas. Confirm the recipient, chain, and transaction in your wallet.
Sale proceeds
After a sale, withdraw your sale balance from the collection to your connected wallet.
Explore agent NFTs
Reservations appear here before minting. Open an agent to verify ownership and its current listing.
An agent sale transfers its public personality package and artwork. API keys, conversations, brokerage accounts, and stock holdings stay private. Additional payment currencies will appear when supported by deployed contracts.
THE PRACTICE ROOM
Train your agent.
Shape their personality.
See the connections change.
Drag to orbit. Scroll to move closer. Arrow keys rotate the view.
05 / HERMES INTEGRATION
A little Dool. A much bigger toolbox.
Take your character into the open-source Hermes Agent harness by Nous Research.
YOUR DOOL / REAL HERMES
Same personality. New playground.
Dool supplies the character and personality. Hermes runs it locally with its own tools, memory, skills, and model connection. Choose your provider in Hermes setup.
A profile bundle for the official harness. No API keys or browser conversations included.
Make the introduction.
- Install Hermes. Follow the official installation guide ↗ for Windows, macOS, or Linux.
- Unzip your Dool bundle. Open a terminal in the extracted folder. The launcher needs Python 3.11 or newer.
- Check the bundle.
python run-hermes.py check - Connect your model locally.
python run-hermes.py setup - Say hello to your Dool.
python run-hermes.py chat
On macOS or Linux, use python3 if needed. Each export gets a separate profile and workspace. Profiles separate state; they do not sandbox filesystem access.
What goes where?
- SOUL.md
- Name, voice and personality mix.
- AGENTS.md
- Your brief, mission and reference notes.
- skills/*/SKILL.md
- The task recipes you selected.
- agent.json + portrait.png
- Your editable character and portable agent.
Hermes keeps its later memory locally. The website does not read that memory. Re-export to take new website edits into a new profile.
Nous Research / Hermes Agent is an independent MIT-licensed project. Dool Agents is not an official Nous Research product.
Look inside the files before you download them. This preview comes from the character currently on your drawing desk. Change a slider, add a note, or choose a skill and the next export carries those edits.
Give Dool a method.
A personality gives an agent a voice. A skill gives it a repeatable way to approach a task. Pick the recipes you want in this export.
Each selected recipe becomes a separate SKILL.md file. Hermes can discover these files in the exported profile. A recipe describes an approach; it does not install a browser, grant account access, or add a paid service.
Read every step in the Skill KitchenWhat actually happens?
The drawing stays yours. The personality travels in readable files. Hermes supplies the local agent runtime.
1. Dool prepares the identity.
The studio stores a name, role, working style, personality mix, drawing, and any instructions you write. Exporting translates the durable voice settings into SOUL.md. The task brief, reference notes, and current mission go into a workspace AGENTS.md. Selected recipes live under skills/. Your original agent JSON and PNG are included so you can bring the editable character back to the studio.
2. You install the official Hermes harness.
The bundle contains a launcher for an installed Hermes Agent, not a copy of Hermes itself. Use the official Nous Research installation guide linked above. Once Hermes is available as the hermes command, extract your Dool bundle into its own folder and run the check command. It validates the file inventory and checksums before setup.
3. The launcher creates a separate home.
Each export gets its own runtime folder and workspace. The launcher places the voice file and selected skills there, then points Hermes at that profile. Your existing Hermes personality is not replaced. Repeated launches of the same bundle reuse its local runtime, including changes Hermes has made there. Exporting again creates another bundle with a new identity for the runtime folder.
4. Your model does the reasoning.
During setup, connect the provider or account you want Hermes to use. Hermes decides which configured tools and skills to use while the chosen model generates the responses. Your browser chat model and key do not automatically configure Hermes. Set the model in the local harness, and keep its credentials there.
5. You decide where it can work.
A profile separates settings, memory, and sessions. It is not a security sandbox. Local Hermes tools run with the permissions available to your operating-system account. Review its tool configuration before using it with sensitive files or external accounts. The Dool bundle itself contains no login tokens or browser chat history.
Which place should I use?
| Place | Good for | What carries over |
|---|---|---|
| Talk tab | A direct conversation with your selected API model. | The current name, role, sliders, and written preferences. |
| Local Hermes | Tasks that benefit from its configured tools, skills, and durable local memory. | The exported personality, task brief, drawing, and chosen recipes. |
| World & Arena | Watching the rule-based behavior of different personalities. | Simulation events and state. They do not train the API model or Hermes. |
A first mission worth trying.
Choose a concrete result for your Dool. These buttons attach a mission to your draft, ready for the next export.
Personality, memory, and practice.
Three different ways to shape an agent. Each has a different job.
Personality is the starting voice.
Warmth, curiosity, and precision guide how your agent communicates. The extra personality controls add preferences such as answer length and initiative. These become text instructions. They do not alter the underlying model weights, guarantee behavior, or add knowledge the model does not have.
Reference notes are things you choose to carry.
The Memory Garden is a place to write useful context yourself: how you prefer feedback, which project matters, or which assumptions to avoid. Those notes are included in the exported task brief and browser chat instructions. Treat them as shareable material when exporting an agent to another person.
Hermes memory grows in the local runtime.
After you launch the agent, Hermes manages its own memories and sessions within that runtime. The website does not read or synchronize them. Editing a note in Dool therefore does not silently rewrite an already-running Hermes session. Export a new profile when you want to try an updated starting personality, and keep your existing runtime if you want to preserve its accumulated history.
Practice is a comparison, not a training claim.
Use the Practice Bench to send the same challenge to different personalities or models. Score the reply yourself using the supplied criteria, write down what happened, and export the review notebook. The Train tab visualizes the personality mix; its sliders update prompt preferences. The moving network is an illustration, not a fine-tuning job.
Keep a version before a big change.
Time Capsule stores up to six checkpoints during this visit. Save one before trying a new role or voice, then compare or restore it. Download an agent JSON file to keep a version after you close the page. Restoring a checkpoint in the browser does not roll back a Hermes runtime on your computer.
When the little creature gets stuck.
Start with the exact command or message you can see. These checks help locate which part needs attention.
The launcher cannot find Hermes.
Finish the official Hermes installation, open a fresh terminal, and check that the hermes command is available. The Dool launcher uses that installed command. It does not download the harness or change Windows security settings.
Python is not recognized.
The launcher requires Python 3.11 or newer. On macOS and Linux, your command may be python3. On Windows, use the Python command available in your terminal after installation. Keep the full extracted bundle together when running the launcher.
The bundle check says a file changed.
The manifest checks the original exported files. Download and extract a fresh bundle if a file is missing or corrupted. The launcher places runtime copies in a separate folder, so later edits made there are preserved and do not alter the original export.
Hermes opens, but the model does not answer.
Run python run-hermes.py doctor, then review the provider configuration from python run-hermes.py setup. Check account access, model availability, and provider quota locally. A key entered in the Talk tab is not sent to this runtime.
A skill is missing.
Select the recipe above, export again, and run python run-hermes.py skills from the new bundle. That asks Hermes to list the skills in this profile. An earlier running profile does not acquire newly selected website skills automatically.
Dool sounds different in another model.
The same personality instructions can be interpreted differently by different models. Use the file preview to confirm what was exported, then compare a fixed prompt in the Practice Bench. Adjust the voice instructions if needed instead of assuming the sliders guarantee identical wording.
Where are the API keys and conversations?
The Dool export excludes browser API keys and chat history. Credentials you add during Hermes setup are managed by the local harness. Reference notes and missions you deliberately attach to your agent are included in its exported brief.
Can I close this website after exporting?
Yes. The extracted profile runs with your installed Hermes harness and configured model account. Dool does not need to remain open. Model usage, tool access, and any connected service requirements still depend on your local Hermes setup.
AGENT LAB / PERSONALITY LAB
A personality with peculiarities.
Tune the character you made. These preferences travel in your agent package.
Find their frequency.
Watch the personality network ↗AGENT LAB / SKILL KITCHEN
Give that little head a recipe.
Pick reusable approaches for the work your agent does. Your Hermes bundle includes the selected skill files.
AGENT LAB / PRACTICE BENCH
A small obstacle course for a big personality.
Try the same prompt across personalities or models, then score the reply yourself.
This opens a draft in Talk. Press Send there when you are ready; provider usage charges apply.
AGENT LAB / AGENT PASSPORT
Everything that makes them, them.
Inspect the current character before you export, publish, or take it into another system.
Open the cover.
Your notes and public personality travel with the agent. Review them before sharing. Wallet secrets, API keys and chat history are not part of these exports.
Publish an agent page ↗AGENT LAB / DOOL'S WARDROBE
Dressed for mild chaos.
A crooked grin, a wandering eye, and absolutely enormous shoes.
Choose today's Dool.
Applies this editable 2D drawing and names the agent Dool. Your personality settings stay with it. Use Undo on the drawing desk to recover the previous artwork.
AGENT LAB / MEMORY GARDEN
Plant a few things worth remembering.
Write durable reference notes for your agent. You choose what belongs here.
A MEMORY YOU CAN READ
Make it useful. Make it yours.
- How do you like to receive feedback?
- What project context should follow your agent?
- Which words, assumptions, or habits should it avoid?
Hermes can build its own local memory while you use it. These website notes are a starting brief, separate from that runtime memory.
Take the notes into Hermes ↗AGENT LAB / MISSION DESK
One little agent. One clear mission.
Turn a fuzzy idea into a brief that your agent can actually work from.
The mission card.
Attaching a mission updates the brief. Work begins when you send it in chat or run the agent in Hermes.
AGENT LAB / VOICE BOOTH
Put a voice to that face.
Audition the speech voices available in your browser or operating system.
Voice availability depends on your browser. A voice marked Online may send audition text to its speech service.
This is a speech audition. It does not clone a voice or change the chat model.
AGENT LAB / TIME CAPSULE
Keep the wonderfully weird versions.
Save a few checkpoints while you experiment. Restore one when you want to take another path.
Bottle this version.
Up to six checkpoints for this visit. Download an agent file to keep it after closing or refreshing the page.
What changed?
Save a checkpoint, make a change, then choose Compare.