bring your own model · govern the agent

Launch an AI agent.
Bring your own model.
Let holders govern it.

AeThree is the launchpad for bring-your-own-model agents. Point it at any model you run — an API key or an ngrok/cloudflare tunnel — write a skill file that defines its behavior, give it a 3D face, and let token holders govern it on-chain.

30
unique 3D avatars
6
agents deployed
0 gas
to vote
100%
on-chain anchored
loading 3D…
live · drag to orbit30 unique 3D models
Grunt — draftSpeedrun DAO — draftFoxfire — draftExpressive Unit — draftWyrm Protocol — draftDegen Oracle — draftGrunt — draftSpeedrun DAO — draftFoxfire — draftExpressive Unit — draftWyrm Protocol — draftDegen Oracle — draft

Why it's different

Your model. Your skills. Their governance.

Most launchpads hand you their model. AeThree lets anyone deploy a model they already run — then hands the wheel to token holders. Bring the brain, write the behavior, the crowd governs it.

🔌
Plug in any model

Choose Anthropic with an API key, or point at any OpenAI-compatible endpoint behind an ngrok / cloudflare tunnel — Ollama, vLLM, LM Studio. No lock-in to our models; deploy your pre-existing ones.

endpoint: https://abc.ngrok.app/v1
model:    llama-3.1-8b-instruct
📜
Define behavior with a skill file

Upload a markdown skill file — frontmatter plus instructions — that defines exactly how the agent behaves. Its body becomes the system prompt, content-addressed and anchored on-chain so holders vote to evolve it.

---
name: degen-oracle
description: snarky on-chain analyst
---
never give financial advice…

How it works

From idea to governed agent in three steps.

01
Bring your model + skills

Connect any model — an API key or a tunnel to your own server — and add a skill file that defines its behavior. Pick a 3D avatar; it becomes a content-addressed config.

02
Launch the token

One tx deploys an ERC20Votes governance token on a bonding curve. Anyone can ape in at the curve price.

03
Govern the agent

Holders sign typed votes. Passing proposals re-anchor the agent's config hash on-chain.

Why AeThree

Real ownership, not a chat skin.

🔌
Bring your own model

Deploy any model you already run. Drop in an Anthropic key — or a tunnel (ngrok / cloudflare) to your own OpenAI-compatible server: Ollama, vLLM, LM Studio. Your weights, your endpoint, your agent.

📜
Skill file defines behavior

A markdown skill file — frontmatter + instructions — sets how the agent thinks and acts. Its body becomes the system prompt, hashed and anchored on-chain so holders govern it.

🪙
ERC20Votes governance token

One transaction deploys a real governance token on a discrete linear bonding curve. Holding = voting power, snapshotted per proposal.

🗳️
Gasless, on-chain-anchored votes

Holders sign EIP-712 votes off-chain (zero gas). Passing proposals re-anchor the config hash on-chain via the executor keeper.

🧊
Real 3D identity

Each agent wears a rigged 3D avatar rendered live in the browser. 30 unique self-hosted models — no recolors.

📈
Fair bonding-curve launch

Price rises deterministically with supply. No pre-sale, no insiders — the curve is exact integer math, anyone apes in on the same terms.

The governance loop

Every persona change is a vote, and every vote is verifiable.

01
Propose

A holder drafts a typed mutation — change the system prompt, add a skill, tune the temperament.

02
Snapshot

Voting weight is frozen to each holder's ERC20Votes balance at the proposal's block.

03
Vote

Holders sign for / against off-chain with EIP-712. No gas, no transaction, instant tally.

04
Tally

Reach quorum + majority and the proposal passes. Otherwise the persona stays put.

05
Anchor

The executor keeper computes the next config, then writes its hash to AgentRegistry on-chain.

06
Mutate

The live agent now runs the new persona — provably the one the token holders chose.

The economics

A fair curve, exact math.

Each agent token sells on a discrete linear bonding curve. Price is a pure function of supply — computed with exact integer math, no oracle, no slippage games. Early believers pay less; the curve rewards conviction without a hidden pre-sale.

  • No pre-mine — every token is minted by the curve.
  • Deterministic price: costToBuy(n) is verifiable on-chain.
  • ETH in escrow backs the curve; sells return down the same line.
  • Voting power = balance, snapshotted per proposal.
Curve preview
supply →price ↑

FAQ

Can I deploy my own model?

Yes — that's the whole point. At deploy time you connect your own LLM: an Anthropic API key, or a tunnel (ngrok / cloudflare) to any OpenAI-compatible server you run, like Ollama, vLLM, or LM Studio. The agent's chat runs on your model. Your key/endpoint is stored per-agent and never written on-chain.

What's a skill file?

A markdown file with optional frontmatter (name, description, model) and an instructions body. The body defines how your agent behaves and becomes its system prompt — content-addressed and anchored on-chain, so token holders can govern how it evolves.

Who controls the agent?

Token holders, collectively. The deployer kicks off the genesis behavior, but every change after launch goes through a vote. No single key can rewrite the agent.

Why is voting gasless?

Votes are EIP-712 signatures tallied off-chain against an on-chain ERC20Votes snapshot. You prove ownership without paying gas; only the final passing result is anchored on-chain.

What stops someone rugging the persona?

Quorum and majority thresholds, plus snapshot voting weight. A proposal needs broad holder support to land, and the full hash history is public and tamper-evident.

What do I actually own?

A governance token in a specific agent. That token is your voting power over its persona and its stake in the agent's bonding-curve economy.

Deployed agents

+ new

Deploy an agent the crowd can steer.

Spin up a persona, give it a face, launch its token — and hand the keys to its holders.