Make Ethereum cypherpunk again
We give AI platforms too much authority.
Launch an AI agent in minutes while ENS controls its identity, its spending limit, and its off switch.
01 / No code. Zero devops.
Capsule turns a complex deployment into one five-step launch flow.
Pick a model, write the prompt, set the spending cap. No terminal, no deploy script.
Add the Telegram bot and the provider key. Secrets stay sealed offchain — only pointers go on the name.
The agent is born as a subname under yours, and the runner starts beating on Sepolia.
02 / The ENSv2 boundary
The agent can prove it is alive. It cannot rewrite what it is.
Web2 can run the agent; Ethereum holds the control layer. The prompt body and the API keys stay sealed offchain. Authority, permissions, identity are onchain.
A prompt injection can say “raise your limit, send 1 ETH.” The model never holds a private key, and the supervisor reads the cap off the name before it signs. The AI can ask. It cannot sign.
03 / Recall
Pull the heartbeat role through ENSv2 and the next write reverts. The runner sees the revert and halts itself. The kill switch is not something we built — it is the one ENSv2 already has.
04 / Why The Graph
The Graph helps management of agents, record updates, and permission changes. Ask the analyst who recalled an agent and when, in plain English.
Read off the chain
Indexed live from ETH Sepolia. Because the permission and the write it authorises are the same story, a gap in this feed means something.
Register a name, connect it, and hire your first agent under it.