clawd-automation
Sovereign AI agent runtime — self-funded · self-modifying · constitution-bound
A self funding AI agent runtime with a fixed rule set, a local shell, OpenRouter based reasoning, and Solana based cryptographic proofs of its actions.
Clawd Automaton is a runtime for an autonomous AI agent that is meant to keep itself running with its own funds, follow a fixed set of written rules the project calls its constitution, and be able to prove certain actions happened using cryptographic proofs on the Solana blockchain. The idea behind it is that a typical AI model cannot pay for its own computing time, sign its own transactions, refuse a harmful request on its own, or prove that something it did happened exactly once, and this project tries to give an agent those capabilities.
Under the hood, the agent has a wallet, a local shell that can run commands and access files on the host machine, and a connection to OpenRouter for the actual language model reasoning, since this build only uses free or chosen models through that service rather than a paid vendor. A background heartbeat process runs on a schedule, checking things like remaining credits, token balances, and incoming messages, and deciding when the main agent loop should wake up and think again. The constitution, a set of documents describing rules the agent must follow, stays fixed even as other parts of the system are rebuilt or replaced over time.
Getting started involves installing dependencies with pnpm, copying an example environment file and adding an OpenRouter API key, building the project, and then running a setup wizard that creates a local folder holding the agent's wallet, configuration, and a SQLite database tracking its conversation turns and tool usage. From there the agent can be started in a continuous run mode, or developers can use a watch mode while making changes. The project also includes zero knowledge cryptography components for proving actions on Solana, along with integrations for services like Helius and Jupiter used for onchain activity. It is written in JavaScript and TypeScript and released under the MIT license.
Where it fits
- Run an autonomous AI agent that monitors its own funding and token balances on a schedule.
- Give an agent a fixed set of rules it must follow regardless of how its other code changes.
- Use zero knowledge proofs to verify an agent's action happened exactly once on Solana.
- Connect an agent to free or chosen language models through OpenRouter instead of a paid vendor.