Skip to content
Logo

Start here

Bloom's current wallet surface, walletFS, is a crypto wallet your AI agent can actually use without ever seeing your private keys.

Instead of giving an agent a pile of Web3 tools or, worse, a seed phrase, Bloom turns wallet and chain interactions into ordinary files and folders. Agents already know how to read, inspect, and write files. Bloom uses that strength as the interface.

  • Reading a file asks the blockchain a question.
  • Writing a file stages a transaction intent. It does not send money.
  • Approving a plan is the step that can actually sign and broadcast.

Your keys are not files in the Bloom filesystem. They live in an encrypted keystore outside the agent's sandbox. The agent can draft work, explain it, and prepare a plan, but signing only happens after the plan passes policy and gets approval.

Who these docs are for

Start here if you use Claude Code, Codex, Hermes, or another coding agent and want it to interact with crypto safely: check balances, stage transfers, inspect positions, or prepare protocol actions without pasting secrets into a chat window.

If you are integrating Bloom into an app or building petals, this first section gives the product model. The later sections go deeper:

The whole model

Your agent proposes. Bloom checks. You approve. Bloom signs.

That is the core safety loop. A prompt-injected or confused agent can stage a bad plan, but it should not be able to steal a private key because the key was never in the agent's readable world.