Skip to content

Non-custodial security model

txcast is a coordinator, not a custodian. It never needs your private key or seed phrase.

  • compiles reusable TypeScript workflow intent into immutable artifacts;
  • enforces frozen capability envelopes at effect registration;
  • freezes exact transaction targets, values, calldata, ABI provenance, and evidence;
  • checks current permissions, account grants, and lifecycle state;
  • prepares Safe typed data or EOA transaction requests;
  • accepts externally produced signatures or transaction hashes; and
  • verifies Safe and onchain outcomes before advancing dependent work.
  • displays the final signing request;
  • protects private keys;
  • signs and broadcasts transactions; and
  • enforces Safe owners, threshold, and nonce rules.

Before approving any wallet prompt, verify:

  1. the connected chain and account;
  2. every destination address;
  3. native-token value and token amounts;
  4. decoded function and arguments;
  5. the Safe nonce and proposal hash, when applicable; and
  6. whether the action is expected at this point in the run.

Authored code has no direct secrets, database, networking, filesystem, clock, or randomness. External work crosses typed effect boundaries that txcast validates and records. AI-agent confirmation prompts improve usability but are not an authorization boundary. Credential scopes, controlled-account grants, live role checks, capability enforcement, cryptographic verification, and wallet confirmation are enforced boundaries.

Do not sign. Close the wallet prompt, cancel the run if possible, and start a corrected run. A cancellation cannot reverse a transaction that has already been broadcast.

For a compromised agent credential, open Profile → API keys and revoke the whole credential. An organisation owner/admin can remove only their organisation under Organisation settings → Agent access. For a compromised wallet, follow the wallet or Safe provider’s recovery procedure immediately.