B3Forge

B3Forge is a visual workflow editor and a browser execution engine. You compose canister calls into a typed workflow in a node graph, and the workflow runs in your browser with a delegated identity: no backend executes it and no data leaves your device. It is in beta on the Internet Computer.

Open B3Forge

B3Forge for EVM is in development

A version for EVM contracts, where any contract with a verified ABI can be a step and each step that writes is clear-signed before you confirm it, is in development. These docs describe the beta on the Internet Computer.

Nodes

A workflow is a graph of three kinds of node:

• Variables define typed input values where the workflow starts, and compose values from literals and earlier outputs.
• CanisterCall calls a canister method, query or update. It follows one edge when the call succeeds and another when the method returns a top-level Err.
• Utility runs arithmetic, guards and list operations in the browser, with no network call: add, sub, mul, div, scale_bps, rescale between token decimals, and guards such as guard_gte that branch when a condition fails.

The node inspector offers each field only the upstream values whose type fits it. A Rust compatibility engine, compiled to WebAssembly, enforces the same rule.

Expressions

A node’s arguments are Candid expressions that refer to earlier nodes’ outputs:

(record {
  owner  = $N0.recipient_account.owner;
  amount = $N1.Ok.amount : nat;
  memo   = opt $N2.memo
})

• $N0 is the first node’s return value.
• .Ok and .Err follow a variant’s branches.
• : nat is an explicit, checked cast.
• opt wraps a value in an option.

The browser resolves the expressions, encodes the call’s arguments and decodes its results with the same Rust semantics as the on-chain runtime.

Sharing

Workflows are stored and shared through the beta’s backend canister: publish a workflow to the community catalog, and others can fork it into a private copy. The backend stores, shares and indexes workflows; it never runs them.

Limits of the beta

• Variables nodes compose values but do not compute: arithmetic and comparisons belong in Utility nodes.
• A call can branch only on a top-level Err result.
• Every node is declared up front, and a call that fails at the network level stops the workflow, with no retry.