NodeFox
The execution platform

Build the process.
Run the whole workflow.

Use the visual graph to build processes that read information, check business rules and prepare updates across your systems. Run each case, inspect the results and reuse the workflow.

NODEFOX / VISUAL GRAPHInvoice example

Invoice INV-2048 · $4,320

240 units · purchase order + receipt attached

  1. Extract

    AI reads

  2. Check

    Code validates

  3. Authorize

    Human decides

Approval releases the AP output

Ready for AP · INV-2048

Supported amount
$4,104
Held for review
$216
Illustrative workflow. Payment separately authorized.
01 / Build and execute

Build the process in the visual graph.

Choose the inputs, checks and actions. Connect the information each step needs and the conditions that let it run. The same saved graph defines the process and runs it—so the logic you review is the logic the runtime follows.

Here, AI reads an invoice, Readers gather the reference records, and code checks the quantities. Data dependencies hold the check until its inputs arrive. A separate approval condition controls the Writer. Change the inputs, checks and actions to build a different process.

Illustrative network using NodeFox node styling. Connections, credentials and approval paths are configured for each implementation.

The same building blocks. Different operational work.

Illustrative process patterns

Client onboarding
  1. 01Read the packet
  2. 02Check required fields
  3. 03Resolve missing information
  4. 04Prepare account setup

A validated handoff to finance and delivery.

Operational investigation
  1. 01Gather linked records
  2. 02Reconcile the evidence
  3. 03Review the recommendation
  4. 04Produce the report

A supported decision with its sources attached.

What can you add to the visual graph?

Bring information in and results out

Reader · Writer

Read files or configured API responses. Produce a file or send an authorized result to a configured endpoint.

Interpret, calculate and structure

Conversation · Code · Data

Give models a defined task, use JavaScript for exact rules, and pass structured values between steps.

Control the next step

Decision · Wait · Buffer · Global

Choose a branch, request a human answer, wait on conditions and manage values needed across the run.

Reuse a working process

Network · Apps

Call a sub-network through mapped inputs and outputs. Give operators an App with inputs and results, without requiring them to edit the graph. Apps are an interface, not a node type.

Execution controls: branches, state, synchronization and reuse

Decision rules select a route, with a default branch and cycle limit where configured. Buffer nodes retain or combine data across cycles; Wait nodes hold execution until their conditions pass.

Network nodes call reusable sub-networks through mapped inputs and outputs. AI-assisted construction can help compose a graph; validate its behavior before operating it.

02 / People and decisions

Give people the facts when a decision is needed.

An App gives operators a form and a result without requiring them to edit the graph. When judgment is needed, the workflow presents the context and waits for the person responsible.

For this invoice, AI extracts the fields and code identifies the quantity difference. AP reviews the evidence and chooses the next path. A held or rejected case does not release the configured write.

Run one case—or a batch

Run a registered CSV column, list or numeric range through the same network. App creation and batches require Automate-tier access. Browser runs need an open session; compatible headless runs can continue unattended.

Illustrative human review · INV-2048

12 units are not on the receipt.

Invoice
240 × $18 = $4,320
Goods received
228 × $18 = $4,104
Proposed action
Prepare $4,104; hold $216

AP approves the supported amount

Source invoice, purchase order and receipt accompany the decision. Payment release remains a separate authorization.

03 / Produce

Finished work your team can use.

Raw input → usable output

Stop assembling the same evidence by hand.

AP receives the supported amount and the held balance together. Staff can review the difference without rebuilding the invoice, order and receipt trail.

Input 1
INV-2048 · Northstar Supply
Input 2
PO-7301 · 240 units at $18
Input 3
Invoice total · $4,320
Input 4
Received · 228 units
ILLUSTRATIVE OUTPUTFrom invoice to approved record

AP-2048 · review completed

Supported amount

$4,104

  • Held amount: $216
  • Payment release: separate authorization

Decision attached

Approve $4,104 for AP preparation

Source references and configured checks accompany the result.

Inspect the structured record
{
  "invoice_id": "INV-2048",
  "currency": "USD",
  "supported_amount": 4104,
  "held_amount": 216,
  "payment_authorized": false
}

Synthetic sample files with inputs, checks and decisions. System updates require configured access and authorization. Documents and reports depend on the chosen Writer format and template.

04 / Connect

Keep your systems. Connect the work between them.

Reader and Writer steps bring files and APIs into the graph. Code handles mappings and specialized requests. Configure access for the particular records and actions the process needs.

Models and tools have a defined role

Choose a provider, model, instructions and output schema for each AI step, using your own credentials. Compatible MCP servers expose configured tools for model-driven tasks. AI responses can vary; the surrounding checks and execution conditions remain explicit.

Provider features, billing and limits vary. Browser MCP connections and OAuth-backed services are not automatically available in the headless runner.

Agree the boundary before the write

Identify the source of truth, permitted fields, credential scope and failure path. Test updates against a safe destination and define duplicate handling before enabling operational writes.

05 / Inspect

See how the result was reached.

See the real editor, execution log and completed review memo. Then explore how recorded inputs and decisions explain the illustrative invoice result.

Actual product recording

Model output moves into code and routing.

The actual visual graph and log show the contract-review network executing. This is the same software shown in the recording, not a simulated workflow film.

Model output moves into code and routing. Still from the NodeFox contract-review recording at 7:10.
Captured software and sample contract · cropped for readability · select the image for full size. Model names and UI reflect the recording.
Explore the illustrative invoice run record
Illustrative recorded-state inspection · sample invoice INV-2048 · no external actions

Incoming invoice

Recorded input

INV-2048 from the configured intake

Recorded output

  • INV-2048 · Northstar Supply
  • PO-7301 · 240 units at $18
  • Invoice total · $4,320
  • Received · 228 units

The input record anchors the rest of this case.

Replay projects a captured artifact. Re-execution can call models and systems again and needs its own authorization, duplicate handling and failure policy. Recorded evidence depends on what the run captured.

06 / Deploy and extend

From a working graph to recurring operations.

The Rust execution engine powers browser-local WebAssembly and supported headless execution. Choose the environment around the systems, operators and operating schedule your workflow needs.

Build and interact in the browser

Author, test and run workflows with operators present. Models and connected services still use their configured endpoints. Some file and network access requires the local gateway.

Run unattended on your machine

Use the headless runner on supported Linux or macOS machines with a schedule, watched folder or webhook. Headless execution is included with Pro and Enterprise. Confirm runner access and connection compatibility for the deployment.

Production ownership, costs and change control

Name the workflow owner and technical operator. Agree monitoring, failure handling, credential maintenance and change review. Concurrency depends on the plan and connected providers; reported model costs are estimates, with provider charges billed separately. Reuse tested connections and sub-networks as adjacent processes move onto NodeFox.

What would your first workflow execute?

Bring the input, the required result and the systems in between. We’ll discuss what can run automatically, where a person should decide and what a realistic implementation would require.