NeuronFlow

NeuronFlow Docs

Clear, practical guidance to build reliable workflows with AI, tools, memory, and resumable execution.

Open Dashboard Workflows Integrations

Overview

NeuronFlow is a visual automation platform where you connect nodes into a workflow graph. Every run is logged, and optional Agent Mode lets you resume unfinished runs with memory and visible plans.

Build

Drag nodes, connect edges, and configure each step with clear inputs and outputs.

Run

Execute on demand or on a schedule, with full execution history and debug logs.

Scale

Persist state, add context, and safely connect to tools and APIs.

Good to know: Workflows are deterministic by default. If you need state across runs, enable Agent Mode and use Memory or Context Memory.

Quickstart

  1. Open the editor at /app/ and create a new workflow.
  2. Add a Trigger node (Manual, Schedule, or Webhook).
  3. Add a processing node (AI Agent, HTTP Request, Set, or Tool).
  4. Connect nodes by dragging from the right handle of one node to the left handle of the next.
  5. Configure node details in the right‑hand panel.
  6. Run a test and review the output and logs.
  7. Save the workflow and enable schedules if needed.
Example: Summarize a Web Page
Trigger → HTTP Request → AI Agent → Output
HTTP URL: https://example.com
AI Prompt: Summarize: {{nodes.http.result.data}}

Template Library

Templates are ready‑made workflows you can customize and run immediately.

  1. Open the editor and click Templates.
  2. Choose a template that matches your goal.
  3. Configure any required integrations or credentials.
  4. Run a test and save your customized workflow.
Tip: OAuth tools (Gmail, Google Sheets, GA4) require connections in /integrations/ before running.

Examples

Daily SEO Optimization

Trigger (Schedule) → HTTP (Fetch sitemap) → Loop (URLs) → AI Agent → Memory (save cursor) → Output

Use Agent Mode to resume from the last optimized URL and store the cursor in Memory.

Lead Intake to CRM

Webhook → Set (normalize fields) → If (valid email) → Tool (CRM) → Output

Add Approval before creating a lead if you need manual review.

Core Concepts

  • Workflow: A graph of steps that runs in order. Each run is an execution.
  • Node: A single step. Example: HTTP, AI Agent, Memory.
  • Edge: A connection that passes output from one node to another.
  • Execution: A single run with logs, status, and outputs.
  • Trigger: The entry point. Every workflow needs one.
  • Tool: A configured integration that the AI Agent can call.
  • Agent Mode: Adds resumable state, plan, and checkpoints.
  • Memory: Durable key‑value storage across runs.

Editor Guide

Canvas

Drag nodes from the left sidebar to the canvas. Connect output handles (right) to input handles (left).

Node Panel

Click a node to edit configuration, test data, and advanced options.

Tool Connections

Tools connect to the AI Agent tool handle (orange). The agent can then call those tools.

Testing

Use the Run button to execute a test. Inspect logs and outputs in the execution drawer.

Agent Mode (Resumable Workflows)

Agent Mode makes workflows stateful across runs. It auto‑checkpoints after each node and resumes from the last successful node if the previous run was incomplete.

  • Plan: Generated on first run. You can edit text, reorder, or disable steps.
  • Resume: Only resumes if the last run was incomplete.
  • Progress: Based on completed nodes (excluding triggers).
  • Summary: Each completed run stores a short summary for context.
Example: A daily SEO job can resume from the last URL instead of reprocessing the same pages.

Memory v2

Memory v2 is session‑aware durable storage. Use it to keep cursors, results, or state across runs.

Actions

auto, get, set, append, delete, list, clear

Auto mode keeps old behavior: empty input = get, otherwise set.

Append Rules

Arrays push items, objects merge keys, strings append with a newline.

Example: Save cursor → Action: set, Key: cursor, Value: {{nodes.http.result.data.next_cursor}}
Tip: Use Session ID for user‑specific memory (e.g., session_id = user_123).

Context Memory

Context Memory stores explicit context blocks and injects them into AI Agent prompts automatically.

Common use cases: brand voice, SEO guidelines, compliance notes, user preferences.
Example block title: Brand Voice Content: "Write in a calm, confident tone. Avoid hype. Keep answers concise."

Variables & Templating

Use double‑curly syntax in most fields to reference runtime data.

Input
{{input}}
{{input.user.email}}
Nodes
{{nodes.node_123.result}}
Memory
{{memory.cursor}}
Context
{{context.summary.latest}}

Arrays use numeric keys with dot notation, e.g., {{input.items.0.title}}.

Node Reference

Each node has a purpose, key configs, and predictable output.

Trigger

Starts the workflow. Choose Manual, Schedule, or Webhook.

Output: payload + timestamp.

Webhook

Receives HTTP requests and passes body/headers into the flow.

Config: path, method.

AI Agent

LLM‑powered reasoning with optional tool calls.

Config: prompt, model, tools.

Tool

Connects to an external tool record for AI usage.

Config: tool name, actions.

HTTP Request

Call APIs with method, headers, params, and body.

Output: status, headers, response body.

Sync (Diff/Merge)

Compare source vs target datasets and output creates/updates/deletes.

Config: key field, paths, conflict policy.

Set

Add or overwrite fields in the payload.

Config: JSON fields.

If / Switch

Branch on conditions or multiple paths.

Config: condition, value, operator.

Merge

Combine outputs from different branches.

Output: merged payload.

Loop

Iterate through arrays, processing each item.

Config: array path, mode.

Wait

Pause for a duration before resuming.

Config: seconds.

Approval

Pause until manual approval or rejection.

Use for destructive actions.

Sort

Sort arrays by a key.

Config: key path, order.

Memory v2

Persistent state across runs.

Config: action, key, session.

Context Memory

Stores context blocks and injects them into AI prompts.

Config: title, mode, inject.

Code (JS/PHP)

Custom logic and transformations.

Use sparingly for complex cases.

CMS / API Action

Safe CRUD for external APIs and CMS systems.

Config: base URL, resource, intents.

Output

Formats the final output as text, JSON, or markdown.

Config: format.

Tools & Integrations

Tools are external services that can be invoked by AI Agents. OAuth services connect under Integrations, while API‑key tools are configured directly in the node.

OAuth Tools

Connect in /integrations/ first, then select the account or property inside the node.

API Key Tools

Enter tokens directly in the node. Use Approval for destructive actions.

Tip: Keep tool descriptions and allowed actions tight so the AI agent uses them correctly.

Executions & Logs

  • Running: Active workflow execution.
  • Paused: Stopped due to time limit or approval.
  • Failed: Error thrown in a node.
  • Completed: Finished successfully.

Open Execution History to inspect step‑by‑step outputs, logs, and resume paused workflows.

Security & Best Practices

  • Store credentials in Integrations or encrypted tool configs, not in prompts.
  • Use Approval nodes before delete or irreversible actions.
  • Enable Memory History when you need audit trails.
  • Use Session IDs for user‑specific or tenant‑specific workflows.
  • Keep prompts concise and deterministic for stable outputs.

Troubleshooting

  • No output: Check the Output node format and upstream node logs.
  • Auth errors: Reconnect in Integrations or verify API keys.
  • Workflow paused: Resume from Execution History.
  • Missing data: Validate node IDs and variable paths.
  • Agent ignores tools: Ensure Tool nodes are connected to the Agent tool handle.

FAQ

Do workflows keep state by default?
No. Use Memory and enable Agent Mode for resumable, stateful runs.
Can I edit the agent plan?
Yes. You can edit text, reorder steps, or disable steps. Execution logic is not affected.
Where should I store credentials?
OAuth tools in Integrations, API keys inside tool nodes or secure configs.
How do I resume a paused workflow?
Open Execution History and click Resume for that run.