Platform · AI Workflows & Agents
Agents and workflows your ops team can actually run
A browser builder where AI takes the judgment steps — reading, classifying, deciding, drafting — inside workflows with triggers, approvals and 700+ integrations. When a step needs a system without an API, a bot does it.
Updated for the October 2026 release.
Agent Builder · invoice-agent.flow
Trigger — every weekday, 7:00
or webhook · or chat message
Read AP inbox
Outlook connector · unread with attachments
AI step — extract & decide
vendor, lines, totals → match against open POs
Human approval if > $5,000
Slack approval to controller · else continue
Post to QuickBooks
create bill · code GL · attach source
In one paragraph
ElectroNeek AI Workflows is a no-code, browser-based workflow builder with 700+ app integrations and an agent step that uses tools: app actions, external MCP servers, other flows and ElectroNeek RPA bots. Workflows start from a schedule, a webhook, an app event, a form, an email or a chat message; approval steps pause the flow in Slack, Teams or email; every run keeps step inputs and outputs and an agent timeline for 30 days. You bring your own model — OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter or a self-hosted endpoint. Billing is per workflow run, never per step or per token.
What it does — concretely
Triggers where work starts
Schedules, webhooks, polled or real-time app events, public forms, inbound email, SFTP drops, chat and an MCP tool trigger that lets other agents call your flow.
Agents with tools
The Run Agent step takes a goal, a model and tools: app actions, flows, external MCP servers and RPA bots. Max steps, structured output and a visible reasoning timeline per run.
Human in the loop as a step
Approval actions for Slack, Teams, Gmail and Outlook; forms and tasks with assignees; a Human Input step for chat. The flow waits, then continues with an audit entry.
700+ integrations and the gaps
Microsoft 365, Google Workspace, Salesforce, HubSpot, QuickBooks, Xero, NetSuite, SAP Ariba, Jira, Zendesk, Slack, WhatsApp Business, Postgres, Snowflake — most with a custom API call for anything missing. No PSA or RMM pieces yet (see limits).
Bring your own model
OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Cloudflare AI Gateway or any OpenAI-compatible endpoint, including self-hosted. Keys are stored server-side, never shown again, masked in logs.
Versions, retries, folders
Draft and published versions with rollback, per-step retry (up to four attempts) and continue-on-failure, JSON import and export, folders, a key-value store and tables up to 10,000 rows.
How it works
01
Pick a trigger and a template
Start from the Explore gallery or a blank canvas. Connect the apps once; connections are reusable across flows and show their health.
02
Add the judgment, then the hands
An agent or an Ask AI / Classify / Extract step for the decisions; app actions for systems with APIs; Run RPA Workflow for SAP GUI, desktop apps and portals without one.
03
Run supervised, then widen
Test a single step or the whole flow on real data. Start with approvals on everything consequential; relax them as the run history earns it.
What it doesn’t do (yet)
Stated here so you don’t find out in a pilot. Roadmap items are marked as such.
- —Run data is retained for 30 days today (extending retention to one year is on the roadmap); a flow paused on an approval cannot resume after that.
- —Access control in AI Workflows is team-wide today; per-client workspaces are in the current roadmap (Partner Console).
- —No native PSA/RMM pieces (ConnectWise, HaloPSA, Autotask, NinjaOne) yet — use the HTTP step or a custom API call.
- —Tables are capped at 10,000 rows × 100 columns. Run memory, log size and timeout limits apply.
- —Image generation is not available with Anthropic models. There is no public management API for workflows; import/export is JSON.
Common questions
What is the difference between an AI workflow, an agent and an RPA bot?
A workflow is the deterministic path: trigger, steps, branches, approvals. An agent is a step inside it that reasons with tools when the path is not fixed — which document is this, who should get it, what should the reply say. An RPA bot is another step: it executes clicks and keystrokes in a system without an API. Most production flows use all three.
Do I pay for the AI model?
You bring your own key — OpenAI, Anthropic, Google, Azure or a self-hosted endpoint — and your provider bills the model calls. ElectroNeek never meters tokens; you pay per workflow run. Document intelligence and Easley AI credits are included in the plan.
Can an agent use a system that has no API?
Yes. RPA bots are tools an agent can call: it decides what to enter, the bot enters it in SAP GUI, AS/400, Citrix or a desktop app, and the result comes back into the flow. The bot runs on a free Bot Runner on Windows.
How do I keep an agent from doing something consequential?
Put an approval step before it. Approvals go to Slack, Teams or email with the context attached, and the run waits. Document fields that fail validation can route to a person the same way. Agents also have a maximum number of steps and structured output you define.
How is this different from n8n, Make or Zapier?
They connect applications with APIs, and do it well. ElectroNeek adds desktop, SAP GUI and portal automation through bots, document intelligence by the page, and — for agencies — a tenant per client. Billing is per run, not per task, step or credit. Self-hosting n8n also makes you responsible for patching: it shipped a CVSS 10.0 unauthenticated RCE in January 2026.
Can I expose a flow as a tool for Claude, ChatGPT or another agent?
Yes: an MCP Tool trigger exposes a flow to agents, and the agent step can consume external MCP servers. A hosted MCP server that exposes every bot with typed inputs is on the roadmap.