Why OpenAgent: a workflow engine and an AI orchestrator in one product, both editable by non-engineers

One product, two engines

Under the hood OpenAgent is two engines sharing one data model. The AI orchestrator answers: it grounds every reply in your knowledge base through RAG, cites its sources, and hands off to a human when it should. The workflow engine decides what happens around the answer: triage by intent, escalation rules, notifications, data capture, calls into your own systems.

The usual alternative is two vendors. A bot product (Fin, Ada, Lyro) answers questions, an automation product (Zapier, Make, or the helpdesk's own rule builder) moves tickets around, and the seam between them is webhooks an engineer maintains. Every rule change crosses the seam. Zapier alone starts at $19.99 a month for 750 tasks, where each action step in a flow is a billed task, so the glue has a meter too.

In OpenAgent the seam does not exist. A workflow step can ask the AI to classify intent, and the AI agent can trigger a workflow when it escalates. Both are configured in the same dashboard, by the same person, who does not need to be an engineer.

What “editable by non-engineers” actually means

Concretely, three things:

• Describe a flow in a sentence, get a draft. Type “when someone mentions urgent, ping Slack and escalate to a human” and the AI drafts the workflow as a step graph. You review it, fill in the specifics (which channel, which team), and turn it on. Nothing goes live without a human approving the draft.
• A step library, not a scripting language. Thirteen step kinds cover the real cases: classify intent, condition, ask the visitor, extract from the conversation, calculate, set variable, reassign team, escalate to human, announce to visitor, send email, notify webhook, call function, delay. If a flow needs your own backend, an admin registers an HTTP function once and every workflow can call it.
• Templates for the flows every team builds first. Intent triage, VIP and urgent escalation, after-hours announcements, lead capture. Start from one, rename the pieces, ship it.

The same applies to the AI side: knowledge ingestion is paste-a-URL or drop-a-PDF, and the agent's behaviour (tone, escalation threshold, greeting) is settings, not code.

Your model, your keys, your data path

OpenAgent does not resell inference. You plug in your own Gemini, OpenAI, Anthropic, or OpenRouter key (or a self-hosted model) and the platform orchestrates it. That one decision carries most of the product's advantages:

• Predictable spend. Tokens at provider list price plus a flat platform fee. No per-resolution meter, no billing surprise when the AI has a good month. The full economics are on the pricing comparison: at 5,000 conversations a month the difference against per-resolution vendors is roughly 30× a year.
• A data path legal can approve. Conversations go to the model provider you have already cleared, under your DPA, in the region you chose. For EU teams and regulated industries (the German Mittelstand pattern: too big for a toy bot, too careful for a US-hosted black box), this is usually the deciding factor before price.
• No model lock-in. When a better or cheaper model ships, you swap a key. Your workflows, knowledge base, and history stay put.
• Dedicated infrastructure when you need it. The Business tier runs your workspace on dedicated infrastructure in the region you need, with an SLA, SSO, and audit log export.

Unlimited seats, because seats are the wrong meter

Per-seat pricing punishes exactly the behaviour good support needs: putting more humans close to customers. OpenAgent charges per site, not per person. Invite the whole team, the founders, the on-call engineer; the bill does not move.

What we deliberately don't do

An honest pitch includes the edges. OpenAgent does not ship marketing automation (Series-style lifecycle campaigns), surveys, a published help-center product (we index the one you already have), or voice. If those are core requirements, Intercom or Zendesk are better fits and the comparison pages say so per vendor. We build the support conversation surface: live chat, the AI agent, the workflows around them, and the CRM underneath.

The short case

• One product where competitors need two plus glue: workflow engine and AI orchestrator on one data model, both editable by the support team itself.
• Your own LLM keys: list-price tokens, provider-level data terms, no lock-in, roughly 30× cheaper at volume than per-resolution billing.
• Flat, boring pricing: $36/site/year billed annually ($5/month billed monthly), unlimited seats.
• A deployment story for careful teams: your choice of model provider and region, and dedicated infrastructure with an SLA on the Business tier.

If your evaluation is mostly about cost, start with the cost comparison calculator and put your own volume in. If it is mostly about trust, read how the grounding pipeline works and then ask us the hard questions.