Short answer: if you are the person who plugs n8n into HubSpot, writes Make scripts for reporting, or automates employee onboarding — you already have everything you need to roll out AI in your organisation. Ragen.ai gives you a REST API where you call an AI Assistant that has access to your company documents, and suddenly your existing automations start doing things they never did before. This post shows exactly how.
The problem that rolling out ChatGPT does not solve
In most companies that start with AI, the script looks like this: leadership buys ChatGPT Team licences, declares an “AI transformation”, and six months later nothing has changed. Employees use ChatGPT to write emails, but business processes look identical to a year ago. Why?
Because ChatGPT is an interface for a human. A human walks in, asks, copies the answer, pastes it into the CRM. That is not automation — it is better autocomplete.
Real automation begins where AI becomes part of the data flow, not the user’s destination. Instead of “the employee opens AI”, AI enters every process that has a decision point dependent on company context. And this is where most market AI integrations fall apart.
The standard APIs from OpenAI or Anthropic give you a model that knows the internet. But it does not know your company. It does not know who your customer is, what your SLAs are, what procedures apply in the complaints team, where the contracts with specific suppliers live. Until you build a RAG pipeline yourself — a six-month project for a RevOps team — AI in your automation behaves like a consultant who just walked in for the first time.
What Ragen API actually gives you
The Ragen.ai API looks like a single endpoint at first glance, but behind it sits everything Ragen does in the UI: access to the organisation’s knowledge base, permission control, hybrid retrieval, multi-model routing.
In practice it looks like this:
import { Ragen } from "@ragenai/sdk";
const ragen = new Ragen({ apiKey: process.env.RAGEN_API_KEY });
const completion = await ragen.chat.completions.create({
assistantId: "sales-assistant",
messages: [
{ role: "user", content: "What is the status of Acme Corp and when did we last talk to them?" }
],
});
console.log(completion.choices[0].message.content);The answer comes back from a Sales Assistant that has access to the CRM, call transcripts, and the offers database. From the calling application’s perspective it looks like a simple question-answer endpoint. Underneath: hybrid retrieval against the knowledge base, the user’s permission check, selection of the appropriate model (GPT, Claude, Gemini, or open source), an answer with cited sources.
The difference versus the OpenAI Assistants API is practical, not theoretical. OpenAI gives you a framework to build an assistant. You have to load documents, manage context, set up retrieval, enforce permissions. Ragen gives you an assistant ready to use — one that already has documents, retrieval and permissions, because you configured it once in the web panel and from then on you just call it from anywhere.
That is the difference between a programming library and a service.
Five automations worth shipping first
Below are concrete scenarios we have seen at clients — they work not after six months of consulting, but after a week of work by someone with intermediate knowledge of n8n or Make.
1. Inbound lead qualification from a website form
A prospect fills in a form. The webhook from the form goes to n8n. n8n calls the Ragen API: “Does this lead match our ICP? Check our ICP documentation and the history of similar leads.” Ragen replies with structured JSON (high/medium/low fit + rationale). Based on the answer: high-fit leads go to HubSpot tagged “hot” with a Slack alert, medium go into the nurturing flow, low go to the long-term database.
What Ragen gives you that OpenAI does not: knowledge of your ICP documentation, knowledge of customer history, a single qualification standard. Without Ragen you would have to feed that documentation into every prompt, or build your own RAG.
Time to deploy: 4–6 hours of work for a RevOps lead with basic n8n knowledge.
2. Auto-summarising customer meetings into the CRM
Fireflies or Otter records the meeting. When it ends, a webhook fires the automation. The automation calls Ragen: “From this transcript, extract next steps, follow-up actions, customer blockers, and update the customer record in HubSpot following our sales notes structure.” Ragen knows the structure because at some point you uploaded a “how we keep sales notes” document to its knowledge base. The result lands in HubSpot via the API.
Time saved per rep: 15 minutes after every meeting. With 3 meetings a day and 20 working days — 15 hours per month per person.
Time to deploy: 6–8 hours.
3. Automated first-line replies to support email
A customer writes to support@. The mail arrives in a shared mailbox. The automation reads the email and asks Ragen: “Is this case suitable for an automatic reply from the FAQ knowledge base, or does it need a human? If automatic — propose an answer from our knowledge base.” Depending on Ragen’s decision, the email gets a draft auto-reply (for human approval) or goes into the team queue.
What Ragen gives you that vanilla GPT does not: the actual FAQ of your company, knowledge of the current returns policy, the B2B/B2C distinction in tone. Without that, the AI sounds generic and the customer instantly knows they wrote to a robot.
Time to deploy: 8–10 hours (more, because it requires an eval set of 50 historical emails).
4. Employee onboarding — generating a personalised plan
A new hire joins the company. HR enters them into Bambee/HiBob. A webhook fires the automation. The automation calls Ragen: “This person is joining department X in role Y. Generate a 14-day onboarding plan based on our onboarding policy and the materials for department X.” The plan comes back as structured JSON. The automation creates tasks in ClickUp, sends the calendar to Google Calendar, generates a list of documents to read.
What Ragen gives you: knowledge of your specific onboarding policy, knowledge of differences between departments, knowledge of the relevant industry regulations.
Time to deploy: 10–12 hours.
5. Daily sales report with AI commentary
Every morning at 8:00 a cron triggers the automation. It pulls data from HubSpot (deals closed yesterday, new leads, scheduled meetings), and calls Ragen: “Based on this data and our Q2 goals, write a short strategic comment for the sales team.” Ragen has access to the document with quarterly goals, so the commentary is not generic — it is specific to your company. The result lands in the #sales-daily channel on Slack.
What changes: a report nobody read (because it was just a number) becomes a report with a perspective. The manager no longer has to interpret the numbers themselves in the morning.
Time to deploy: 4–5 hours.
Comparison with OpenAI and Anthropic API
A fair conversation, because each platform has its place.
OpenAI Assistants API — broadest ecosystem, the most tool-calling tooling, the best documentation. Pick it when AI is the heart of your product (a consumer chatbot, a tool like Cursor). Limitations: company documents land at OpenAI, permission control is your problem, multi-model does not exist.
Anthropic Messages API — top reasoning quality, best for long context, excellent tool use. Pick it when the task needs subtlety (legal analysis, high-quality copy). Limitations: a single model, no native RAG, documents also land at Anthropic.
Ragen API — pick it when you need AI with your organisation’s context without building your own RAG layer. Data stays inside European infrastructure. A single endpoint gives access to Assistants trained on your documents, with permission control, with model choice (open source or commercial). Limitation: if the task does not need company context, OpenAI or Anthropic will do.
In practice we see hybrid architectures at clients — Ragen for tasks that require company context (lead qualification, summarisation, customer support replies), OpenAI/Anthropic for generic tasks (translation, summaries of public materials). Each tool where the value-to-cost ratio is best.
What’s next
If you already have a team writing automations — in n8n, Make, Zapier, or your own code — Ragen API is the fastest path to add AI with company context to those automations. You do not need a new stack, a new model provider, or a new RAG pipeline. You need one API key and a defined AI Assistant.
The documentation lives at docs.ragen.ai. The official TypeScript SDK is published as @ragenai/sdk, OpenAI-compatible — and plain REST works just as well.
The first step is not picking the API — it is identifying the process that makes the most sense to automate. Often that is not the process that looks obvious at first glance. That is why for clients where AI automation is a bigger project we start with an AI Mapping workshop — the output is a prioritised process map and a concrete plan for the first production deployment.
Related posts: