Set Up Cypher LM Step by Step

This path walks through building a Cypher LM agent from a blank account to a live, working assistant, in the order the wizard actually asks for things. If you just want the big picture first, read Getting Started with Cypher LM before this one — it introduces the five products this setup will eventually use.

Creating an account takes one click. From the Cypher LM homepage, choose "Sign In" in the top-right corner and continue with Google — there's no password to create or remember, and Cypher LM only requests your email and name. You're redirected straight to your dashboard afterward, and every agent you build from that point on is private to your account.

Before you open the wizard, it helps to have a few things ready: your business website address if you have one, a short description of what you do, and any documents you'd want a new employee to read on their first day — a price list, an FAQ, a policy document. None of these are required to start, but having them nearby will make the later steps faster.

The wizard opens from the "Build Your Agent" button on your dashboard or the homepage. It's a single guided flow — you can't lose progress by navigating between steps, and nothing goes live until you explicitly deploy at the end. The next lesson covers the first real step: describing your business.

It's worth setting expectations before you dive in: this is not a technical integration project. There is no code to write, no server to configure, and no separate tool to learn — everything happens inside the same guided flow, in plain language, and most people finish it well within ten minutes. If you've ever filled out a detailed onboarding form for a new piece of software, this will feel familiar, except that at the end of it you have a working voice agent instead of a saved draft.

One more piece of groundwork: decide who in your business should see the finished agent before it's shared with customers. Even a single colleague listening to a couple of test calls and flagging anything that sounds off is enough to catch most issues early. You don't need a formal review process — just a second pair of ears before the link goes out widely.

If this is your first time building any kind of AI agent, resist the temptation to research extensively before opening the wizard. The questions it asks are deliberately ordinary — about your business, not about AI — and the fastest way to learn the system is to answer them honestly for a real or even a test version of your business and see what comes out the other end.

The first set of questions in the wizard are about your business, and they matter more than any other step — everything your agent knows later traces back to what you enter here. Be specific. A vague answer produces a vague agent.

• Business name — your official company or brand name, exactly as customers know it.
• Business website — if you have one, enter the URL. Cypher LM visits the site, reads its content, and uses it to pre-fill later answers, which saves significant time.
• Industry or business type — healthcare, retail, real estate, education, and so on.
• What your business does — two to four sentences, written as real sentences rather than keywords.
• Target customers — who you typically serve, described concretely rather than generically.
• Key products or services — three to five of your most important offerings, which the agent will know by heart.

If you did enter a website and let Cypher LM scrape it, review every pre-filled answer before moving on. Auto-filled content is a starting point, not a guarantee — correct anything that's out of date or slightly off before it becomes part of your agent's knowledge.

The next screen asks for contact information: a phone number the agent can hand out when a caller asks how to reach you directly, with the country code included, plus an optional support email. This number is not one the agent itself answers — it's the number your agent tells customers to use if they need a human.

It helps to think of this whole section as writing the business profile you'd hand to a new hire on day one, not as filling out a form for a database. The target-customers field, for example, is far more useful as "homeowners in their first five years of ownership looking for routine maintenance" than as "homeowners." The agent uses this context to decide how to phrase things and which of its key services to lead with when a caller's request is ambiguous, so concrete, real sentences consistently produce a sharper agent than short keyword answers.

Businesses with more than one location or more than one distinct service line sometimes wonder whether they need a separate agent for each. There's no fixed rule — a single agent that knows about every location and service usually works fine for a smaller business, while a larger one may prefer a dedicated agent per location so each one can have its own local contact number and slightly different knowledge.

The next section is where the agent stops being a form and starts becoming a character. Think of it as writing a short job description for a new employee who will answer your phone.

Give the agent a human name — something a caller would be comfortable hearing, like "Sara" or "James." Write its first message: the exact greeting it opens every call with. A short, natural line works best — something like "Hello! Thanks for calling Riverside Clinic. I'm Sara, how can I help you today?" Then choose a tone of voice: Professional, Friendly, Formal, or Casual. Formal suits legal, medical, or financial businesses; Friendly suits most everyday retail and service businesses.

Next comes behaviour. List the agent's core responsibilities in plain language — answering product questions, taking orders, booking appointments, whatever applies to you. Then list topics it should avoid entirely, such as discussing competitor pricing or giving medical advice it isn't qualified to give. Define a fallback line for when it genuinely doesn't know an answer — something honest like "I'll note your question and have our team follow up with you shortly" works better than letting the agent guess. Finally, decide whether and when it should hand a caller off to a person, such as when someone is upset or repeatedly dissatisfied.

Choosing a language

The last field in this section is language: English only, Urdu only, or English and Urdu together. Bilingual mode is the right default for most Pakistani audiences, because callers often move between the two languages mid-sentence, and the agent follows naturally when you've selected both.

Tone and behaviour are easy to under-invest in because they feel less important than the factual questions earlier in the wizard, but they're what separates an agent that sounds like your business from one that sounds like a generic assistant reading a script. A clinic's agent and a retailer's agent should not sound the same, even if both are friendly — the clinic's fallback line probably routes more cautiously toward a human, while the retailer's can confidently handle most questions on its own. Take the extra minute here; it shows in every call the agent answers afterward.

A good first message is shorter than most people initially write. Callers calling a business expect a quick, natural greeting, not a long recitation of hours and services — save that detail for when the caller actually asks for it, and let the opening line simply welcome them and invite the question.

After the wizard's questions are done, Cypher LM generates a system prompt — the full instruction set that ties together everything you entered: your business, the agent's personality, its rules, and its language. This is the single most important artifact in the whole setup, because it is literally what the agent reads before answering any call.

You'll see it on the Summary page with three options: read it carefully to confirm it reflects your business accurately, copy it if you want to keep a version elsewhere, or edit it directly in the text box to add, remove, or sharpen any instruction. Spending two or three minutes here, checking that the business name, services, and tone are all correct, is worth more than any other step in the wizard.

Below the prompt is an optional but strongly recommended step: uploading knowledge files. These are documents the agent can draw answers from directly — a product catalogue, a price list, an FAQ document, a return policy. Click the upload button next to Copy, choose one or more files in PDF, DOCX, CSV, or TXT format (up to 4 MB each, up to ten files per agent), and a confirmation chip appears once each file is processed.

Once deployed, the agent can answer directly from the content of every file you uploaded — so a customer asking about a specific price or policy gets a precise answer instead of a vague one. The next lesson covers deployment itself and testing the agent before anyone else talks to it.

If you don't have a polished document ready, don't let that stop you from deploying — a short, plain-text list of your most commonly asked questions and their correct answers is often more useful to the agent than a long, formally formatted brochure. You can always add more files later; knowledge files are not a one-time setup step, they're something you'll likely return to add to as you notice what customers actually ask.

One practical tip: name your files so you'll recognize them months from now — "2026 Price List" rather than "final_v3_use_this_one.pdf." It costs nothing at upload time and saves real confusion later when an agent has accumulated several files and you need to replace just one of them.

Neither the system prompt nor the uploaded files are locked once the agent is live. You can return to an agent's settings at any point to refine the prompt further or swap in an updated document, which means this step doesn't need to be perfect — it needs to be good enough to deploy and test honestly.

Once you're satisfied with the system prompt and any knowledge files are uploaded, deployment is a single click: "Deploy Agent" at the bottom of the Summary page. You'll see a short animated screen while Cypher LM creates your live assistant using everything you configured — this usually takes ten to twenty seconds.

When it finishes, you land on the Agent Ready page, confirming your agent is live, and a confirmation email arrives with the details, including the agent's identifier. Deployment is a one-time action per agent — if you need to make changes afterward, you can build a new agent with corrected settings, or adjust the existing agent's details from the Agents section in your dashboard.

Before sharing the agent with real customers, test it yourself. From your dashboard, find the agent in your list and choose "Talk to Agent," which opens its dedicated page. Click "Start Call," speak naturally using your browser's microphone, and the agent responds in real time with its configured voice. Click "End Call" when you're done. Run a few test conversations, asking the kinds of questions a real customer would, before pointing anyone else at it.

You can also share the agent's direct link with teammates so they can test it themselves, in the format cypherlm.com/agent/[agent-id]. Once you're confident in how it answers, turn on booking or payments for the agent in its settings if you want it to schedule appointments or take payments mid-call — covered in Let Cypher S Book and Confirm Meetings for You and Get Paid on Calls with Cypher P, the next two product paths in this hub.

A good habit once the agent is live is to revisit it after its first handful of real calls, not just after your own test calls. Open the transcript of each one in Calls and read it the way a customer experienced it. You'll usually spot one or two small adjustments worth making — a phrase to add to the knowledge files, a line in the system prompt to tighten — and making them early, while call volume is still low, is far easier than discovering the same gap after dozens of customers have already hit it.

From here, the natural next step is deciding what the agent should do beyond answering questions — booking appointments, taking payments, or both. Those are covered in their own paths in this hub, and turning either on is a short settings change, not a rebuild of the agent you just created.