Build your first agent
In this page you'll find a comprehensive guide to build your first agent.
Last updated
In this page you'll find a comprehensive guide to build your first agent.
Last updated
With Calk AI, creating an AI Agent is fast, flexible, and designed to fit your exact needs. Whether youβre building an assistant for onboarding, sales support, or document summarization, hereβs how to do it in three simple steps:
This is where you teach your agent what to do.
πΉ Example: For an assistant that summarizes sales calls, your prompt might be: "Summarize this conversation and highlight key objections, buying signals, and next steps."
π§© Tips for writing a good prompt
Be clear and direct.
Break tasks into steps.
Use bullet points and examples.
Try loading from our Prompt Library to get started fast.
β¨ You can test the prompt live before continuing.
Calk gives you 3 quick tools to build great prompts, even if you're not a prompt expert:
Check out our range of tools designed to assist you effectively!
This is where your agent gets context and capabilities.
π Connect Integrations
We support two types of integrations:
Knowledge Integrations β Notion and Google Drive These fetch documentation, playbooks, notes, and structured files.
Information Integrations β Slack, Intercom, HubSpot, GDrive, Notion, Web pages... These provide up-to-date business context directly into the agentβs hands.
π§ Agents with connected integrations = smarter, faster, more relevant.
π Upload Local Files
Want to train your agent on a few PDFs or internal memos? Just drag and drop.
π Add a Website
Need your agent to "read" a public page? Add any URL and let it ingest the content directly.
π οΈ Enable Tools
Web Search β for live internet browsing
Web Scraper β for turning sites into internal knowledge
Reasoning Engine β for more logic-heavy tasks
Data Tools β for generating charts from insights
A possible use case ‡οΈ
Spy on your competitors blog or website with a scraper!
Now define who your agent is.
π§βπΌ Name & Icon Make it easy to recognize and reuse!
π§ Model Choice Choose from top LLMs: GPT-4o, Claude, Mistral, Cohere. Each has strengths β pick the best for your use case.
π Tone Slider
0
= Serious and direct
1
= Creative and human-like
It adapts your agentβs language to fit the brand voice or target audience.
π·οΈ Tag by Domain Organize by team or function: Marketing, HR, Customer Care, Dev, etc.
Donβt build one general-purpose agent. Specialize.
One agent = one task = better results
Shorter prompts, cleaner outputs
Easier to troubleshoot and iterate
π Use agents in sequence Build workflows where agents feed each other:
Ask
@accountSummary
to scan your Hubspot and extract client insights. Then tag@coldemailer
to generate a hyper-personalized outreach with those insights from Hubspot.
This layered logic allows your team to mix domain expertise with automation, boosting precision and saving time.
You just have to click on one with made by : Calk Team
and tailor them to your needs!
π§ How it works You write a clear instruction, also known as a prompt. This tells the agent its job. The more specific, the better. You can read our Class on how to prompt : !
β¨ See examples in the β from Marketing to Onboarding.οΈ
You have pre made Assistants on your Assistants Page here β .