If you use AI chat tools daily, you have probably developed a ritual: open a new session, paste your standard context paragraph, explain who you are and what you need, then finally ask your actual question. Three minutes of setup before you can do the one thing you opened the tool to do.
Custom agents in Kolbo Chat eliminate that ritual entirely. You define the context, instructions, and knowledge once. Every session with that agent already knows your role, your standards, and how you want the output structured. This guide walks through exactly how to build them and what they are genuinely useful for.
What a Custom Agent Actually Is
A custom agent is not an autonomous bot that runs tasks in the background. It is a smarter chat session with persistent context baked in.
When you create an agent, you write instructions that describe its role, writing style, what it should and should not do, and what output format it should follow. You can also upload knowledge files: PDFs, DOCX documents, CSVs, code files. Every session you open with that agent starts with all of that context already loaded.
The result is a dedicated assistant that behaves consistently every time, without you re-explaining a single thing.
The Problem It Solves
Three specific patterns waste significant time for teams using general-purpose AI chat:
Re-explaining context every session. "I work in marketing at a B2B SaaS company, our audience is IT managers, we write in a direct conversational tone, we avoid jargon." You have probably typed some version of that dozens of times.
Copy-pasting prompt templates. Most professionals who use AI regularly maintain a separate document with their best prompts. Every session means opening that document and pasting from it. Custom agents make those templates the default behavior.
Inconsistent tone across team members. When five people on a team use the same general-purpose AI, they get five different writing styles in their output. A shared custom agent trained on your brand voice produces consistent output for everyone.
How to Create a Custom Agent in Kolbo Chat
Step 1: Name It Precisely
Open the Chat tool and navigate to the Agents section. When naming your agent, specificity matters. "Marketing Content Writer" is immediately useful. "AI Writing" is not. The name tells you and your team exactly what to open when you need a specific kind of help.
Step 2: Choose an Icon
When you have multiple agents, a visual icon makes the right one instantly identifiable. A small choice that saves friction when you are switching between agents throughout the day.
Step 3: Write the Instructions
This is the most important step. Good instructions have four components.
Define the role clearly. "You are a marketing assistant for [Company Name]. Your job is to write marketing copy, social media posts, and email content."
Specify writing style with concrete rules. "Write in second person. Use short sentences, under 15 words. Avoid corporate jargon. Write like a person, not a press release."
Define what the agent does NOT do. "Do not explain what you are about to write. Do not add a summary after the output. Do not offer multiple options unless specifically asked."
Specify output format. "When asked for a headline, always provide three options, each on its own line, with no preamble."
Step 4: Upload Knowledge Files
This is where agents become genuinely powerful. Upload the documents your assistant should reference without you having to paste them every session. Brand guidelines and tone-of-voice documents work especially well here, as do product descriptions, approved example copy, FAQ documents, templates, or a CSV of product names and SKU codes.
The agent reads these files and references them in its responses. Someone can ask "what's our return policy for international orders?" and get the correct answer pulled from your uploaded policy document.
Step 5: Share with Your Team
A finished agent can be shared with your team via a link. Everyone gets the same agent, the same settings, and the same consistent output. No one needs to recreate the setup.
Five Agent Types Worth Building First
1. Brand Voice Content Writer
Upload your brand guidelines, approved copy examples, and a list of banned phrases. Every team member who needs to write now has a single source of truth for your voice. Drafts come out consistent whether they were written by your most experienced copywriter or your newest hire.
2. Internal Knowledge Base
Upload your FAQ documents, return policies, service terms, onboarding guides, and HR policies. Team members can ask natural questions and get accurate answers instead of searching through Google Docs or asking a manager for the fifth time.
3. Code Reviewer
Upload your team's coding standards, preferred patterns, and style guidelines. The agent reviews submitted code in plain language, flagging issues against your actual standards rather than generic best practices.
4. Social Media Manager
This agent knows your active campaigns, target audiences, and platform-specific tone requirements. Brief it in two lines and get platform-specific versions for Instagram, LinkedIn, and X. Without an agent, producing three platform versions from one brief takes 15 to 20 minutes. With one, it takes 90 seconds.
5. Content Editor
Train an editor agent on your writing examples and style. Use it to clean up drafts, remove translation artifacts if your team writes in a second language, and maintain a consistent voice across all published content.
How to Write Instructions That Actually Work
The most common mistake is writing instructions that are too general. "Write well" and "be helpful" do not produce reliable behavior. Concrete rules do.
Compare these two instructions for the same agent:
Vague: "Write in a friendly tone and keep it readable."
Specific: "Write sentences under 15 words. Use contractions naturally. Avoid passive voice. Never use the word 'utilize.' Always start responses with the output, not a sentence explaining what you are about to write."
Both intend the same thing. Only the second one reliably produces it.
Before sharing any agent with your team, test it with at least three different types of requests covering the range of things people will actually ask. If it behaves oddly on any of them, adjust the instructions and test again.
One Agent Per Purpose
An agent that writes marketing copy AND reviews code AND answers HR questions ends up doing none of those things particularly well. Instructions for different domains pull in different directions.
Build one agent per distinct use case. The small amount of extra setup time pays off in output quality. A focused agent gives better answers than a general one, every time.
Where to Start
Think about what you re-explain to an AI tool every single day. What prompt do you always copy from a separate file? What context paragraph do you paste at the start of every session? That is your first agent.
Build it, test it for a week, then build the next one. Most teams identify their most valuable agent within the first day of using them.
Build your first custom agent and start every session with context already in place: app.kolbo.ai
