Skip to main content

​
Agents

Agents are custom AI assistants built for specific tasks, teams, or workflows. They combine a tailored system prompt, curated knowledge sources, selected tools, and guardrails into a reusable conversational experience that anyone in your organization can use.

​
What are agents?

An agent is a configured AI persona that goes beyond a simple prompt template. Agents can:
  • Follow detailed instructions specific to your use case
  • Access a knowledge base of uploaded documents (RAG)
  • Use connected tools and integrations
  • Enforce guardrails on what topics to address or avoid
  • Be shared across your organization or kept private

​
Example agents

AgentDescription
HR Policy AssistantAnswers employee questions using your HR handbook and policies
Code ReviewerReviews pull requests against your team’s coding standards
Sales EngineerAnswers technical questions using product documentation
Onboarding BuddyGuides new hires through your company’s setup process
Security AuditorReviews configurations against compliance frameworks

​
Creating an agent

Navigate to Agents in the sidebar and click Create Agent to open the agent builder.

​
Basic configuration

1

Name and description

Give your agent a clear, descriptive name and a short description that helps users understand what it does.
2

Avatar

Choose an icon or upload a custom image to visually distinguish your agent in the sidebar and conversations.
3

System prompt

Write the core instructions that define your agent’s behavior. This is the most important part of agent creation.
4

Model selection

Choose which AI model powers your agent. Consider the tradeoff between capability and cost.

​
Writing effective system prompts

The system prompt is the foundation of your agent. It tells the AI who it is, what it should do, and how it should behave. Best practices:
  1. Start with the role — “You are a [role] that helps [audience] with [task].”
  2. Define the scope — Clearly state what the agent should and should not help with.
  3. Set the tone — Specify whether responses should be formal, casual, technical, etc.
  4. Provide examples — Include example questions and ideal responses.
  5. Add guardrails — Specify topics to avoid, disclaimers to include, or escalation paths.
You are the Tegendo HR Assistant. You help employees find answers
to questions about company policies, benefits, and procedures.

Rules:
- Only answer questions based on the uploaded HR documents
- If you don't know the answer, say so and direct the user to hr@company.com
- Never provide legal advice — always recommend consulting with HR directly for sensitive matters
- Be friendly and professional in tone

​
Knowledge sources (RAG)

Upload documents to give your agent specialized knowledge beyond its base training data. When a user asks a question, the agent searches the knowledge base using retrieval-augmented generation (RAG) to find relevant passages and ground its response in your actual documents.

​
Supported file types

  • PDF documents
  • Word documents (DOCX)
  • Text files (TXT, MD)
  • CSV / XLSX spreadsheets
  • HTML pages

​
How RAG works

  1. Documents are chunked into overlapping segments
  2. Each chunk is embedded into a vector representation
  3. When a user asks a question, the query is embedded and compared against the chunks
  4. The most relevant chunks are injected into the context alongside the system prompt
  5. The model generates a response grounded in the retrieved content
Knowledge base documents are processed at upload time. Large documents may take a few minutes to index. You can upload up to 100 MB of documents per agent.

​
Tools

Enable tools to give your agent the ability to take actions beyond text generation:
  • Web search — Search the internet for real-time information
  • Google Drive — Read and search files in connected Google Drive accounts
  • Slack — Search and post messages to connected Slack workspaces
  • Custom MCP servers — Connect to any tool via the Model Context Protocol
Tools are configured at the organization level by admins and can be selectively enabled per agent.

​
Guardrails

Guardrails define boundaries for your agent’s behavior:
  • Topic restrictions — Limit which subjects the agent will discuss
  • Output filters — Prevent the agent from generating certain types of content
  • Escalation rules — Define when the agent should suggest speaking with a human
  • Disclaimer requirements — Automatically append disclaimers to certain response types

​
Using agents in conversations

​
Starting an agent conversation

There are three ways to start a conversation with an agent:
  1. Sidebar — Click on an agent in the Agents section of the sidebar
  2. Plus menu — Click the + button in the input bar and select an agent
  3. Slash command — Type @agent-name in the input bar to invoke an agent inline

​
Agent conversations vs regular chats

When you start a conversation with an agent, the agent’s system prompt, knowledge base, and tool configuration are automatically applied. The conversation is tagged with the agent’s name and icon so you can easily identify it in your history. You can switch away from an agent mid-conversation by selecting a different model or agent. The previous agent’s context will be preserved in the conversation history.

​
Agent marketplace

The agent marketplace is where your organization’s agents are published and discoverable. Browse agents by category, search by name or description, and see usage statistics.

​
Publishing an agent

When you create an agent, it starts as a draft visible only to you. To make it available to your organization:
  1. Open the agent in the agent builder
  2. Click Publish
  3. Choose the visibility: Organization (everyone) or Team (specific teams)
  4. Add tags to help users discover the agent

​
Agent analytics

Published agents include usage analytics:
  • Total conversations started
  • Active users per week
  • Average conversation length
  • User satisfaction ratings

​
Deploying agents

Agents can be deployed beyond the Tegendo.AI interface:

​
Embed widget

Generate an embeddable chat widget to add your agent to any web page. The widget supports full conversation functionality including streaming, file uploads, and tool usage.

​
API access

Every agent is accessible via the Tegendo.AI API. Use the agent ID to start conversations programmatically. See the API reference for details.

​
Slack bot

Deploy an agent as a Slack bot that responds to messages in specific channels or via direct message.