AI Agents

Create personalized AI agents that are bound to your workspace and use your memories as their knowledge base.

Table of Contents

• What Are Agents?
• Creating an Agent
• Agent Settings
• Memory Blocks
• Permissions
• Using Agents
• Best Practices

What Are Agents?

Agents are stateful, self-modifying AI assistants that evolve through conversation. Unlike standard AI chats, Kybernesis agents:

• Remember everything - Access your workspace memories for context
• Learn about you - Store information about you in their memory blocks
• Evolve their personality - You can tell them "be more casual" or "call yourself Max"
• Persist across sessions - Conversations and learnings are saved permanently

Example Interaction:

You: "From now on, be more casual and call yourself Max"
Max: "Hey! Got it. I'll keep it chill from here on out."
[Agent's name and persona auto-update in database]

Creating an Agent

Step 1: Navigate to Agents

1. Click Agents in the sidebar navigation
2. Click CREATE_AGENT button

Step 2: Choose Workspace(s)

Select which workspace(s) the agent should have access to:

• Agents can read memories from selected workspaces
• You can select multiple workspaces for broader knowledge access
• The first selected workspace is the agent's "primary" workspace

Step 3: Identity

Give your agent a name and description:

Step 4: Persona

Define the agent's personality in the persona memory block:

I'm Samantha, a creative and data-driven marketing manager.
I'm professional but friendly, and I always think about brand consistency.
I use clear, engaging language.

Persona tips:

• Be specific about tone (casual, formal, technical)
• Include the agent's name and role
• Describe communication style
• Mention domain expertise

Step 5: Model & Temperature

Step 6: Memory Scope

Configure which workspace memories the agent can access:

• Tags: Filter memories by specific tags (e.g., marketing, brand)
• Context Limit: How many memories to include (10-50)
• Leave tags empty to access all workspace memories

Step 7: Permissions

Step 8: Create

Click Create Agent to finish. You'll be taken to the agent detail page where you can start chatting.

Agent Settings

Accessing Settings

1. Navigate to Agents in sidebar
2. Click on an agent card
3. The Settings Sidebar appears on the right

Editable via Sidebar

These settings require the sidebar UI:

• Model selection
• Temperature
• Memory tags and context limit
• Permissions

Editable via Chat

These can be changed by talking to the agent:

• Name ("call yourself Max")
• Personality traits ("be more casual")
• Communication style ("use bullet points")
• Behavioral instructions ("always greet me first")

Memory Blocks

Agents have structured memory blocks that persist across conversations:

Default Blocks

Viewing Memory Blocks

In the agent detail page:

1. Look at the Settings Sidebar on the right
2. Scroll to the Memory Blocks section
3. See current values with character counts

How Agents Update Blocks

When you tell an agent something important, it uses internal tools to update its memory:

You: "I prefer bullet points over paragraphs"
Agent: "Got it! I'll use bullet points going forward."
[Agent internally updates "human" block with your preference]

Manual Editing

1. Click EDIT on any memory block
2. Modify the content
3. Click SAVE
4. Changes take effect immediately

Permissions

Read Memories

When enabled, the agent can search your workspace memories:

You: "What's our brand voice?"
Agent: [Searches workspace for brand guidelines]
Agent: "Based on your brand docs, your voice should be..."

Write Memories

When enabled, the agent can save important information to workspace:

You: "Save this meeting summary to my workspace"
Agent: [Creates new memory with title and tags]
Agent: "Saved to workspace as 'Q4 Planning Meeting Summary'"

Trigger phrases:

• "Save this to memory"
• "Add this to my workspace"
• "Remember this for later"
• "Store this information"

Search Conversations

When enabled, the agent can search its past conversations:

You: "When did we discuss the Q3 budget?"
Agent: [Searches conversation history]
Agent: "We discussed Q3 budget on November 15th. You mentioned..."

Use Connectors

When enabled, the agent can trigger syncs from connected sources (Google Drive, Notion).

Using Agents

In the Web UI

1. Navigate to Agents in sidebar
2. Click an agent to open detail view
3. Type in the chat input at the bottom
4. Agent responds using workspace context

Via Claude Code Plugin

1. Install the Kybernesis Claude Code plugin
2. Sync your agents: npx kybernesis-plugin sync
3. Use the kybernesis-agents skill
4. Chat with agents by name

See Claude Code Plugin for setup instructions.

Via MCP

1. Generate an agent-specific API key in the MCP Connection section
2. Configure your MCP client with the agent endpoint
3. Use simplified tools: chat, search_memories, etc.

See MCP Setup for configuration.

Best Practices

Naming Agents

• Use memorable, distinct names
• Names should reflect the agent's role
• Examples: "Max" (engineering), "Samantha" (marketing), "Alex" (HR)

Writing Personas

Good persona:

I'm Max, a senior engineer focused on backend systems and API design.
I'm direct and technical, preferring code examples over long explanations.
I ask clarifying questions before diving into solutions.

Avoid:

I am an AI assistant that helps with engineering tasks.

Scoping Permissions

• Start restrictive: Only enable permissions as needed
• Read-only first: Most agents only need to read, not write
• Monitor usage: Check token usage in the sidebar

Organizing Workspaces

• Create dedicated workspaces for different agent domains
• Use tags to segment knowledge within a workspace
• Example: Marketing agent only sees marketing, brand, campaign tags

Troubleshooting

Agent Not Finding Memories

• Check the agent's workspace access permissions
• Verify tag filters aren't too restrictive
• Ensure memories exist in the workspace
• Try increasing the context limit (10 → 25)

Agent Forgetting Context

• Agents use memory blocks for persistent context
• Long conversations may exceed context limits
• Start a new conversation for fresh context
• Check if memory blocks have relevant information

Persona Not Updating

• Wait a few seconds for the update to save
• Refresh the page to see latest state
• Check the memory block history in sidebar
• Verify you're not at the character limit

High Token Usage

• Reduce context limit (fewer memories retrieved)
• Use more specific tag filters
• Switch to a faster model (Haiku) for simple tasks
• Archive old conversations

What's Next?

• Claude Code Plugin - Use agents in Claude Code
• Agent Chat - Advanced chat features
• MCP Setup - Integrate with other AI tools

Need help? Check the API Reference for technical details, or join our Discord community.