Agent Chat

Learn how to effectively chat with your Kybernesis agents and leverage their full capabilities.

Table of Contents

• Chat Interface
• Conversation Basics
• Using Workspace Context
• Agent Tools
• Saving to Memory
• Memory Block Updates
• Conversation History
• Tips & Tricks

Chat Interface

Web UI Layout

The agent detail page has a split layout:

+----------------------------------+-------------------------+
|                                  |  Settings Sidebar       |
|   Chat Interface                 |                         |
|   -----------------              |  [Agent Settings]       |
|                                  |  [Memory Blocks]        |
|   [Agent messages appear here]   |  [Usage Stats]          |
|                                  |  [MCP Connection]       |
|                                  |                         |
|   -----------------              |                         |
|   [Type a message...]     [Send] |                         |
+----------------------------------+-------------------------+

Sending Messages

1. Type your message in the input field at the bottom
2. Press Enter or click Send
3. Wait for the agent's response (usually 1-3 seconds)

Response Metadata

Each agent response shows:

• Memories used: How many workspace memories informed the response
• Token count: Input/output tokens consumed
• Time: When the message was sent

Conversation Basics

Starting a Conversation

When you open an agent, you can:

• Start a new conversation immediately
• Continue a previous conversation (select from history)

Conversation Context

Agents maintain context within a conversation:

You: "What's our Q4 revenue target?"
Agent: "Based on the planning docs, it's $2.5M."

You: "How does that compare to Q3?"
Agent: "Q3 was $2.1M, so Q4 is a 19% increase."
[Agent remembers you're discussing quarterly targets]

Ending Conversations

Conversations are automatically saved. You can:

• Start a new conversation for a fresh context
• Return to old conversations from the history panel

Using Workspace Context

Automatic Retrieval

When you ask a question, the agent automatically searches your workspace:

You: "What's our refund policy?"
Agent: [Searches workspace for "refund policy"]
Agent: "According to your Customer Service Guidelines document,
        refunds are processed within 5-7 business days..."

How It Works

1. Your message is analyzed for intent
2. Relevant memories are retrieved via hybrid search
3. Memories are included in the agent's context
4. Agent responds using both knowledge and workspace context

Memory-Informed Responses

The agent will cite or reference workspace content:

Agent: "Based on your Brand Guidelines doc, the primary color
        should be #3B82F6 (blue) with #10B981 (green) as accent."

Scoped Retrieval

If your agent has tag filters configured:

• Only memories with matching tags are retrieved
• Use this to focus agents on specific domains
• Example: Marketing agent only sees marketing, brand tags

Agent Tools

Agents have internal tools they use automatically:

Memory Tools

You Don't Call These Directly

Tools are used automatically based on conversation:

You: "Remember that I prefer dark mode"
Agent: "Got it! I'll note that preference."
[Agent internally calls memory_insert to update "human" block]

Saving to Memory

Triggering Save

Ask your agent to save information to your workspace:

Trigger phrases:

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

What Gets Saved

When you ask to save, the agent:

1. Identifies the content to save
2. Creates a descriptive title
3. Adds relevant tags
4. Stores in your workspace

Example

You: "Here's the summary of our meeting: We decided to launch
      the new feature in March, focus on mobile first, and
      allocate $50k budget. Save this to memory."

Agent: "I've saved this to your workspace as 'Meeting Summary -
        Feature Launch Decision' with tags: meeting-notes,
        feature-launch, q1-planning."

Verification

Saved memories appear in:

• Your Topology graph as a new node
• Retrieval search results
• Future agent conversations

Memory Block Updates

Real-Time Updates

When an agent updates its memory blocks, changes appear immediately in the sidebar:

1. Agent decides to update a block
2. Sidebar refreshes automatically
3. You see the new value with character count

What Agents Update

Example: Persona Update

You: "From now on, call yourself Max and be more casual."

Agent: "Hey! I'm Max now. What's up?"

[Sidebar shows updated persona block]:
"I'm Max! I keep things casual and friendly..."

Viewing History

Click the history icon on any memory block to see:

• Previous values
• When changes occurred
• Whether changed via chat or sidebar

Conversation History

Accessing History

1. In the agent detail page, click View All Conversations
2. Or use the dropdown to select a previous conversation

Conversation List Shows

• Date/time of last message
• Preview of conversation topic
• Message count

Continuing Old Conversations

Click any conversation to:

• Load the full message history
• Continue where you left off
• Agent remembers the context

Searching Conversations

If the agent has Search Conversations permission:

You: "When did we talk about the budget issue?"
Agent: "We discussed budget on December 3rd. You mentioned
        concerns about Q1 allocation..."

Tips & Tricks

Be Specific

# Good
"What are our brand colors according to the Brand Guidelines?"

# Less good
"What are our colors?"

Reference Documents

You: "Based on our Q4 Planning doc, what are the key milestones?"
[Agent specifically looks for Q4 Planning document]

Set Context First

You: "I'm working on the product launch. From now on, focus on
      launch-related topics and prioritize marketing docs."
Agent: "Got it! I'll focus on product launch context."
[Agent updates objectives block]

Ask for Formats

You: "Give me a bullet-point summary of the meeting notes."
You: "Create a table comparing Q3 vs Q4 metrics."
You: "Write this as a formal email."

Check Memory Usage

Look at the Usage section in the sidebar:

• Messages this month
• Tokens consumed
• Memories retrieved
• Estimated cost

Use for Analysis

You: "Analyze the last 3 quarterly reports and identify trends."
[Agent searches for quarterly reports, synthesizes findings]

Common Patterns

Knowledge Q&A

You: "What's our vacation policy?"
Agent: [Retrieves HR docs, provides answer]

Document Summarization

You: "Summarize the main points from the Product Roadmap doc."
Agent: [Finds document, provides summary]

Research Assistant

You: "Find everything we have about competitor pricing."
Agent: [Searches workspace, compiles findings]

Meeting Prep

You: "I have a meeting about the Q1 budget. What should I know?"
Agent: [Retrieves budget docs, past meeting notes, provides briefing]

Content Creation

You: "Write a blog post outline about our new feature,
      using information from the Product Specs doc."
Agent: [Retrieves specs, creates outline]

Troubleshooting

Agent Not Using Workspace Context

• Check agent has Read Memories permission
• Verify workspace has relevant memories
• Try being more specific: "Based on my documents..."

Slow Responses

• Reduce context limit in settings
• Use faster model (Haiku vs Opus)
• Start new conversation (clear context)

Agent Forgetting Things

• Memory blocks persist; check if info is there
• Conversation context is limited; start fresh if needed
• Explicitly ask to save important information

Wrong Information Retrieved

• Check tag filters in agent settings
• Verify source documents are up to date
• Be more specific in your questions

What's Next?

• AI Agents - Create and configure agents
• Claude Code Plugin - Use agents in Claude Code
• MCP Setup - Connect to other AI tools

Need help? Check the API Reference for technical details.