Agentic Agents
Overview
The Agentic Agents section is the core AI workspace within your project. It allows you to create and manage agent tabs — whether they are CLI-based, SDK-based, or ACP-based. Each tab runs an independent AI agent session, so you can work with multiple agents or conversations simultaneously.
Kaiju Hub AI ships with built-in support for Anthropic Claude Code, Google Gemini, and OpenAI Codex. The Pro version also supports custom CLI additions like OpenCode.
The agent that opens by default is determined by the provider and mode you selected in the Project Info area. Every time you left-click the + button, a new tab opens using that default configuration.
Kaiju Hub AI Context System
How It Works
Kaiju Hub AI uses a unique context bundling system to communicate with AI agents. Instead of sending raw messages directly, every prompt you send is bundled into a structured context file that the AI reads as a single package. This bundle includes your prompt text, any attached images, attached files, and references to blocks or projects you have included.
When you send a message, Kaiju Hub AI assembles all of these pieces into one bundle file and passes it to the AI agent. The agent reads the entire bundle at once, giving it full access to everything you have attached in a single, organized context.
What Gets Bundled
Prompt Text
Your message to the AI agent is included as the primary instruction.
Attached Images
Any images you attach are included in the bundle. You are not restricted to a small number — you can send many images in a single message.
Attached Files
Files you attach are bundled so the AI can read and reference their contents directly.
Project References
If you attach a project from the Project Hub, its path is included so the AI can reference and work with that project directly.
Block References
Blocks from the Block Library can be attached to a message. The AI will read and utilize the block content as part of its context.
Cross-Session Memory
Every bundle is stored on the backend after it is sent. This means the AI can reference context from previous tabs within the same session, and it can also look back at previous chat sessions entirely. For example, if you had a conversation last week about a specific feature, the AI can recall what you discussed and what you asked for.
This cross-session memory works across all tabs. If you send something in one tab, the context is available for the AI to reference from any other tab or future session.
Why This Matters
Reduced Token Usage
By bundling context into a structured file rather than appending raw history, the system cuts down on token consumption significantly.
More Context Material
You can include far more context — many images, multiple files, project references, and blocks — without hitting typical limitations.
Persistent History
The AI retains awareness of past conversations across sessions, so you do not need to repeat yourself or re-explain prior work.
Add Tab
Overview
The + button in the agent toolbar lets you create new agent tabs. Left-clicking the + button instantly opens a new tab using the default agent selected in the Project Info area. Right-clicking the + button opens a dropdown where you can choose a specific agent type.
Supported Agent Types
Markdown Viewer
Overview
The Markdown Viewer tab renders markdown content within the agent area. It is used to display formatted documentation, agent instructions, skill files, and other markdown-based content directly in the workspace without switching to an external editor.
Features
Rich Rendering
Supports full markdown syntax including headings, lists, code blocks with syntax highlighting, tables, links, and images.
Agent Documentation
View agent instruction files (CLAUDE.md, agents/*.md, skills/*/SKILL.md) rendered in a readable format.
Inline Display
Content renders inline within the agent tab area, so you can read documentation alongside your active agent sessions.
Agent Area
Overview
The Agent Area is the main workspace for interacting with AI agents. It displays chat messages, renders code with syntax highlighting, shows file and component context, and provides controls for managing the conversation. Each agent tab has its own independent agent area.
Tab Components
Code Terminal Tab
For CLI-based agents (Claude Code, Gemini CLI, Codex CLI). Shows a terminal-like interface where the agent executes commands and displays output in real-time.
SDK Chat Tab
For SDK-based agents (Claude SDK, Gemini SDK, OpenAI SDK). Provides a conversational chat interface for interacting with the agent.
ACP Chat Tab
For ACP (Agent Communication Protocol) agents. A specialized chat interface for communicating with dynamic protocol-based agents.
Agents & Skills Tab
Browse and manage agent capabilities, custom agents, and skill definitions for the current project.
Session History Panel
View a log of all past sessions for the current agent type, with the ability to load and review previous conversations.
Image Gallery
Browse images generated by AI agents during the current project session.
Context History
Overview
The Context History panel tracks all context that has been shared with AI agents throughout your project. It separates context into internal and external sources, giving you full visibility into what information agents have accessed.
Context Types
Internal Context
Context generated from within your project — selected files, checked code, component selections, and file content that you explicitly provided to agents during chat sessions.
External Context
Context from outside your project — web search results, fetched URLs, imported documentation, or other external data sources that agents referenced or that you attached.
Features
Chronological Log
Context entries are listed in chronological order so you can trace what was shared and when.
File Registry
An internal file registry tracks all files that have been shared with agents, including enhanced metadata and unique IDs for each file.
Context Indicators
The chat input area shows a context indicator displaying the number of files currently attached, so you always know what context is active.
Context Modal
Click the context indicator to open a detailed modal showing all current context — files, components, and projects — with options to add or remove items.
Internal & External
Overview
The Internal / External toggle in the agent toolbar controls how CLI-based agents are launched, including custom agents. This determines whether the agent runs inside the built-in workspace terminal or in a separate external terminal window on your system.
Modes
Internal
The agent runs inside the built-in workspace terminal. Output is displayed inline within the agent area, and the session is fully managed by Kaiju Hub AI.
External
The agent launches in a separate system terminal window (e.g. PowerShell, Terminal, iTerm). Useful when you need a full terminal environment or want to run agents alongside other command-line tools.
Shell Selector
When in External mode, a dropdown appears letting you choose which shell to use (e.g. PowerShell, Command Prompt, Git Bash). The selected shell is used when launching external CLI agent sessions.
Enhanced Chat Input
Overview
The Enhanced Chat Input is the main input area at the bottom of the agent workspace. It goes beyond a simple text box — offering image attachments, screen recording, file attachments, component selection, voice input, and granular permission controls for AI context.
Attachment Buttons
Gallery (Image Collection)
Opens the Image Upload Modal where you can browse, select, and upload images to include in your message. Supports multi-select with thumbnails and drag-and-drop upload.
Image Popup Modal
Click any attached image thumbnail to open it in a full-size centered overlay viewer. Close with the X button or by clicking outside the modal.
Screen Recording
Capture screen recordings directly from the chat input. The recorder captures frames at configurable FPS (30 or 60), tracks frame count and duration, and generates a manifest with metadata. Recorded frames can be attached to messages for visual context.
File Attachment
Attach project files directly to your message using the paperclip icon.
Workspace Context
Add workspace context using the brain icon — includes project files and metadata for richer AI understanding.
Component Selection
Select components or blocks from the library using the puzzle icon to include as context.
Include Context
Overview
The Include Context button is the first icon on the left sidebar of the chat input area. It allows you to attach project context — such as checked files, selected components, or workspace metadata — to your message before sending it to the AI agent.
How to Use
Left Click
Opens the context panel where you can browse and select which files and project context to include with your next message.
Right Click
Instantly includes the current context in your message without opening the panel. This is a quick shortcut to attach whatever files and context are currently selected.
Project Images
Overview
The Project Images panel is a centralized gallery for all images associated with your project. Upload screenshots, mockups, reference images, or any visual assets you want to attach to AI conversations or keep organized within the workspace.
Features
Upload Images
Click the upload area or drag and drop images directly into the panel. You can also paste images from the clipboard using Ctrl+V / Cmd+V.
Image Grid
Uploaded images are displayed as a thumbnail grid for easy browsing and selection.
Selection Controls
Select individual images by clicking, or use Select All / Deselect All to manage bulk selections. Selected images can be deleted with Delete All.
Attach to Messages
Selected images can be included as visual context when sending messages to AI agents, giving them reference material for your requests.
Screen Recordings
Overview
The Screen Recordings panel lets you capture screen recordings and configure capture settings. Recorded frames are saved as images that can be attached to AI agent conversations, providing visual context for your requests.
Capture Settings
Capture Speed
Choose between Slow (2 FPS) and Normal (4 FPS). Use Slow if recording is too fast or you are missing content. Normal captures more detail but uses more storage.
Frames Per Image
Control how many frames are combined per image. Set to 1 for individual screenshots, or higher values for grid compositions that show multiple frames in a single image.
Grid Columns
Set the number of columns in the grid layout when combining multiple frames into a single image. Use 1 for vertical stacking or higher values for wide grids.
Capture Source
Select which screen or monitor to capture. Shows available monitors with resolution and position details. The primary monitor is labeled for easy identification.
Recordings Tab
The Recordings tab displays all previously captured recordings. Browse past sessions, review captured frames, and attach them to new AI conversations for visual reference.
Settings
Overview
The Settings panel provides configuration options for the active AI agent. Access it from the gear icon on the chat input sidebar. Settings control permission modes and other agent-specific behavior.
Permission Mode
Skip All Permissions
When enabled, bypasses all permission checks for the agent (equivalent to --dangerously-skip-permissions). This lets the agent execute file edits, terminal commands, and other actions without confirmation prompts.
Restart Required
Changes to permission settings take effect on the next session. Close and reopen the agent tab to apply the new configuration.
Voice Input
Overview
The Voice Input button is the microphone icon at the bottom of the chat input sidebar. It provides platform-native speech-to-text functionality, letting you dictate messages to AI agents instead of typing.
How It Works
Activate
Click the microphone icon to start voice input. The icon shows an active state while listening.
Dictate
Speak naturally and your words are transcribed into the chat input field in real time using your system's speech recognition (Windows Speech on Windows, Dictation on macOS).
Send
Once done dictating, review the transcribed text and click send or press Enter to submit your message to the agent.
Shortcuts
Middle Mouse Button
On Windows and Mac (if your mouse has a middle button), press the middle mouse button to start recording. Click it again in the chat input area to stop recording and automatically send the message.
Custom Shortkey
You can set a custom keyboard shortcut for voice input in the Voice tab within Settings. This lets you start and stop voice recording without using the mouse.