AI Modes
Understand the three ways to work with AI in Omnilib — Chat, Agent, and Plan mode.
Omnilib's AI operates in three distinct modes. Each mode gives the AI a different level of autonomy and is suited to different kinds of work. Switch between them using the dropdown in the chat input area.
Chat Mode
In Chat mode, the AI converses with you without using any tools. It reads the context you provide — files you mention with @, attachments, and project memory — but it does not independently read files, run commands, or make changes.
Use Chat mode when:
- You want a quick explanation or answer
- You are brainstorming or exploring ideas
- You want to review a piece of text or code without the AI modifying anything
- You need a fast response without the overhead of tool execution
Chat mode is the fastest mode because the AI responds directly without orchestrating tool calls.
Agent Mode
In Agent mode, the AI has full access to a set of tools and can use them autonomously to complete tasks. It reads files, writes and edits code, runs shell commands, searches your project semantically, and more.
Use Agent mode when:
- You want the AI to implement a feature or fix a bug end to end
- You need the AI to search across multiple files to find something
- You want the AI to run a build or test command and react to the output
- You are delegating a task that requires several interdependent steps
Tool Execution Badges
Every tool call the AI makes appears as a badge in the chat timeline. Each badge shows the tool name, status, and an expand option to view the result or diff. For file modifications, you can accept or reject each change individually.
Approval for Destructive Actions
Agent mode requires your approval before executing actions that are difficult to reverse:
- Deleting files
- Running shell commands that modify system state
- Writing to files outside the project directory
The AI presents a confirmation prompt with a description of what it will do. You can approve, reject, or ask the AI to take a different approach.
Available Tools
| Tool | What the AI can do |
|---|---|
| File read | Read any file in the project |
| File write | Create new files |
| File edit | Make targeted edits to existing files |
| Bash execute | Run shell commands in the project terminal |
| Vector search | Find relevant files using semantic similarity |
| LaTeX compile | Compile .tex files and return errors or PDF output |
| Graph operations | Query and update the project knowledge graph |
| Bibliography | Search, import, and cite references |
| Memory | Query and update project memory |
| Notebook | Execute notebook cells and read output |
| Tags | Add, remove, and query file tags |
| Comments | Read and post inline comments on files |
Plan Mode
In Plan mode, the AI first generates a detailed implementation plan and presents it to you for review. Only after you approve the plan does the AI begin executing it, step by step.
Use Plan mode when:
- The task involves changes across many files
- You want to understand and influence the approach before any code is written
- You are working in a sensitive codebase where unexpected changes would be disruptive
- You want a structured, reviewable audit trail for a complex task
How Plan Mode Works
- Describe your task in the chat input
- The AI analyzes the project and produces a numbered implementation plan
- Review the plan — you can edit it, ask the AI to revise specific steps, or remove steps you do not want
- Click Execute Plan to begin
- The AI executes each step in sequence, showing tool execution badges for every action
- If any step fails or produces unexpected output, the AI reports back before continuing
Plan mode uses the same tools as Agent mode, but it commits to an approach first rather than deciding step by step.
Switching Modes
Open the mode dropdown in the chat input area and select Chat, Agent, or Plan. Your selection persists for the current session. You can switch modes mid-conversation — the chat history is preserved, and the new mode applies from your next message.
Related
- AI Chat — Full chat interface reference
- Sub-agents — Parallel agents spawned during complex tasks
- Agent Behaviors — Configure how the AI approaches tasks
- Models — Choose which AI model powers each mode