4. OpenClaw TUI Interaction

TUI (Text User Interface) is OpenClaw's "command line chat window". When you can't easily open the web console on Jetson, or you're connected via SSH remotely, you can use TUI to complete most daily interactions in the terminal: conversations, model switching, status checking, session management, skill invocation, etc.

OpenClaw TUI 界面（可替换为你的截图）

1. How to Enter TUI

Execute in terminal:


openclaw tui

If you have already started openclaw gateway (e.g., gateway/server) in another terminal, you can directly execute openclaw tui to enter the interactive interface

2. What is the TUI Interface

You can understand TUI as a "chat box with commands":

Top: Conversation history (your messages, model replies, tool/skill outputs)
Bottom: Input box (enter questions or send slash commands)
Send method: Press Enter after typing

Recommended first-time actions:


xxxxxxxxxx
/status

3. Slash Commands

Slash commands generally need to be sent as standalone messages starting with /, used to control sessions and run operations.

Command	Function	When to Use	Example
`/help`	View help and command list	When you don't know available commands	`/help`
`/status`	Check current status	When suspecting disconnection, want to see context usage	`/status`
`/new`	Start new conversation line (new session)	Topic switching, avoid context interference	`/new`
`/reset`	Reset session (equivalent to restarting, same effect as new)	Conversation goes off track, want to clear context	`/reset`
`/model` / `/models list`	View/switch models	When wanting faster or stronger models	`/models list`
`/think medium`	Set thinking depth	Daily recommendation: use low or disable	`/think medium`
`/stop`	Stop current task	When stuck, want to stop immediately	`/stop`
`/context detail`	View currently injected context	Troubleshoot "why it remembers/forgets certain content"	`/context detail`
`/compact`	Compress earlier history into summary	Conversation too long, context nearly full	`/compact`
`/skill <name>`	Run a specific skill	When you know skill name and want to run directly	`/skill web_search`

Two additional "daily usage" recommendations:

Write only one slash command per message for best reliability
To quickly browse all commands, use /help or /commands

4. Sessions and Memory

In OpenClaw, these two concepts have clear divisions:

Session: Manages "where this conversation has progressed". When you use /new, /reset in TUI, you're essentially switching/reopening sessions.
Memory: Manages "content that should be retained long-term across sessions". It writes to disk (Markdown) and retrieves back to current session when needed.

4.1 Where are Session Files

Sessions are saved to disk in JSONL format, common locations (actual installation path may vary):

Index: ~/.openclaw/agents/<agentId>/sessions/sessions.json
Transcripts: ~/.openclaw/agents/<agentId>/sessions/<SessionId>.jsonl

A more convenient way for actual troubleshooting is to first execute:


xxxxxxxxxx
openclaw status

It usually provides the store path and recent session information.

4.2 How to "Use" Memory

You can understand Memory as "letting the robot take notes":

memory/YYYY-MM-DD.md: Daily process records, suitable for recording progress, temporary decisions
MEMORY.md: Long-term stable information (preferences, fixed rules, project constraints, etc.)

Practical recommendations:

To make a conclusion available across sessions: directly tell the robot "write this to MEMORY.md"
Before /new to switch to new topic: first use /compact or write key decisions to Memory, then switch sessions

5. Quick Start Workflow

Enter TUI: openclaw tui
Confirm status: /status
Select model: /models list → /model <number or name>
Daily conversation: directly enter questions and press Enter
Topic switching: /new
Context too long: /compact
Key conclusions need long-term retention: have robot write to Memory

6. Common Troubleshooting

6.1 No Response After Entering Commands

Confirm commands are sent as standalone lines (don't mix with regular text)
First use /help to see which commands are supported in current version

6.2 Feeling "It's Starting to Talk Nonsense or Can't Understand" or Being Interfered by History

Start new session: /new or /reset
Check what context was injected: /context detail

6.3 Responses Getting Slower

Reduce thinking depth: /think low or /think minimal
Compress history: /compact