Magic commands are special instructions prefixed with / that let you directly control conversation state without waiting for the AI to interpret your intent.
| Command | Wait | Compressed Summary | Long-term Memory | Response Content |
|---|---|---|---|---|
/compact |
⏳ Yes | 📦 Generate new | ✅ Background save | ✅ Compact complete + Summary |
/new |
⚡ No | 🗑️ Clear | ✅ Background save | ✅ New conversation prompt |
/clear |
⚡ No | 🗑️ Clear | ❌ No save | ✅ History cleared prompt |
| Command | Response Content |
|---|---|
/history |
📋 Message list + Token stats |
/message |
📄 Specified message details |
/compact_str |
📝 Compressed summary content |
/dump_history |
📁 Exported history file path |
/load_history |
✅ History load result |
Display a list of all uncompressed messages in the current conversation, along with detailed context usage information.
/history
Example response:
**Conversation History**
- Total messages: 3
- Estimated tokens: 1256
- Max input length: 128000
- Context usage: 0.98%
- Compressed summary tokens: 128
[1] **user** (text_tokens=42)
content: [text(tokens=42)]
preview: Write me a Python function...
[2] **assistant** (text_tokens=256)
content: [text(tokens=256)]
preview: Sure, let me write a function for you...
[3] **user** (text_tokens=28)
content: [text(tokens=28)]
preview: Can you add error handling?
💡 Tip: Use
/historyfrequently to monitor your context usage.When
Context usageapproaches 75%, the conversation is about to trigger auto-compact.If context exceeds the maximum limit, please report the model and
/historylogs to the community, then use/compactor/newto manage context.Token calculation logic: ReMeInMemoryMemory implementation.
View detailed content of a specific message by index.
/message <index>
Parameters:
index- Message index number (starting from 1)
Example:
/message 1
Output:
**Message 1/3**
- **Timestamp:** 2024-01-15 10:30:00
- **Name:** user
- **Role:** user
- **Content:**
Write me a Python function that implements quicksort
Display the current compressed summary content.
/compact_str
Example response (when summary exists):
**Compressed Summary**
User requested help building a user authentication system, login endpoint implementation completed...
Example response (when no summary):
**No Compressed Summary**
- No summary has been generated yet
- Use /compact or wait for auto-compaction
Manually trigger conversation compaction, condensing all current messages into a summary (requires waiting), while saving to long-term memory in the background.
/compact
Example response:
**Compact Complete!**
- Messages compacted: 12
**Compressed Summary:**
User requested help building a user authentication system, login endpoint implementation completed...
- Summary task started in background
Unlike auto-compaction,
/compactcompresses all current messages, not just the portion exceeding the threshold.
Immediately clear the current context and start a fresh conversation. History is saved to long-term memory in the background.
/new
Example response:
**New Conversation Started!**
- Summary task started in background
- Ready for new conversation
Immediately clear the current context, including message history and compressed summaries. Nothing is saved to long-term memory.
/clear
Example response:
**History Cleared!**
- Compressed summary reset
- Memory is now empty
⚠️ Warning:/clearis irreversible! Unlike/new, cleared content will not be saved.
Save current conversation history (including compressed summary) to a JSONL file for debugging and backup.
/dump_history
Example response:
**History Dumped!**
- Messages saved: 15
- Has summary: true
- File: `/path/to/workspace/debug_history.jsonl`
💡 Tip: The exported file can be used with
/load_historyto restore conversation history, or for debugging analysis.
Load conversation history from a JSONL file into current memory. Existing memory will be cleared first.
/load_history
Example response:
**History Loaded!**
- Messages loaded: 15
- Has summary: true
- File: `/path/to/workspace/debug_history.jsonl`
- Memory cleared before loading
Notes:
- File source: Loaded from
debug_history.jsonlin the workspace directory - Maximum load: 10,000 messages
- If the first message in the file contains a compressed summary marker, the summary will be restored automatically
- Current memory is cleared before loading — make sure to backup important content
⚠️ Warning:/load_historyclears current memory before loading. Existing conversation will be lost!
In chat, send /daemon <subcommand> or use short names (e.g., /status is equivalent to /daemon status). From the terminal, run copaw daemon <subcommand>. These run without going through the Agent.
| Command | Description |
|---|---|
/daemon status or /status |
Show runtime status (config, working directory, memory service, etc.) |
/daemon restart or /restart |
In-process restart (channels, cron, MCP) when in chat; from CLI prints instructions only |
/daemon reload-config |
Re-read and validate config (channel/MCP changes require /daemon restart or process restart) |
/daemon version |
Version number, working directory, log path |
/daemon logs or /daemon logs 50 |
View last N lines of console log (default 100, from copaw.log in working directory) |
From the terminal:
copaw daemon status
copaw daemon version
copaw daemon logs -n 50