Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/lvndry/jazz/llms.txt

Use this file to discover all available pages before exploring further.

When you’re chatting with an agent, you can use special slash commands to control your session, view information, and configure your agent. All commands start with / and can be typed at any point during your conversation.

​
Quick reference

Type /help during any chat session to see the full list of available commands.

​
Session management

​
/new

Start a new conversation and clear the context. 
/new
This command:
  • Generates a new conversation ID
  • Clears all conversation history
  • Keeps the system prompt and agent configuration
  • Resets token usage tracking
Use this when you want a fresh start with the same agent without carrying over previous context.

​
/fork

Fork the conversation into a new branch from the last message. 
/fork
This command:
  • Creates a new conversation ID
  • Keeps only the system prompt and last user message
  • Clears all intermediate conversation history
  • Allows you to explore a different approach to your most recent question
Use this when you want to try a different approach to your last question without losing your main conversation.

​
/clear

Clear the terminal screen. 
/clear
This command clears the terminal display but preserves your conversation history in memory.

​
/exit

Exit the chat session. 
/exit
You can also use Ctrl+C to exit.

​
Agent management

​
/agents

List all available agents. 
/agents
Displays:
  • Agent name and ID
  • Description (if configured)
  • Model and provider
  • Reasoning effort level
  • Current agent indicator

​
/switch

Switch to a different agent in the same conversation. 
/switch <agent-name-or-id>
You can also run /switch without arguments to see an interactive agent picker. Examples: 
/switch coder
/switch code-reviewer
/switch 123e4567-e89b-12d3-a456-426614174000
When you switch agents:
  • Conversation history is preserved
  • The new agent can see all previous messages
  • Tool availability changes based on the new agent’s configuration

​
Configuration

​
/model

Show or change the model and reasoning effort. 
# Show current model
/model

# Change model
/model <provider>/<model>

# Change reasoning effort
/model reasoning <level>
Examples: 
/model openai/gpt-4o
/model anthropic/claude-3-5-sonnet-20241022
/model reasoning high
/model reasoning disable
Reasoning levels:
  • low - Faster responses, less thorough
  • medium - Balanced speed and quality
  • high - Slower responses, more thorough
  • disable - No extended reasoning

​
/config

Show or modify agent configuration. 
# Show full configuration
/config

# Configure tools interactively
/config tools
The /config tools subcommand opens an interactive checkbox interface where you can enable or disable tools for the current agent.

​
/mode

Switch between tool approval modes. 
# Interactive mode selector
/mode

# Switch to safe mode (require approval for every tool)
/mode safe

# Switch to yolo mode (auto-approve all tools)
/mode yolo

# Auto-approve specific command prefix
/mode allow <command-prefix>

# Remove auto-approval for command prefix
/mode disallow <command-prefix>
Examples: 
/mode safe
/mode yolo
/mode allow git status
/mode allow npm test
/mode disallow git push
Approval modes:
  • safe - Require approval for every tool call (default)
  • yolo - Auto-approve all tool calls
You can also selectively auto-approve specific command prefixes while staying in safe mode. Commands added with /mode allow persist only for the current session.

​
Information and monitoring

​
/tools

List all tools available to the current agent, organized by category. 
/tools
Displays:
  • Tools grouped by category (File System, Git, Web, etc.)
  • Total tool count
  • Web search provider (if configured)

​
/skills

List all available skills. 
/skills
Displays skills from:
  • Built-in - Skills shipped with Jazz
  • Global - Skills in ~/.jazz/skills/
  • Agents - Skills in ~/.agents/skills/
  • Local - Skills in ./skills/
Skills are packaged expertise that agents can load on demand for specialized tasks.

​
/workflows

List available workflows. 
/workflows
Displays:
  • Workflow name and description
  • Schedule (if configured)
  • Assigned agent (if configured)
  • Workflows grouped by source (local, global, built-in)
You can also ask the agent to create a workflow by typing /workflows create.

​
/context

Show context window usage and token breakdown. 
/context
Displays:
  • Visual grid showing context usage
  • Model and provider
  • Total tokens used vs. available
  • Breakdown by category:
    • System prompt
    • System tools
    • Skills
    • Messages
    • Free space
    • Autocompact buffer (16.5% reserved)
Token estimates are approximate and based on the 4-character-per-token heuristic. Actual token counts may vary by model.

​
/cost

Show token usage and estimated cost for the current conversation. 
/cost
Displays:
  • Input tokens used
  • Output tokens used
  • Total tokens
  • Pricing per million tokens (from models.dev)
  • Estimated cost breakdown (input, output, total)
The cost is calculated based on the accumulated token usage since the conversation started (or since the last /new command).

​
/stats

Show session statistics and usage summary. 
/stats
Displays:
  • Current agent name and ID
  • Model and provider
  • Reasoning effort level
  • Tool count
  • Working directory
  • Session duration
  • Message count
  • Token usage (input, output, total)
  • Estimated cost

​
/mcp

Show MCP server status and connections. 
/mcp
Displays for each configured MCP server:
  • Connection status (connected/disconnected)
  • Enabled/disabled state
  • Transport type (stdio/http)
  • Command or URL
MCP servers are configured in ~/.agents/mcp.json.

​
Context optimization

​
/compact

Summarize conversation history to save tokens. 
/compact
This command:
  • Requires at least 5 messages in history
  • Keeps the system prompt
  • Summarizes all other messages into a single high-density summary
  • Preserves key information while reducing token usage
  • Uses the agent’s LLM to generate the summary
The compaction process:
  1. Reads all messages from conversation history
  2. Analyzes content and extracts key information
  3. Generates a high-density summary
  4. Replaces history with system prompt + summary
Use this when you’re running low on context space but want to continue the conversation.
Jazz automatically compacts context when usage reaches the autocompact buffer threshold (83.5% of context window).

​
Utilities

​
/copy

Copy the last agent response to clipboard. 
/copy
Requires pbcopy (macOS) or equivalent clipboard tool.

​
/help

Show available commands with descriptions. 
/help
Displays a quick reference of all slash commands.

​
Command conventions

  • All commands start with /
  • Commands are case-insensitive (/NEW works the same as /new)
  • Arguments are space-separated
  • Multi-word arguments are joined (e.g., /switch my agent name)
  • Unknown commands show an error and suggest /help

​
Tips

  • Use /context to monitor token usage before running expensive operations
  • Use /compact when approaching context limits to continue long conversations
  • Use /fork to explore alternative approaches without losing your main conversation
  • Use /mode allow <prefix> to auto-approve repetitive safe commands
  • Use /stats to track session duration and costs
  • Use /switch to change agents mid-conversation while preserving context

Build docs developers (and LLMs) love

Get started for freeTalk to us