Admin Commands

Quick Reference

Status & Information

/status

Display a comprehensive overview of the agent's current state.

Behavior: Returns a status message containing all of the following information:

• Active conversations : number of currently tracked chat sessions
• Provider & model : the active LLM provider (e.g. OpenAI, Anthropic) and model name
• Max iterations : the current agentic loop limit (tool-call cycles per request)
• DM policy : who can message the agent in direct messages
• Group policy : who can interact with the agent in group chats
• Pause state : whether the agent is currently paused
• Mention settings : whether the agent responds to @mentions in groups

Example:

/status

Agent Status
---
Active conversations: 3
Provider: openai
Model: gpt-4o
Max iterations: 10
DM policy: allowlist
Group policy: open
Paused: No
Respond to mentions: Yes

/help

Display the full list of available admin commands with their usage syntax.

Behavior: Returns a formatted list of every admin command recognized by the agent, grouped by category. Each entry includes the command name, expected arguments, and a short description. This is the same set of commands documented on this page.

Example:

/help

/ping

Quick health check to verify the agent process is running and responsive.

Behavior: The agent immediately replies with Pong!. No LLM call is made, this is a direct handler response. If you do not receive a reply, the agent process is likely down or unreachable.

Example:

/ping

Pong!

Model & Loop Configuration

/model

View the current LLM model or hot-swap to a different model at runtime without restarting the agent.

Behavior:

• Without arguments: Displays the currently active LLM model name and provider.
• With a model name: Immediately switches the agent to use the specified model for all subsequent requests. The change takes effect on the next incoming message. No restart required. The model name must be a valid identifier supported by the configured provider.

Examples:

# View current model
/model

# Switch to GPT-4o
/model gpt-4o

# Switch to Claude Sonnet
/model claude-sonnet-4-20250514

# Switch to a local model
/model llama-3.1-70b

/loop

View or set the maximum number of agentic iterations (tool-call cycles) the agent is allowed per single user request.

Behavior:

• Without arguments: Displays the current max iteration limit.
• With a number: Sets the maximum tool-call cycles. Each cycle is one round of the agent calling a tool and receiving its output. Higher values allow the agent to complete more complex multi-step tasks but consume more tokens and time. Values outside the 1-50 range are rejected.

Examples:

# View current loop limit
/loop

# Set to 10 iterations (good default)
/loop 10

# Set to maximum for complex tasks
/loop 50

# Set to 1 (single tool call, no chaining)
/loop 1

Access Policy

/policy

View or change who is allowed to interact with the agent in DMs and group chats.

Behavior:

• Without arguments: Shows the current DM policy and group policy side by side.
• With arguments: Changes the specified policy at runtime. The change is immediate.

DM policy values:

Group policy values:

Examples:

# View current policies
/policy

# Restrict DMs to allowlisted users
/policy dm allowlist

# Open group access to everyone
/policy group open

# Disable all DM interactions
/policy dm disabled

# Disable group interactions
/policy group disabled

Pause & Resume

/pause

Pause the agent so it stops responding to all non-admin messages.

Behavior: Immediately pauses the agent. While paused:

• The agent will not respond to any user messages, mentions, or triggers.
• Admin commands (/resume, /status, etc.) continue to work.
• Incoming messages are silently ignored and are not queued for later processing.
• The paused state is reflected in /status output.

Example:

/pause

Agent paused. Use /resume to continue.

/resume

Resume the agent after it has been paused.

Behavior: Resumes normal operation. The agent will begin responding to messages again immediately. If the agent is not currently paused, this command has no effect.

Example:

/resume

Agent resumed.

Wallet

/wallet

Display the agent's TON wallet address and current balance.

Behavior: Queries the TON blockchain for the agent's wallet and returns:

• Wallet address : the full TON address configured for the agent
• Current balance : the TON balance at the time of the query

The wallet must be configured in the agent's settings for this command to work. If no wallet is configured, an error message is returned.

Example:

/wallet

Wallet: EQD...abc
Balance: 12.45 TON

Trading Strategy

/strategy

View or adjust the automated NFT trading thresholds expressed as a percentage of the collection floor price.

Behavior:

• Without arguments: Displays the current buy and sell thresholds as percentages of the collection floor price.
• /strategy buy <percent>: Sets the maximum price the agent will pay to buy an NFT. Value must be between 50 and 150. For example, /strategy buy 90 means the agent will only buy items priced at or below 90% of the floor.
• /strategy sell <percent>: Sets the minimum price at which the agent will list an NFT for sale. Value must be between 100 and 200. For example, /strategy sell 130 means the agent will list items at 130% of floor price or higher.

Examples:

# View current strategy
/strategy

# Buy at up to 90% of floor price
/strategy buy 90

# Sell at 130% of floor price or above
/strategy sell 130

# Conservative: buy low, sell high
/strategy buy 70
/strategy sell 180

Conversation Management

/clear

Clear the conversation history for a chat, resetting the agent's memory of that conversation.

Behavior:

• Without arguments: Clears the conversation history for the current chat (the chat where the command was sent). The agent will have no memory of previous messages in this conversation.
• With a chat ID: Clears the conversation history for the specified chat. Useful for remotely clearing a group conversation from a DM with the bot.

This only affects the agent's stored conversation context. It does not delete Telegram messages.

Examples:

# Clear current conversation
/clear

# Clear a specific chat by ID
/clear -1001234567890

/task

Give the agent a free-form task to execute. The agent interprets the instruction and decides how and when to carry it out.

Behavior: The agent receives the task description and autonomously determines:

• Schedule : when to execute (immediately, at a specific time, or on a recurring basis)
• Payload type : what tools and actions to use
• Execution plan : the sequence of steps to accomplish the task

Tasks are persisted and survive agent restarts. Use /status to see active scheduled tasks.

Examples:

# One-time alert
/task Check TON price and alert me if above $5

# Recurring monitoring
/task Every hour, check my wallet balance and log it

# Complex multi-step task
/task Find the cheapest NFT in the Teleton collection and buy it if under 2 TON

/boot

Run the agent bootstrap sequence for first-time setup.

Behavior: Loads and executes the BOOTSTRAP.md template, which guides the agent through an initial setup conversation. This typically includes:

• Setting up the agent's personality and system prompt
• Configuring initial preferences
• Walking through available features

You can re-run /boot at any time to repeat the setup conversation.

Example:

/boot

Debugging

/verbose

Toggle verbose debug logging on or off.

Behavior: Toggles verbose mode. When enabled, the agent outputs detailed debug information including:

• Raw tool call payloads and responses
• LLM prompt construction details
• Token usage per request
• Timing information for each agentic loop iteration

Run the command again to disable verbose logging. The current state is shown in /status output.

Example:

# Enable verbose mode
/verbose

# Disable verbose mode (run again)
/verbose

Tool RAG

Tool RAG (Retrieval-Augmented Generation) uses semantic search to select the most relevant tools for each request instead of sending the full tool list to the LLM. This reduces token usage and improves response quality when the agent has many tools loaded.

/rag

Toggle Tool RAG on or off, or manage its settings.

/rag : Toggle

Behavior: Toggles Tool RAG on or off. When enabled, the agent performs semantic search over the tool index before each LLM call to select only the most relevant tools. When disabled, all tools are sent to the LLM on every request.

/rag status : View status

Behavior: Displays a detailed status report including:

• Enabled/disabled : whether Tool RAG is currently active
• Indexed status : whether the tool index has been built
• Top-K value : the maximum number of tools returned per query
• Total tool count : how many tools are registered in the index
• Always-include patterns : tool name patterns that bypass RAG filtering and are always included

/rag topk : Set top-K

Behavior: Sets the maximum number of tools retrieved from the index per LLM call. A higher value sends more tools to the LLM (more capable but more tokens). A lower value is cheaper but may miss relevant tools.

Examples:

# Toggle Tool RAG on
/rag

# Check current RAG configuration
/rag status

# Set top-K to 20 tools per request
/rag topk 20

# Set top-K to maximum for complex requests
/rag topk 200

Module Permissions

/modules

Manage per-group permissions for tool modules. Modules are logical groupings of related tools (e.g. "telegram", "ton", "dex").

/modules : List all modules

Behavior: Lists every registered module with:

• Module name
• Tool count : number of tools in the module
• Permission level : current access level for this group
• Protected status : whether the module's level can be changed

/modules set : Set permission level

Behavior: Changes the permission level of a module for the current group. Protected modules cannot be modified.

/modules info : Module details

Behavior: Shows detailed information about a specific module including the full list of its tools, each tool's scope, and the module's current permission level.

/modules reset : Reset permissions

Behavior: Resets module permissions back to the default open level. If a module name is given, only that module is reset. If no argument is provided, all modules in the current group are reset to open.

Examples:

# List all modules and their status
/modules

# Restrict DEX trading tools to admins only
/modules set dex admin

# Disable TON blockchain tools in this group
/modules set ton disabled

# View details of the telegram module
/modules info telegram

# Reset the dex module to open
/modules reset dex

# Reset all modules to open
/modules reset

Plugin Secrets

/plugin

Manage secrets (API keys, tokens, credentials) for installed plugins. Secrets are stored securely and never displayed in plain text after being set.

/plugin set : Set a secret

Behavior: Stores a secret key-value pair for the specified plugin. If the key already exists, its value is overwritten. The plugin may need to be reloaded or the agent restarted for the new secret to take effect.

/plugin unset : Remove a secret

Behavior: Removes a previously set secret from the named plugin. If the key does not exist, the command has no effect.

/plugin keys : List configured keys

Behavior: Lists all secret key names configured for the specified plugin. For security, the actual values are never shown, only the key names are displayed.

Examples:

# Set an API key for the "weather" plugin
/plugin set weather API_KEY sk-abc123xyz

# Set a webhook URL for a notifications plugin
/plugin set notifications WEBHOOK_URL https://hooks.example.com/notify

# Remove a secret
/plugin unset weather API_KEY

# List all keys configured for a plugin
/plugin keys weather

System

/stop

Emergency shutdown of the agent process.

Behavior: Initiates an emergency shutdown sequence:

1. The agent acknowledges the command.
2. After a 1-second delay, a SIGTERM signal is sent to the agent process.
3. The process terminates gracefully, closing open connections and flushing pending writes.

Example:

/stop

Shutting down in 1 second...