Tools Overview

Tools give the AI agent the ability to act: run commands, read/write files, search the web, manage memory, and interact with hardware. Each tool implements the hu_tool_t vtable.

Tool vtable interface

Method	Description
execute	Run the tool with JSON args, return hu_tool_result_t
name	Tool identifier (e.g. shell, file_read)
description	Human-readable description for the LLM
parameters_json	JSON Schema for the tool’s parameters
deinit	Optional cleanup

Available tools (87+)

Tool	Description
shell	Execute shell commands (allowlisted, sandboxed)
file_read	Read file contents from workspace
file_write	Write content to a file
file_edit	Search-and-replace or structured edits
file_append	Append to a file
git	Git operations (status, diff, commit)
web_search	Web search (DuckDuckGo by default)
web_fetch	Fetch URL content (HTTPS only, size limited)
http_request	HTTP/HTTPS requests
browser	Browser actions: open, read, screenshot, click, type, scroll
browser_open	Open approved HTTPS URLs in default browser
image	Image operations (load, process)
screenshot	Capture screenshots
memory_store	Store a fact or memory entry
memory_recall	Recall memories by query
memory_list	List stored memories
memory_forget	Remove a memory by ID
message	Send a message (e.g. to a channel)
delegate	Delegate tasks to sub-agents (routes to claude_code)
spawn	Spawn child processes
cron_add	Add a cron job
cron_list	List cron jobs
cron_remove	Remove a cron job
cron_run	Run a cron job immediately
cron_runs	List recent cron execution history
cron_update	Update a cron job
claude_code	Delegate coding tasks to Claude Code CLI (sub-agent)
composio	Composio integration (list tools, execute actions)
hardware_memory	Read/write hardware memory via probe-rs or serial
schedule	Scheduled tasks (create, list, cancel, pause)
schema	Validate JSON schemas, list tool schemas
pushover	Push notifications via Pushover
hardware_info	System hardware info
i2c	I2C bus read/write
spi	SPI bus operations

Tool filtering

Enable or disable tools via config:

{
  "tools": {
    "enabled_tools": ["shell", "file_read", "memory_store"],
    "disabled_tools": ["browser_open", "spawn"]
  }
}

• enabled_tools: If non-empty, only these tools are available
• disabled_tools: These tools are always disabled
• Empty enabled_tools = all tools except disabled_tools

Security

• Shell: Commands are allowlisted; execution is sandboxed (Landlock, Firejail, Bubblewrap, or Docker when available)
• File operations: Restricted to workspace by default
• HTTP: HTTPS required; HTTP URLs are rejected
• browser_open: Only allowlisted domains

Workspace scoping

Most file and shell tools respect workspace_dir. The agent cannot access paths outside the workspace unless explicitly allowed.