Built-in Tools

LeafMesh ships a set of built-in tools across several categories, all automatically registered when LeafMesh loads. Every LLM agent is additionally pre-injected — with no configuration — with schedule_followup (see Self-Scheduled Follow-ups), and, when the corresponding config is present, recall_memory (memory) and query_knowledge (knowledge).

Math Tools

`calculator`

Safe mathematical expression evaluation.

Parameter	Type	Description
`expression`	string	Math expression (max 500 chars)

Supports: +, -, *, /, //, %, unary operators.

`random_number`

Generate random integers.

Parameter	Type	Default	Description
`min_val`	integer	1	Minimum value
`max_val`	integer	100	Maximum value
`count`	integer	1	How many (max 1000)

Time Tools

`current_time`

Get current date and time.

Parameter	Type	Default	Description
`timezone`	string	"UTC"	Timezone name
`time_format`	string	"iso"	"iso", "human", "timestamp", or strftime

`time_difference`

Calculate time between two ISO timestamps.

Parameter	Type	Description
`start_time`	string	ISO format start
`end_time`	string	ISO format end
`unit`	string	"seconds", "minutes", "hours", "days"

Text Tools

`text_analyzer`

Analyze text properties: character count, word count, sentence count, etc.

Parameter	Type	Default	Description
`text`	string	—	Text to analyze
`analysis_type`	string	"all"	"basic", "detailed", or "all"

`text_formatter`

Format text: uppercase, lowercase, title, reverse, capitalize, remove_spaces, add_prefix, add_suffix, wrap_lines.

Parameter	Type	Description
`text`	string	Text to format
`format_type`	string	Formatting operation
`options`	object	Additional options

Web Tools

`web_request`

HTTP requests (GET, POST, PUT, DELETE). Requires confirmation. SSRF-protected.

Parameter	Type	Default	Description
`url`	string	—	Target URL
`method`	string	"GET"	HTTP method
`headers`	object	{}	Request headers
`data`	object	{}	Request body
`timeout`	number	10	Timeout seconds

Data Tools

`json_formatter`

Format and validate JSON: "pretty", "compact", "validate".

`data_converter`

Convert between formats (json ↔ text).

Reasoning Tools

`chain_of_thought`

Step-by-step reasoning through problems. Styles: analytical, creative, logical, intuitive, systematic, exploratory.

`metacognitive_reflection`

Evaluate reasoning quality: strengths, weaknesses, bias check, improvement suggestions.

thinking: true on an LLM agent gives it these two tools plus apply_edits. They are text only — none of them runs code.

Sandbox Tool

`sandbox_exec`

Runs Python or shell in a contained sandbox and returns {stdout, stderr, exit_code, error, truncated}. It is given automatically to an LLM agent with sandbox: set — you do not list it under tools: — and only when the server can contain code. No network unless you allow hosts. See LLM Agents → Running Code.

Parameter	Type	Default	Description
`code`	string	required	The complete program
`language`	string	`python`	`python` or `shell`

System Tools

`system_info`

Platform info, Python version, current time, working directory.

Board & Experience Tools

These are given automatically — you do not list them under tools:.

An agent on a shared board gets view_board, add_task, pull_task, update_checklist, update_task and finish_task to take, track and close work.

An agent with experience gets remember and recall_experience to keep and recall what it learned from its own work.

An agent with a playbook gets playbook to load one of its responsibilities when a request needs it.

Assigning Tools to Agents

agents:
  research_agent:
    name: "research_agent"
    tools: ["web_request", "calculator", "current_time"]
    tool_choice: "auto"
    max_tool_calls_per_message: 5
    allow_parallel_tool_calls: true

When an agent requests several independent tools in one step, LeafMesh runs them concurrently (order-preserving). The concurrency ceiling is set by the limits.max_parallel_tool_calls config field (default 8). Set allow_parallel_tool_calls: false on an agent to force its tools to run one at a time.

Every LLM agent is also pre-injected with schedule_followup, which lets it resume its own work later in the same session — no configuration required.

Tool Execution Flow

Tool schemas sent to LLM alongside the prompt
LLM decides when to call tools (with tool_choice: "auto")
LeafMesh validates permissions and runs the tool
Results returned to LLM for follow-up response
Loop continues until no more tool calls or max_tool_calls_per_message reached

Next Steps

Custom Tools — Register your own tools
Tool Permissions — Access control details
Tool System API — API reference

LeafMesh — 17 built-in tools