Tools

Tools are what turn the model into an agent. Instead of only producing text, the Pathvela agent can act — read a file, run a command, search your project, fetch a page — observe the result, and keep going. It decides which tools to use for a task; you just describe the goal.

This page is the catalog of what it can do. Which tools are available depends on where the agent runs and how much access you've granted — see Permissions & safety and, for the browser, Browser vs Node.

Files

Tool	What it does
Read	Open and read a file.
Write	Create or overwrite a file.
Edit / Multi-edit	Make precise in-place changes to one or many parts of a file.
Patch	Apply a diff-style patch.
List / Glob	List a directory or find files by pattern.

Shell and code

Tool	What it does
Bash / Terminal	Run shell commands and read their output — build, test, run scripts.
LSP	Use language-server intelligence (definitions, references, diagnostics) for code-aware work.

Risky shell commands are gated — see Permissions & safety.

Search

Tool	What it does
Search (grep)	Search file contents across a project.
Session search	Look back through the current and past conversations.

Web and browser

Tool	What it does
Web fetch	Pull in a page or docs URL a task needs.
Web search	Search the web for current information.
Browser	Drive a real browser to navigate and interact with pages.

Task management

Tool	What it does
Todo	Track a multi-step task as a checklist so long jobs stay coherent.
Background tasks	Kick off long-running work and stream its output.
Batch	Run several tool calls together.
Question	Ask you a clarifying question when it needs a decision.

Memory and skills

Tool	What it does
Memory	Remember and recall facts across sessions — see Memory.
Skills	Load and apply reusable know-how — see Skills.

Integrations

The agent can also reach beyond your machine when configured:

MCP servers add external tools through the Model Context Protocol; they appear as mcp__<server>__<tool>.
Connected services (like email and scheduling) become tools when you've set them up.

Read-only vs full access

Not every session gets the write tools:

A headless lily run works in a read-only profile — it can read, search, and run read-only commands, but won't change files, because no one is present to approve edits.
An interactive session (the lily terminal UI, or the browser via auto mode) gets the full tool set, with risky actions gated by approval.

See Permissions & safety for exactly how that gate works.