Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/jorgeurtubiam-ship-it/Gulin_ia/llms.txt

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

GuLiN ships a set of built-in tools that the AI agent can invoke autonomously as part of answering your requests or completing multi-step tasks. Rather than just suggesting what you should do, the agent can act directly — running a command, reading a web page, editing a file, or saving a memory — all within a single conversation turn. Every action that modifies state requires your explicit approval, keeping you in control at every step.
Approval gates: All tools that write, execute, click, or type — term_run_command, web_click, web_type, and all file-write tools — require your explicit manual confirmation in the chat before they run. GuLiN will never execute a destructive or state-changing action without first presenting it to you for approval. Read-only tools (web_read_page, capture_screenshot, brain_list, brain_search, apimanager_list) run without a confirmation prompt.
If the agent is heading in the wrong direction mid-execution, use Rewind & Merge: send a new corrective message at any time. GuLiN atomically cancels the current task, terminates all pending tool calls, rewinds the history, and merges your new instruction. See the AI Overview page for details.

​
Tool Reference

term_run_command

Execute a shell command in any open terminal widget. The agent proposes the command and waits for your approval before running it. Output is captured and fed back into the conversation automatically.

web_read_page

Read the full text content of any web page open in a GuLiN browser tab. Ideal for summarizing documentation, following tutorials, or extracting data without leaving the terminal.

web_click

Click a DOM element on an open web page — buttons, links, checkboxes, and dropdowns. Requires your approval before any click is sent.

web_type

Type text into a form field on an open web page — search boxes, input fields, and text areas. Requires your approval, so the agent never submits forms without permission.

capture_screenshot

Capture the visual state of any GuLiN widget as an image. The AI uses this to inspect rendering output, UI bugs, or dashboard states that are not representable as text.

brain_update / brain_list / brain_search

The three Gulin Brain memory tools. Save new knowledge, list all stored memories, or run a deep semantic vector search. See the Memory page for full details.

apimanager_list / apimanager_call / apimanager_register / apimanager_delete

Manage and invoke stored API connections. List registered APIs, call an endpoint, register a new connection, or remove one from the secure API Manager vault.

File Tools

Read, write, edit, and create files in your workspace. The agent can read any file to gather context (no approval needed) and proposes specific writes or edits that you confirm before they are applied.

​
Approval Requirements

ToolCategoryApproval Required
term_run_commandTerminal✅ Yes
web_read_pageWeb❌ No
web_clickWeb✅ Yes
web_typeWeb✅ Yes
capture_screenshotSystem❌ No
brain_updateMemory❌ No
brain_listMemory❌ No
brain_searchMemory❌ No
apimanager_listAPI Manager❌ No
apimanager_callAPI Manager✅ Yes
apimanager_registerAPI Manager✅ Yes
apimanager_deleteAPI Manager✅ Yes
File readFiles❌ No
File write / edit / createFiles✅ Yes

​
Terminal Tool: term_run_command

The terminal tool allows the agent to run any shell command in any open terminal widget inside GuLiN. The workflow is always:
  1. The agent decides a command is needed and proposes it in the chat.
  2. An approval prompt appears — you can inspect the exact command before it runs.
  3. You approve or reject.
  4. On approval, the command runs in the target terminal and its output is captured back into the AI’s context automatically.
This means the agent can orchestrate multi-step workflows — building a project, running tests, checking logs, fixing errors — without you having to manually copy-paste each command. 
# Example: You ask "Run the test suite and summarize the failures"
# The agent proposes:
npm test -- --reporter=verbose

# After your approval, it runs the command, reads the output,
# and summarizes the failed tests directly in the chat.

​
Web Tools: web_read_page, web_click, web_type

GuLiN’s integrated browser is a first-class agentic surface. The AI can navigate and interact with web pages open in browser tabs inside GuLiN.
  • web_read_page: Reads the full text of any open tab — articles, documentation, dashboards, or API responses — and uses it as context. No approval needed since it is read-only.
  • web_click: Clicks DOM elements. Useful for navigating paginated content, expanding accordions, or interacting with web apps. Always requires approval.
  • web_type: Types into input fields, search boxes, or forms. Always requires approval so the agent never submits data on your behalf without confirmation.
The User-in-the-Loop layer for web actions is non-negotiable: GuLiN will never send a form submission, click a confirm button, or type into a password field without first showing you exactly what it is about to do.

​
Screenshot Tool: capture_screenshot

The capture_screenshot tool gives the AI literal vision over your terminal workspace. It captures the current rendered state of any GuLiN widget as an image. This is particularly useful for:
  • Debugging visual rendering errors in dashboards or custom widgets.
  • Verifying that a CSS change or layout adjustment looks correct.
  • Inspecting the output of a terminal command that produces graphical or tabular output that is hard to represent as plain text.
The AI uses the screenshot to identify visual issues and can then propose targeted fixes.

​
API Manager Tools

The API Manager is GuLiN’s secure vault for managing external service connections — REST APIs, cloud credentials, and custom endpoints. The AI can interact with the vault through four tools:
  • apimanager_list: List all registered API connections (names only, not secrets). Read-only, no approval needed.
  • apimanager_call: Invoke a registered API endpoint. Requires approval so you can inspect the request before it is sent.
  • apimanager_register: Add a new API connection to the vault. Requires approval.
  • apimanager_delete: Remove a connection from the vault. Requires approval.

​
File Tools

The agent can interact with files in your workspace as a natural part of any task — reading configs to understand the project, writing new files, or making targeted edits.
  • Read: The agent reads any file to gather context. This is automatic and requires no approval.
  • Write / Create: Propose creating a new file with full content. You review and approve before the file is written to disk.
  • Edit: Propose a targeted change to an existing file. The agent shows you the before/after diff for approval.
File operations work seamlessly with the rest of the agentic loop: the agent can read a config, run a command based on what it finds, observe the output, and edit the file to fix a problem — all in one conversation turn, with approvals at each write step.

Build docs developers (and LLMs) love

Get started for freeTalk to us