Built-in Tools

Enable KorinAI built-in tools to extend your assistant without client secrets—configure toggles per agent/environment, then invoke actions in chat with optional confirmation and logged outputs.

Overview

Built-in tools extend the assistant’s capabilities without additional code. Enable only what you need to augment search, retrieval, or computation for specific user requests.

Environment scope: toggles are saved per agent/environment; changes apply to that agent’s chat sessions.

When to enable

• Routine tasks you want the AI to perform on demand (search, generate images/slides/docs, quick scripts).
• You want server-side execution without exposing client keys.

When to keep disabled

• Actions with side effects you’re not ready to audit or confirm.
• Expensive operations you only want via Custom Tools with stricter controls.

Key Features

• Context-aware execution with optional user confirmation.
• Composable tooling (e.g., Web Search ➜ Web Crawl).
• Centralized control via the dashboard; no client-side keys exposed.

Security & Execution

• Server-side execution: tools run on the backend; API keys (if required) are kept off the client.
• Confirmation recommended for operations with network or data-side effects.
• Logs appear in the chat stream for transparency during testing and review.

Confirmation strategy

• Prefer confirmation for networked or write-like operations (posting data, sending emails, persistent changes).
• Skip confirmation for read-only, low-risk tasks (formatting, summarization, non-network transforms).

Usage Guide

1. Enable desired tools in Dashboard → Settings → Built-in Tools.
2. In chat, request an action covered by the enabled tool.
3. Approve confirmation prompts if configured.
4. Review tool invocation and results in the message list.

Tool Catalog

Detailed Tool Guides

Web Research

Web Research performs web search and, when needed, multi‑page crawling to gather current, public information from reputable sources.

Typical prompts

• "Summarize recent techniques for vector search; include sources and dates."
• "Compare provider X and Y pricing; cite pages and call out hidden fees."
• "Find docs and tutorials for deploying Next.js on Vercel; include official links only."
• "Aggregate 3 credible sources on RAG evaluation best practices; summarize pros/cons."

Example

• "Give me a 5‑bullet brief on the latest LangChain features. Include official docs and one blog analysis."

Best for

• Market scans and competitive intel with citations and timestamps.
• Engineering/documentation reviews with direct references.
• Quick executive briefs that link to primary sources.

Capabilities

• Web search, targeted crawl of selected results, citation extraction.

Inputs/Outputs

• Input: plain-text query and optional focus terms.
• Output: summary with links and minimal metadata (title/url/date when available).

Limits & tips

• Rate limits and site policies may reduce coverage; prefer precise queries.
• Use confirmation if downstream actions depend on external content.

Image Generation

Image Generation creates visuals from textual direction—great for mockups, icons, and lightweight diagrams.

Typical prompts

• "Create a minimalist icon for a finance dashboard in 2 variants (outline/filled)."
• "Generate a header illustration: night skyline, neon palette, 3:1 aspect ratio."
• "Produce a small architecture sketch: client → API → DB (monochrome, simple)."

Example

• "Design a square hero image for a documentation site: abstract shapes, cool palette, subtle gradients."

Best for

• Marketing visuals, UI placeholders, lightweight concept art, diagrams for docs.
• Early-stage ideation where speed matters more than pixel-perfect brand fidelity.

Capabilities

• Text-to-image with style, size/aspect, and simple layout hints.

Inputs/Outputs

• Input: prompt text; optional guidance (style, aspect).
• Output: image file(s) with preview in chat; downloadable asset links.

Limits & tips

• Not a brand system; expect variability between generations.
• No confirmation usually needed; large batch generations may require.

Slide Generation

Slide Generation creates and edits slide decks—insert or delete slides, restructure sections, and update content programmatically.

Typical prompts

• "Create a 6‑slide deck on RAG with agenda, overview, architecture, pros/cons, use cases, references."
• "Insert a new slide after slide 3 titled 'Roadmap Q4' with 4 bullet points."
• "Summarize this doc into 8 slides; each slide gets a title + 3 bullets."

Example

• "Turn this meeting summary into a deck: 1) agenda 2) decisions 3) risks 4) next steps."

Best for

• Pitch drafts, internal briefings, training decks, recurring status updates.
• Rapid iteration on outline/structure before design polish.

Capabilities

• Generate new decks, insert/move/delete slides, rewrite slide content.

Inputs/Outputs

• Input: outline or source text to transform; edit commands (insert, update).
• Output: slide deck artifact with download link and diff summary in chat.

Limits & tips

• Formatting is functional; final polishing may require a design pass.
• Use confirmation for bulk edits on existing decks.

Note Generation

Note Generation produces structured notes and summaries (sections, bullets, action items) and supports insert/replace/view operations for ongoing documents.

Typical prompts

• "Summarize yesterday’s meeting into decisions, action items, open questions."
• "Append these 3 tasks to the 'Action items' section with owners and due dates."
• "Refactor this long note: split into sections 'Context / Findings / Open Questions'."

Example

• "Create a crisp post‑mortem from this chat transcript: timeline, root cause, remediation, follow‑ups."

Best for

• Meeting minutes, research digests, SOP drafts, onboarding docs.
• Consolidating multi‑thread discussions into a single source of truth.

Capabilities

• Summarize, outline, and restructure notes; insert new sections; maintain action lists.

Inputs/Outputs

• Input: raw text, meeting summaries, specs; simple edit directives.
• Output: updated note contents with clear sections and optional action items.

Limits & tips

• Ensure source text is provided or referenced from knowledge; otherwise, results may be generic.
• Confirmation optional; enable when notes are shared broadly.

Document & PDF Generation

Generates professional HTML documents optimized for PDF export (CVs, reports, invoices, certificates) with clean layout and print‑friendly styles.

Typical prompts

• "Generate an invoice PDF with company, client, items, totals, due date."
• "Create a certificate template (A4, margins 1in, serif headings, logo placeholder)."
• "Build a simple multi‑page report with TOC, headers/footers, and page numbers."

Example

• "Produce a one‑page CV with two columns (experience, skills). Use subtle divider lines and good spacing."

Best for

• Invoices, reports, certificates, formal documents needing reliable PDF output.
• Any document where typography, margins, and repeatable layout matter.

Capabilities

• Structured HTML generation with CSS for PDF; header/footer, TOC, pagination support.

Inputs/Outputs

• Input: content sections and layout requirements.
• Output: HTML artifact and exported PDF.

Limits & tips

• Complex brand layouts may need manual CSS adjustments.
• Use confirmation when generating billable documents (e.g., invoices).

Computer

Computer assists with full‑stack coding, data analysis, conversions/optimizations, and scraping & capture—running code in a sandbox with previews and artifacts.

Typical prompts

• "Write a script to parse CSV and compute cohort retention; include a quick chart."
• "Spin up a minimal Next.js app with a health endpoint and a landing page."
• "Fetch a page, extract <h2> headings, and save them as JSON + a CSV backup."

Best for

• Full‑stack scaffolding (Next.js app, API routes, UI components) and rapid prototyping.
• Data processing (CSV/JSON parsing, stats, charts) with visual artifacts.
• File/image/document conversions (optimize images, merge PDFs, transcode text to HTML/PDF).
• Web scraping & capture (fetch pages, extract elements, save screenshots/PDFs).
• Spinning up a local web preview to validate flows end‑to‑end inside the sandbox.

Examples

• Create a Next.js app

  1. "Create a minimal Next.js app with a /api/health endpoint and a landing page."
  2. Tool generates scaffold (package.json, next.config, app/route, API route).
  3. Runs the dev server, returns a preview URL and logs.

• Quick scraper

  1. "Fetch https://example.com and extract all <h2> headings. Save as JSON."
  2. Tool writes a small Node script using fetch/Cheerio.
  3. Runs it, stores headings.json, streams the JSON back.

• CSV analysis

  1. "Load sales.csv, compute monthly totals and render a bar chart."
  2. Tool installs needed libs, creates a small analysis script.
  3. Outputs a PNG/SVG chart and a brief summary.

Capabilities

• Code execution in an isolated environment; file IO, simple web preview, artifact storage.

Inputs/Outputs (high level)

• Input: natural language task; the tool writes/executes code to satisfy it.
• Output: logs, files (JSON/CSV/PNG), optional preview URL.

Limits & tips

• Sandbox timeouts and resource limits apply; keep tasks scoped.
• Use confirmation for network calls, data writes, or destructive operations.

Examples

Research — Web Search + Web Crawl

1. Enable both tools.
2. Ask: "Summarize recent posts on retrieval-augmented generation with sources."
3. Confirm any prompts if required.
4. The result should include links to collected pages.

Quick Calculation — Python Execution

# Example snippet a Python execution might run
values = [1.2, 3.4, 5.6]
mean = sum(values)/len(values)
print({"mean": mean})

Troubleshooting

• "Tool not enabled": Confirm toggle in Built-in Tools.
• Missing confirmation prompts: Check the tool’s confirmation setting.
• Permission errors: Verify user/account policy and API keys.
• Rate/credit errors: Review Dashboard → Credits and model limits.
• Not visible in chat: Ensure you toggled the correct agent/environment; refresh the chat page after updates.

Best Practices

• Enable a minimal set of tools per environment.
• Require confirmation for external/networked operations.
• Log and audit tool invocations on the server.
• Rotate API keys regularly and scope them minimally.

Related Topics

• Feeding Knowledge
• Agent Orchestration
• Custom Tools
• Prompt Templates