Overview

Product video

Key features

• Answers with sources — every answer shows which document and section it came from, users can rate it 👍 or 👎, and old chats are searchable
• Easy to chat with — suggested follow-up questions, regenerate or edit a question, attach a picture, talk instead of typing, have answers read aloud, and see numbers as charts
• Works on phones — the chat fits any screen, has a dark mode, switches between Thai and English, and can be installed as an app, even one app per assistant
• Assistants for every team — build one per department, let AI draft the instructions and starter questions, give it a picture, and tune how it answers
• Assistants that work together — a main assistant can hand questions to specialists like HR or IT and bring their answers back
• Try before you launch — unfinished assistants are only visible to their creator and admins; when ready, open them to everyone, some departments or a few people
• Test answer quality — write test questions with the answers you expect, or pull them from real chats, and let an AI score each one; you get a warning before going live if fewer than 80% pass
• Benchmark page — see how satisfied users are with each assistant, how often it answers correctly and how often it had to switch to a backup model, and read every 👎 along with the user's reason
• Knowledge base — upload PDF, Word, Excel, CSV or text files; search works by meaning and by keyword, tables and scanned pages are read too, and you can review the extracted text before it goes live
• Folder sync — point it at a folder on your server and new or changed files flow into the knowledge base on a schedule; spot a wrong answer in the chat history and save the corrected version as an FAQ
• Ask your databases — connect PostgreSQL, MySQL, SQL Server, SQLite, MongoDB or MinIO/S3 read-only; the system works out how tables relate, so people just ask in plain language, no SQL needed
• Call other systems — set up API requests the way you would in Postman and let assistants call them while answering, with the usual sign-in methods like API keys, OAuth 2.0 and JWT
• Connect tools over MCP — use existing MCP servers, or build your own inside the platform from Google Workspace, Microsoft 365, database or file templates, or your own code
• Search the web — assistants search through your own SearXNG, read pages, and can be limited to the sites you allow
• Skills — add abilities by configuration, such as HTTP calls or ready-made prompts; no user code is ever run
• Pick any model — OpenAI, Anthropic, Gemini, Azure OpenAI, DeepSeek, Mistral, Groq, OpenRouter, xAI, Z.ai, or models you run yourself with Ollama, vLLM, LM Studio, llama.cpp, LocalAI or LiteLLM; choose a model per assistant and which models each department may use
• Still answering when a model is down — line up backup models, and if the main one fails the system switches over without users noticing
• A gateway for your models — other apps in your company and coding tools like Claude Code or OpenCode can use models through Askly AI, with keys, permissions and budgets in one place
• Website chat button — paste one line of code, pick your colors and text, and choose which sites may use it
• LINE, Teams, Azure Bot and Telegram — people chat with assistants in the apps they already use; on LINE they see a typing indicator while they wait, and even slow answers get through
• Share a link with outsiders — customers can chat with an assistant after signing in with Google or LINE, and you set when the link expires and how many people can use it
• Scheduled runs — for example, have an assistant summarize the day every morning and post it to LINE, Telegram, Slack, Teams or Discord
• API for developers — chat with full or streamed replies, upload files, read history, keep each of your users' history separate under one key with per-user limits, plus code samples and Swagger to try it out
• Keep sensitive data in — detect ID card numbers, phone numbers, emails, credit card numbers or your own banned words, then block the message or hide just that part, on both questions and answers
• Dashboards and chat history — usage, response times and problem answers by channel, model, assistant or user over 7, 30 or 90 days; search chat history, export it to Excel, and decide how long logs are kept
• Control costs — daily or monthly token quotas per department, per person or per API key, optionally per model
• Sign in with company accounts — Active Directory or LDAP, Microsoft Entra ID, Google, Keycloak, Okta or GitHub, with permissions set from people's groups, and staff lists imported from Excel
• Permissions and audit trail — detailed role-based permissions, control over who uses which assistant, and a record of every change
• Secure and self-hosted — keys, passwords and secrets are encrypted before they are stored, and the whole platform runs on your own servers with Docker

System requirements