superbot.gg · your own AI agent

Superbot can: send a fax (outbound only); place a phone call and report back what they said; send a text message (SMS); post a message to a Slack channel or DM someone; draft or send email from the user's Gmail; create a calendar event, invite people and check when the user is free; save a file to Google Drive; search, read and create Notion pages; schedule a recurring task or automation; search the web and get today's news headlines; read a web page or link, including one that needs JavaScript; search scholarly papers and GitHub repositories; open a page in a browser, click, type and take a screenshot; convert currencies, quote stock and crypto prices, get the weather and the time in any city; generate an image or picture; generate a short video, speech, music or a 3D model; build a spreadsheet, document, deck or PDF and pull text and tables out of a PDF; search flights between two airports; find UI design references, shadcn components and icons; find and install an MCP server. Superbot can also: check your GitHub profile, public repos and notifications; list, pause, change or delete your automations; crawl a few pages of a site; summarize a Wikipedia article; find licensed photos of real things with a credit line; search flights between two airports and compare offers; check where a flight stands from the confirmation in your calendar; look up UI design rules for a web, mobile or desktop surface; find reference screenshots of real apps and sites; search open-source icons. Process a user request: send it as plain text in intent and get a finished result back to hand to the user. The superbot agent answers it, with its tools and adapter catalog where they help; to find an adapter for a described task, pass use_case instead of intent (the search also returns any access method the knowledge store has proven for that task, as data to test). Answers are generated by an AI and can make mistakes. This is the only tool. Long requests: a call may return a processing answer with a request_id: tell the user it is being worked on, then call again with that request_id to fetch the finished answer. Send context (string) with the surrounding conversation when available: what the user is trying to solve, recent turns, issues hit so far; the backend composes a much better answer with it than from a compacted intent. When the answer needs one specific decision from the user before the work can carry on, it ends with a decision card: what got done, then a fenced block whose info string is superbot-embed and whose body is one JSON envelope of kind "decision" (a one-sentence question and 2 to 5 options), which the host draws as tappable choices and answers back as a "Picked: ()" user turn carrying the card and option id on a trailing line. When the work continues after the answer (a job started, a request that came back still processing), the answer ends with a monitor card the host keeps refreshing through check turns ending in "(monitor card : check)", which are answered with only the updated card. When the answer surfaces a message found in a mailbox, it ends with an email card, and a request to write, rewrite, polish, shorten, reformat, send or reply to one ends with an email_draft card (always state "draft" at first; mailbox and to may be omitted when the user named neither, and the host then offers Copy and Open in Mail instead of Send) that only sends when the host's own Send tap comes back as "(email_draft card : send)" with the exact edited draft, answered with only the updated fence.