Telegram, iMessage and SMS
This is planned. You will link Telegram or a phone number with a one-time code and chat with any of your bots from there. Messages land in the same bot thread, and you can answer approvals with a button or a reply.
A JamBots bot keeps one continuous conversation, remembers what it learns, runs routines on a schedule and works on its own cloud computer with a browser, a shell and a desktop. It connects to your apps through MCP, splits big jobs across subagents, and asks for approval before anything consequential.
A bot is a continuing teammate with one conversation that never resets. You give it a name, an avatar and standing instructions, then talk to it like a colleague. You can change its settings by asking it in chat.
Each bot writes down preferences and facts as it learns them. You can read, edit and delete any entry, see the revision history and roll back. Every bot in a workspace can also read the workspace's shared memory.
Yes. A routine is a prompt with a schedule: one time, every interval, or a cron expression with a time zone. Each run posts its result or a question in the bot's thread, and starts within 60 seconds of its scheduled time.
Each bot gets its own cloud computer with a shell, a file system, a browser and a desktop. It starts when the bot needs it and sleeps when idle, keeping files and sign-ins. You can watch the screen live and take control.
Not for anything consequential. Before a bot sends, buys, deletes, posts or writes to an outside system, it posts an approval card with Allow once, Always allow and Deny. An approval nobody answers expires after 10 minutes and the action does not run.
JamBots works with OpenAI, Anthropic, OpenRouter and any OpenAI-compatible endpoint. Each bot has a chosen model and an optional fallback used when the first one fails. Every call records tokens and cost, and the workspace shows usage by bot and by day.
Yes. You connect remote MCP servers over HTTPS, including ones that sign in with OAuth. A connection belongs to you or to the workspace, and you choose which bots can use it. Large tool catalogs are searched rather than loaded all at once.
Yes. A bot can start short-lived subagents for parallel subtasks and collect their results. It runs at most 4 at a time, one level deep, and at most 20 per run. The backend enforces those limits, not the model.
Every bot gets these tools. A tool its model cannot use, such as desktop control on a model without vision, is not offered to that bot.
These are planned and not available yet. After them come bot-to-bot work, shareable bot templates, skills a bot learns by watching you, and one-click connectors for common apps.
This is planned. You will link Telegram or a phone number with a one-time code and chat with any of your bots from there. Messages land in the same bot thread, and you can answer approvals with a button or a reply.
This is planned. You will dictate messages, hear replies spoken aloud and hold a live voice call you can interrupt. Speech uses your workspace's own voice provider key, with OpenAI and ElevenLabs supported.