Each agent gets one job — and exactly the access it needs. Your public Telegram bot stays read-only. Your private coordinator handles everything else.
Give every agent exactly the access it needs — and nothing more.
Sees and responds — nothing else
Can do
Cannot do
Research, draft, and coordinate — no exec
Can do
Cannot do
Full access — for your trusted coordinator
Can do
Pick a template and your agents are live — no configuration needed.
Community Team
Safe public support with a private powerhouse behind it
Startup Team
Ship faster with a full-stack AI team
Dev Team
Build, secure, and document — all in one team
Creator Team
Produce content that actually performs
Research Team
Deep investigation, structured insights
Anywhere a public AI needs safe boundaries.
Your Telegram group has thousands of members. Beacon (read-only) handles the public questions. Host routes anything complex to Concierge behind the scenes. Members get fast answers; your private systems stay safe.
A read-only bot greets new members and answers FAQs. A safe-mode assistant drafts mod responses for review. The coordinator handles escalations privately. No bot can accidentally ban a user or wipe channels.
Your public support bot reads the knowledge base and answers product questions. It never touches customer data or runs queries. Sensitive lookups go to a safe-mode agent working in the background.
No. Read-only agents have exec, file writes, browser automation, and agent delegation blocked at the framework level — not just by instructions. Even if a user tries to trick the bot, the tools simply aren't available to it.
The coordinator agent uses agentToAgent delegation to hand off tasks to specialists by name. Specialists respond directly to the coordinator, which synthesizes and replies to the user. Read-only agents cannot initiate delegation.
Yes. Every agent can have its own preset — unrestricted, safe, or read-only — regardless of how many agents are on the team. You can create a fully custom team from scratch or modify a pre-built template.
Safe mode blocks exec (terminal commands) and file writes — the agent can still browse the web, delegate to other agents, and do research. Read-only adds those restrictions too: no browser automation and no delegation. It's the strictest tier.
Yes. The agent system including all capability presets is available on Starter, Pro, and Business plans. Pro and Business plans support additional instances and more agents per instance.
Pick a pre-built template or configure each agent individually. Capability presets are applied automatically.
Get started free