Agents that do the work, and stop to ask for your OK.
Every area has an agent that reads your base and prepares real work. Anything that goes out to a client waits for you — and every step is logged.
14-day trial · No card · Then a free plan
Sara, Support: “Answer Lumen's ticket about portal access.”
I found the wiki article on resetting access and an identical ticket solved in August. I drafted the reply — shall I send it to Lumen? Reply to Lumen's ticket with the steps. Replied. The ticket is now awaiting the client and the action is in the agent's log.
Studio Bravo is a demo company. Made-up data.
AI at work, before and after
The chatbot sounds great, but it can't see the client, the contract or the ticket.
The agent reads the same base that holds your clients, projects, invoices and tickets.
You copy the AI's answer into the email, the CRM and the spreadsheet.
The agent does the work in the modules: creates the task, drafts the proposal, logs it in the CRM.
Nobody knows what the automation did last night at 11 pm.
Every tool call is logged, with its arguments, result and duration.
Each AI tool learns on its own and forgets the next day.
What one agent learns about the company goes into a memory every agent uses.
Agents with access, limits and memory
The agents come with the platform and work in the modules you switched on. You decide what each one may do.
One per area
An agent for every part of the company
Sales, support, finance, marketing, projects and development each come with an agent. Talk to them, mention them in a team chat, or let an event call them — a new ticket, an overdue invoice.
- Ready-made agents per area, or a custom one
- Triggered by a chat, an event or a schedule
- On WhatsApp, Instagram, email and website chat
Approval
Nothing sensitive goes out without you
Issuing an invoice, sending a proposal, publishing a page or touching an ad budget pauses the agent until a person approves that call, with the exact arguments on screen.
- Supervision level per agent
- Tools switched on or off per agent
- Pause every agent in one click
Audit
Every call is on record
Every tool an agent uses goes through the same path: the organization is injected by the system, never chosen by the AI, and the call is stored with what went in and what came out.
- Run history per agent
- Arguments, result and error of each call
- Relevant actions in the organization's audit log
Memory
A company memory, not a chat's
Facts about the company live in a memory shared by the agents, and they search project documents, contracts and solved tickets. An agent talking to a client only gets what is safe for that client.
- Memory shared by every agent
- Search across the company's documents and history
- Internal knowledge kept out of client conversations
Agents are worth what they can see
- Your other tools, over MCPConnect your team's MCP servers once and use them in every agent — each tool off, automatic or asking for approval.
- From ticket to answerThe support agent answers from your wiki and routes to the right department when she doesn't know.
- From lead to meetingThe sales agent qualifies, follows up and logs everything on the lead's record.
Modules you can switch on today
Read from the same catalog as the module store: if it's listed here, it's in the product.
- AI connectionsfrom R$59/month
- Organization AI connectionUse your organization’s Vercel AI Gateway, Z.AI or Anthropic account for compatible native agents. Native inference and its included assistants use your selected account. Images, embeddings, voice and independent BaseOS services keep their current charges.
- MCP serversfrom R$99/month
- MCP serversConnect your team's remote MCP servers (Linear, Notion, Sentry, your own ERP) once and reuse them across agents.
- Per-tool approvalSwitch each tool on per agent as Off, Automatic or Ask for approval — approval is per call, showing the exact arguments.
- Agents operate your productYour support agents act inside your own product for the customer they are talking to — resend a reset link, read and change their settings — after the customer proves who they are with a support code your product shows them.
Can the agent mess up my data?
It only sees your organization and only uses the tools you allowed. Irreversible or client-visible actions — issuing an invoice, sending a proposal, publishing — wait for a human, whatever the automation level.
How do I know what it did?
Every run shows the tools it called, with arguments and results. Relevant actions also go into the organization's audit log.
Do I pay extra for the agents?
The agents come with the platform; AI usage draws on prepaid credits. If you prefer, the AI Connections module runs on your company's own Vercel AI Gateway, Z.AI or Anthropic account.
Does it answer my clients on its own?
In the conversation the client is in, the agent replies and calls a human when needed. Sending anything outside that conversation, like an email or a proposal, goes through your approval.
Be the owner again.Not the integrator.
Start with the area that hurts most. In 14 days you'll know whether BaseOS stays — no card, and a free plan afterward.
14-day trial · No card · Then the free plan