
Everything in the panel, also in code.
An API, an MCP server and exportable agents, with changes through pull requests and tests.
What Developers does.

const res = await fetch("https://api.baguals.ai/v1/tables/containers/rows", {
method: "POST",
headers: {
Authorization: `Bearer ${process.env.BAGUALS_TOKEN}`,
"Content-Type": "text/csv",
},
body: await readFile("containers-in-transit.csv"),
});
// { "inserted": 1284, "flagged": 7 }{
"mcpServers": {
"baguals": {
"url": "https://mcp.baguals.ai/mcp"
}
}
}
// Claude opens the sign-in page, you choose the mode
// (read or read and write) and the folders it can reach.curl https://api.baguals.ai/v1/agents/customer-agent/versions/7/export \
-H "Authorization: Bearer $BAGUALS_TOKEN" \
-o customer-agent-v7.json
# instructions, skills, tools, tests and model profile# tests/quote-reefer.yaml
customer: "I need a quote for a 40' reefer to Callao"
must_call: quote_freight
must_mention: "plug-in fee"
must_not_say: "guaranteed price"
✓ 42 tests passed · 0 failed
→ Waiting for human approval to merge
A future-ready operation connects to everything else.
You hand us the process; we do the rest. For your IT team, that means agents that fit into existing systems with clear permissions, not another tool to babysit.
Explore the approachQuestions, answered.
Is the API public?
Not as self-service yet. During a pilot, we give your team access to the parts of the API the project needs, with tokens tied to your company and a role. The endpoints shown on this page are illustrative. Public documentation comes once the API is stable. If your IT team needs a specific integration, we define it together in the workshop.
What can we do through the MCP server?
Connect Claude, or another assistant that supports MCP, to your company’s memory. A person can ask about a policy or update a process from their usual chat, and an agent can read the context it needs. Every connection signs in with OAuth, works only in the mode and folders you approved, and every call is recorded.
Can we take our agents with us if we leave?
Yes. Every agent can be exported with its instructions, skills, tools, tests and model profile, along with your tables and memory. The platform’s core runs on portable TypeScript and standard Postgres, so it isn’t tied to a single cloud provider. What you built belongs to your company, and we agree on the export and the deletion of the rest when a contract ends.
Who approves changes to our agents?
A person, always. Changes arrive as pull requests with tests, whether a member of our team or a development agent wrote them. A failing test blocks the change, and a passing one still waits for human approval before being merged. Published versions are never edited: each change becomes a new version that has to pass the tests before going live.
Can it run in our own AWS account?
Not in the pilot. Today Baguals runs as a shared platform with each company isolated and data in São Paulo. The platform is built so it can run on AWS when a customer requires it, but installing it in your own cloud account isn’t something we offer yet. If your company needs it, we talk about it and quote it separately.
One OS, three modules.




