TwelveAI logoTwelveAI
Developers

Give your AI access to financial infrastructure.

One call turns a message into a structured, validated financial action. Bring your own rails, wire your tools, and test in a live sandbox before you ship.

API-first by design

Send a message and a set of tools; get back an intent, extracted parameters, and whether confirmation is required. The hard parts, grounding, tool-calling, and guardrails, are handled for you.

Point tools at your existing endpoints so conversational banking runs on your rails, your rates, and your compliance.

For developers

One call, a structured financial action.

A message in, a validated intent out, with confirmation controls built in.

request.js
const response = await twelveai.agent.run({
  user: userId,
  message: "Send ₦50,000 to Sarah",
  tools: ["accounts", "transfers", "transactions"]
})
response.json
{
  "intent": "transfer",
  "amount": 50000,
  "currency": "NGN",
  "recipient": "Sarah Okafor",
  "requires_confirmation": true
}

Built for engineering teams

Composable tools

Declare tools and bind them to your APIs. Test each endpoint live before it goes near a customer.

Sandbox

Run the model against stubbed tool responses so you can build the whole flow first.

Structured output

Every turn returns a clean, typed action you can act on with confidence.

Guardrails in the loop

Confirmations, limits, and policy are enforced by the platform, not your prompt.

Developers with TwelveAI
12Built on real financial rails
Developers in practice

Infrastructure, not a demo.

Everything here runs on the same platform we operate for real customers, grounded answers, gated writes, and a full audit trail.

  • Composable tools. Declare tools and bind them to your APIs. Test each endpoint live before it goes near a customer.
  • Sandbox. Run the model against stubbed tool responses so you can build the whole flow first.
  • Structured output. Every turn returns a clean, typed action you can act on with confidence.
  • Guardrails in the loop. Confirmations, limits, and policy are enforced by the platform, not your prompt.

Start building.

Read the docs and make your first call.