Governed AI tools

Give AI hands.
Keep the keys.

Connect an AI assistant to your real systems — orders, bookings, tickets — and put it in front of your customers, without ever exposing API keys or private data to the model. Import any API from a single cURL command. Govern every field it may touch. Meter every action it takes.

Free development plan · no card required · upgrade when you go live

anatomy · import from cURL
curl -X POST https://api.yourshop.com/v1/orders \
  -H "Authorization: Bearer sk_live_9f24…" \
  -d '{
    "customer_id": "cus_58421",
    "item": "espresso beans 1kg",
    "quantity": 2
  }'

Secret

Encrypted in the vault. Injected server-side. The AI never sees it.

Customer context

Resolved per customer at runtime — never asked from the model.

AI-fillable

The only fields the model may fill, validated on every call.

AES-256-GCM encrypted secrets
Write-only provider keys
Fail-closed tool governance
Per-field allowlists
Full execution audit trail
Hard usage limits

Import any API in minutes

Paste a cURL command and the platform dissects it: secrets, AI-fillable fields, customer context, fixed values — automatically classified, human-reviewed, published as an immutable tool version.

Governed execution

The AI can only call tools you bound, only fill fields you allowed. API keys stay encrypted in a vault and are injected server-side at the moment of execution — the model and the browser never see them.

Your customers, your channels

Web chat, WhatsApp, Telegram, Messenger, Instagram, Slack, Discord, Teams, Webex, SMS, phone calls, and email — every channel drives the same governed runtime with one customer identity.

How it works

From cURL to customer-facing AI

01

Create a flow

Pick any OpenAI-compatible provider and model. The provider key is write-only from that moment.

02

Import your API

One cURL command becomes a reviewed, versioned AI tool.

03

Bind and test

Attach tools to your flow and watch it act in the playground.

04

Go live

Embed the widget on your site. Every message and action is metered and audited.

Built for the people who answer for it

Operators review what the AI may touch. Teams publish tools together. Owners watch adoption and spend — all in one console.

A team reviewing an AI tool draft together around a laptop
Review together, publish versioned
An operator reviewing approval workflows on a dashboard
Approve what the AI may touch
A founder reading customer growth and usage analytics
Watch usage, spend, and growth
Tools Anatomy — governed AI tools

Security model

Built so your security team says yes

Keys the AI cannot leak

Provider and API keys are AES-256-GCM encrypted at rest and injected server-side per call. Nothing sensitive ever reaches the model, the browser, or a log line.

Fail-closed by construction

Every tool call is validated against its published contract and tenant scope. An unresolvable secret or an unbound tool means no call at all — not a best guess.

Tenant isolation everywhere

Workspaces are isolated per request, key-scoped, end to end. Cross-tenant reads collapse to not-found — even for malformed IDs.

Limits before money

Hard usage limits are enforced before a provider is ever charged, with per-meter usage and alerts visible in the console.

Everything on the record

Every import, publish, binding, execution step, and secret access is audit-logged with content-free metadata.

Observable in real time

Live usage meters, event feeds, and execution traces — see exactly what your agent did and why.

Channels

One governed runtime, every channel

Connect a platform from the console, bind a flow, and the same tools, prompt, guardrails, audit trail, and customer identity answer there. Platform credentials go to your Secret Manager — never to the model.

Web chat widget

A two-line embed puts your agent on any website — origin-locked and session-scoped.

Web
WhatsApp

WhatsApp Business

Meta Cloud API: customers message your WhatsApp number and your flow answers.

Messaging
Telegram

Telegram

Paste the BotFather token; the webhook is registered for you and private chats are answered.

Messaging
Messenger

Facebook Messenger

People who message your Facebook Page get answered by your flow.

Messaging
Instagram

Instagram

DMs to your professional Instagram account, answered by the same governed runtime.

Messaging
Slack

Slack

A Slack app bot that answers direct messages and @mentions in your workspace.

Messaging
Discord

Discord

A bot for DMs and @mentions in your server — no public URL needed.

Messaging
teams

Microsoft Teams

An Azure Bot Framework bot that answers chats and @mentions in Teams.

Messaging
Webex

Cisco Webex

A Webex bot that answers direct messages and @mentions in spaces.

Messaging
Twilio

Twilio SMS + Voice

SMS replies and a turn-based voice agent on your Twilio numbers.

SMS · Voice

CallinAI voice

Phone calls answered by your flow, with SMS on the same number when you want it.

Voice · SMS

Email

A dedicated mailbox on Gmail, Microsoft 365, or any IMAP/SMTP provider; replies land in-thread.

Email

Live today
All twelve channels ship today. Voice is turn-based (the platform handles speech, your flow answers each turn); streaming voice is on the roadmap. Any OpenAI-compatible model provider works.

Start free, pay when you go live

Every account starts on the free Development plan — real flows, real tools, real playground. Upgrade in the console when your customers arrive.

Documentation

All docs →

From cURL to customer-facing AI in an afternoon