Skip to Content
Get started

Get started

almyty turns the software your company already runs into AI assistants that do real work, and puts them wherever your staff and customers already talk.

For example: a support assistant that looks up an order in your order system and answers a customer on WhatsApp, a sales assistant in Slack that pulls up an account from your CRM before a call, or a nightly check that posts a short report to your team. You build them yourself, on screen, without writing code.

Start on the hosted platform  (no card required), or run it on your own servers.

How it works, in three steps

1. Connect the systems you already use. Under APIs, give almyty the description of a system: your order system, CRM, help center or intranet. (An API is the way one piece of software talks to another; most business software has one.) almyty works out what the system can do, such as “look up an order” or “issue a refund”, and turns each into a tool an assistant can use.

A connected system, with what it can do

2. Build an assistant. Under Agents, describe the job in plain words, pick the AI model it runs on and tick the tools it may use. You can run it on almost any AI model and switch whenever you like, without starting over. If you prefer to lay out every step yourself, draw it on the canvas instead.

An assistant, with its models and checks

3. Put it where people already are. On the assistant’s Channels tab, publish it as a web chat with its own link, a chat bubble on your website, in Slack, Microsoft Teams, WhatsApp, email, text messages or iMessage, as a desktop or terminal app, or for other AI tools to call.

What keeps it safe

  • Keys stay locked away. The keys to your systems are stored under Credentials, and are never shown to the people using your assistants.
  • You decide who may use what. Each assistant, tool and key is private, shared with a team, or open to the whole organization.
  • Spending caps and visitor limits. Set how much an assistant may spend per answer, per day and per month, and how many messages one visitor may send.
  • A person approves what cannot be undone. An assistant can stop and ask before a big refund, a payment or an email to a customer.
  • A record of every action. Every answer, every tool it used and every change to it is kept, with its cost.

Your first assistant

Two full tutorials take you from an empty account to a working assistant, every click and field, with the text to paste and what to do when something goes wrong:

The step-by-step guides build more assistants, each with a screenshot for every step:

Inside the app, Guide at the bottom of the sidebar shows the same path as a checklist: give an AI your API, build an agent, put it where people are, and run it on your machines. It ticks steps off as you do them.

For developers: your API inside your coding assistant

The same tools can go straight into Claude Code, Cursor or any other client that speaks MCP (Model Context Protocol, the common way AI assistants plug in outside tools). You publish them through a gateway: one address that serves a set of tools.

  1. Under APIs, click Connect an API, pick the kind (OpenAPI / Swagger, GraphQL, SOAP / WSDL, gRPC / proto, SDK / npm or Manual HTTP), give it the description as a link, a file or pasted text, and click Connect API. Every operation becomes a typed tool.
  2. Under Gateways, click Create gateway, pick MCP, pick the API and click Create gateway.

A gateway with its tools

  1. The gateway page shows its access key once, next to a ready-made setup for Claude Code, Cursor and Claude Desktop. For Claude Code:
claude mcp add petstore --transport http \
  --header "x-api-key: gw_live_..." \
  https://api.almyty.com/your-org/petstore

The client setup on the gateway page

Ask Claude to do something and it calls your API through the gateway:

Claude calling the gateway

A client that only speaks stdio can use the gateway through a local proxy: npx @almyty/mcp-server your-org/petstore. Everything in the app can also be done from the CLI:

npm install -g @almyty/cli
almyty login

An assistant you build also has an OpenAI-compatible endpoint, so anything that talks to OpenAI can talk to it. See the API reference, the CLI and Control almyty via MCP.

Where to next

Overview