Any AI. Any API.

TML is an open format that makes it easy for AI agents to discover and use public APIs — using a fraction of the tokens of existing standards.

Thousands of public APIs exist — weather, maps, finance, government data. AI agents need a simple way to know what's available and how to call it. TML gives every API a lightweight, portable spec file that any LLM can read. No server. No SDK. Just a file.

There's no simple way to tell AI what APIs exist

Today, developers hardcode tools or use verbose specs designed for humans, not agents. There's no standard for API discovery that's lightweight enough for AI. And the formats that do exist waste thousands of tokens on schema overhead — every message, every turn.

Projected from 3-tool benchmarks. Actual token counts vary by API complexity.

One file per API. Any AI can read it.

A .min file is a complete, portable tool spec. Publish it on GitHub, share it in a prompt, or load it into any agent framework. Every AI gets the same tools — in a fraction of the tokens.

{
  "openapi": "3.0.0",
  "paths": {
    "/forecast": {
      "get": {
        "operationId": "getCurrentWeather",
        "summary": "Get current weather...",
        "parameters": [
          {
            "name": "latitude",
            "in": "query",
            "required": true,
            "schema": { "type": "number" },
            "description": "Location latitude"
          },
          {
            "name": "longitude",
            "in": "query",
            "required": true,
            "schema": { "type": "number" },
            "description": "Location longitude"
          },
          // ... 5 more params, each 6 lines
        ],
        "responses": {
          "200": {
            "content": {
              "application/json": {
                "schema": { /* nested */ }
              }
  // ... repeat for 3 endpoints

7 formats tested. Only one wins on both size and accuracy.

5 real scenarios, 35 API calls, all against live Open-Meteo weather API. Every format given the same 3 tools.

Token counts measured from Claude API input_tokens for 3 weather tools, 5 live scenarios (Open-Meteo, March 2026). TML achieved 100% tool selection accuracy across all tested scenarios.

Complementary, not competitors

TML and MCP solve different layers of the tool stack. Use both.

Two formats, one spec

TML Source is human-readable. .min Index is token-optimized. Both describe the same tools.

@tool.id
desc: What this tool does and when to use it
GET /endpoint
auth: none|bearer|apikey
in:
  param:str!          "description"
  param:num=default   "description"
out: {response_shape}
tags: keyword tags for search
---

#oatr1|API Name|https://base.url
tool.id|METHOD /path|param:t!,param:t=default|{output}|tags

Real registries, real APIs

These .min files describe production APIs. Each one is a complete tool spec.

Compile, test, and chat with your tools

A local dev tool with three panels: compile OpenAPI to .min, manage your registry, and test tools with a live AI chatbot.

Start in 3 steps

@users.list
desc: List all users with pagination. Use when user asks to see users or search the directory.
GET /users
auth: bearer
in:
  page:int=1       "Page number"
  limit:int=10     "Results per page"
out: {users:[{id,name,email}]}
tags: users list directory search
---

@users.get
desc: Get a single user by ID. Use when user asks about a specific person.
GET /users/{id}
auth: bearer
in:
  id:str!          "User ID"
out: {id,name,email,role}
tags: users get profile detail
---

You have access to these API tools:

#oatr1|My API|https://api.example.com
users.list|GET /users|page:i=1,limit:i=10|{users:[{id,name,email}]}|users list
users.get|GET /users/{id}|id:s!|{id,name,email,role}|users get profile

Select the right tool and construct the API call.

Our goal: every public API has a .min file that any AI can read.