Invite-only beta · Model routing

One API. Less work. Smarter spend.

For teams running AI agents, MUSCLE Flex keeps up with changing models and selects a qualified, cost-aware route for each supported task. Less provider upkeep. More deliberate spend.

Why Flex

Fewer moving parts between your agent and the right model.

The right route for the work.

Each supported task gets a path that qualifies for it, not the same model for everything.

Send a supported request to muscle/auto. Flex checks what the task needs, then chooses a qualified route for speed, reasoning, tools, and cost.

Stop chasing model releases.

We test new models and provider paths before they earn a place in the roster.

We test new models and provider paths against agent workflows before they earn a place in the roster. Your team can skip the release watching, repeat benchmarks, and provider switching.

Control cost without slowing agents.

Routine work uses efficient qualified paths. Demanding steps use stronger ones.

Routine work can use efficient qualified paths; demanding steps can use stronger ones. Supported streams, tool calls, and continuations stay in your agent. Your application keeps its tools and approvals.

A low score cannot make an incompatible route safe. Hard requirements stay hard.

How Flex qualifies a route

How it works

How Flex chooses a route.

Quickstart

Keep your agent. Change one endpoint.

Three steps

Point your client at Flex.

Set the base URL of a supported OpenAI or Anthropic SDK, or a tested coding agent, to Flex. Add your Flex key. Nothing else in your agent changes.

Ask for `muscle/auto`.

Set the model to muscle/auto and send the request you already send. Streams, tool calls and continuations keep the shape your SDK expects.

Flex picks the route.

Flex checks the request, selects a qualified route for the task, returns the response and records usage against your balance. Every response carries a request ID for support.

Illustrative example

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: `${process.env.FLEX_BASE_URL}/v1`,
  apiKey: process.env.MUSCLE_FLEX_API_KEY,
});

const res = await client.chat.completions.create({
  model: "muscle/auto",
  messages: [
    { role: "user", content: "Summarize this test failure." },
  ],
});

import Anthropic from "@anthropic-ai/sdk";

// Host only: the SDK adds /v1 itself
const client = new Anthropic({
  baseURL: process.env.FLEX_BASE_URL,
  apiKey: process.env.MUSCLE_FLEX_API_KEY,
});

const msg = await client.messages.create({
  model: "muscle/auto",
  max_tokens: 1024,
  messages: [
    { role: "user", content: "Summarize this test failure." },
  ],
});

# Check the key and balance. This call is never charged.
curl "$FLEX_BASE_URL/v1/account" \
  -H "Authorization: Bearer $MUSCLE_FLEX_API_KEY"

Two variables

FLEX_BASE_URL: Flex endpoint
MUSCLE_FLEX_API_KEY: Your key
Model: muscle/auto

Developers

Full quickstart

Endpoints, profiles and checks.

Spend

A clearer way to control spend.

Spend less time managing providers and avoid using maximum capability for every task. The beta uses one prepaid balance with usage visible in one place.

Less upkeep

One integration.

One Flex key and endpoint while we maintain the qualified route roster.

Smarter spend

Pay for the fit.

Cost-aware routing uses efficient qualified paths when the task allows.

Beta pricing

Prepaid. Visible. Controlled.

Fund a balance first. New work stops when it runs out, and the portal shows where it went.

Prepaid credits, added to your balance by the team during the beta. Beta pricing is provisional and may change with notice; savings vary by workload. The beta covers supported coding-agent and tool-use workflows.

Pricing

See beta pricing

The family

The MUSCLE family.

Invite-only beta

MUSCLE Flex

One API for qualified, cost-aware model routing.

Coming later

MUSCLE Tool

A future layer for coding-agent workflows.