Build a chat or text feature on the FLOP gateway API

The simplest way to use FLOP compute is to call it from an app. Flop runs a public gateway at https://api.flop.network/v1 that "speaks the OpenAI-style gateway API," according to its SDK reference page. If your product already sends prompts to a hosted model and reads text back, the gateway is the shortest path to trying the same feature on FLOP miners.

Status, November 10, 2026: the gateway serves the devnet, which earns nothing. The testnet is not live. See the compute use cases hub for the latest status.

What the gateway does

An app sends a chat completion request, the gateway routes it to a miner serving that model, and the response comes back in the familiar OpenAI shape. The @flop-labs/pi README says the gateway "lists an offer only while an attested, identity-verified, priced, and healthy compute session serves it," and that each response "is settled through a FLOP compute channel." Underneath, it is the same request, match, prove and settle cycle described in our FLOP AI agents guide.

Flop treats the gateway as a convenience layer. On November 7 Flop Labs wrote that the "gateway is a convenience on top of actual settlement logic, we do encourage to interact with chain directly." Apps that need full control of payment, receipts and disputes should use the chain SDK, covered in agent budgets.

What models can an app call?

On November 10 the model list held one entry: qwen/qwen2.5-0.5b-instruct, named "Qwen2.5 0.5B Instruct (devnet-ts)". It is text in, text out, with a 32,768-token context window and per-request limits of 24,576 prompt tokens and 8,192 completion tokens. Its prices are tagged "devnet-nominal".

That is a test model, so don't judge quality from it. Flop's demand FAQ says miners choose which open-weight models to register, from "frontier-scale open models on multi-GPU clusters down to roughly 20B-parameter models on a single consumer card," and that "frontier proprietary models are excluded at launch." Which models are live at any moment "depends on which miners have loaded them."

The agent integration doc adds a warning for anyone who reads the catalog as a promise: "A gateway model catalog is discovery data and is not proof that inference is available."

How to wire it into an app

Flop's official client is @flop-labs/client, version 0.2.2 on npm when we checked on November 10. It is a small fetch-based client for browsers and Node.js 22 or later. The docs' example is three calls:

npm install @flop-labs/client
import { FlopClient } from "@flop-labs/client";
const client = new FlopClient({ apiKey: process.env.FLOP_API_KEY });
const models = await client.listModels();

Then createChatCompletion with a model ID and a list of messages. Streaming is supported. The SDK page says "Model discovery needs no key; chat completions do." Errors come back as FlopHTTPError. If you write your own client, @flop-labs/contracts ships JSON Schema and validators for the same API. Our FLOP SDK guide lists every official package.

How do you get a gateway API key?

Flop's public docs pages don't describe a signup. The Pi package README is the most specific source we found. It tells users to set FLOP_API_KEY to "a bounded child credential (flop_ek_… from POST /v1/keys/ephemeral or POST /v1/connect/grants) with a spend cap, rather than an owner key." Follow that advice in any app: give the app a capped credential, never a key that controls your main wallet. Get keys only through Flop's own endpoints and docs, never from a reseller or DM.

What would you do on testnet day one?

1. Prove the integration on devnet now with the builder quickstart.
2. When the testnet opens, switch to the testnet gateway or RPC that Flop publishes on its status page. Don't guess endpoints.
3. Run your feature on real prompts and watch latency and failures. Flop's demand FAQ says buyers pick miners "on price, latency and region".
4. Keep a fallback. The Pi README notes that answers from a non-FLOP fallback model are "not FLOP-verified," which is a reasonable rule for any app.

Does gateway usage earn the airdrop?

Possibly, but the docs don't say how gateway calls are attributed. The airdrop page shares the agent pool "pro rata to compute purchased in settled sessions," and the timeline FAQ describes an agent with a DID key that claims faucet FLOP and submits "an inference request signed with that key." We found nothing in Flop's docs that explains whether calls made under a gateway API key are credited to your agent identity. If the airdrop matters to you, also follow the signed-request path in FLOP for AI agents.

The usual limits apply: devnet earns nothing, faucet grants alone earn nothing, and the testnet page says spend sent to a miner you control isn't counted. Agent allocations unlock at 3:1 after genesis. Read the airdrop guide for the full rules.

Limits to plan around

• The devnet is rate-limited, and Flop has said it is expanding capacity slowly. See devnet limits.
• One small devnet model is not a production model. Plan for the catalogue miners bring to the testnet.
• The gateway is a convenience layer. Receipts and disputes live on chain.
• No FLOP token exists yet. Nobody can sell you gateway credits for real money today; see FLOP scams.

FAQ

Is the FLOP gateway compatible with OpenAI clients?

Flop's SDK page says the gateway speaks an OpenAI-style API and that its default base URL is https://api.flop.network/v1. The official client is @flop-labs/client. Model listing works without a key, and chat completions need one.

Which models does the FLOP gateway serve?

On November 10, 2026 it listed one devnet model, Qwen2.5 0.5B Instruct, text to text with a 32,768-token context. Flop's docs say miners will serve open-weight models they choose, and frontier proprietary models are excluded at launch.

Do gateway calls count toward the FLOP airdrop?

Flop's docs don't say how gateway calls are credited. The agent pool is shared by compute purchased in settled testnet sessions, and the timeline FAQ describes inference requests signed with the agent's DID key. Devnet calls earn nothing.