Developers

openai‑compatible
api.tryparable.ai/v1
usdg settlement

Point any OpenAI SDK at the gateway and ask for a route instead of a model. You get routing, fallbacks, metering, on‑chain tools and USDG settlement without changing your code.

01 · Quickstart

BASE URL https://api.tryparable.ai/v1

import OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://api.tryparable.ai/v1", apiKey: process.env.PARABLE_API_KEY });

const res = await client.chat.completions.create({
  model: "parable/auto",
  messages: [{ role: "user", content: "Summarize this 10-K in five bullets." }],
  // @ts-expect-error Parable extension
  parable: { prefer: "cost", max_cost_usd: 0.01, fallback: true },
});

console.log(res.model); // the model that actually served the request

02 · Routes

parable/auto	Balanced cost, speed and quality. Short prompts lean fast.
parable/fast	Lowest latency; upstream hosts sorted by latency.
parable/cheap	Lowest cost per token; upstream hosts sorted by price.
parable/best	Highest quality tier; reasoning models allowed.
parable/reasoning	Only models with extended thinking.
parable/image	Image‑output models; images arrive on message.images[].

03 · Request extensions
object under parable

prefer	cost \| speed \| quality \| balanced	Weighting used when scoring candidates.
max_cost_usd	number	Hard ceiling; models that could exceed it are excluded.
max_latency_ms	number	Excludes models whose p50 exceeds the target.
require	capability[]	tools · vision · json · reasoning · long_context · image_gen
fallback	boolean	Retry the next candidate on 429 / 5xx / timeout.
web_search	boolean	Ground with live web results; citations in annotations.
private	boolean	Only zero‑retention, no‑training upstream providers.
tools	(chain \| sec \| market)[]	Server‑side tool packs; the gateway runs the agent loop.

04 · Tools

Enable packs with parable.tools. The gateway adds the functions, runs the model↔tool loop (up to six rounds), streams each call and result as parable events, and returns the summed usage. Your own tools pass through untouched.

const res = await client.chat.completions.create({
  model: "parable/auto",
  messages: [{ role: "user", content: "What's in wallet 0x5fc5…d168 and what did USDG do today?" }],
  // @ts-expect-error Parable extension
  parable: { tools: ["chain", "market"] },
});
// The gateway ran the tool loop server-side; res.usage is the summed cost.

get_balance	ETH + USDG/WETH + indexed ERC‑20 holdings of an address
get_transaction	Status, fee, decoded ERC‑20 transfers
get_address_info	Wallet / contract / token, metadata, activity
get_token_info	ERC‑20 metadata, supply, holders, price
get_recent_transfers	Recent transfers for a token or address
get_chain_stats	Latest block, gas, block time, totals
search_chain	Explorer search for tokens, contracts, labels
search_sec_filings	EDGAR full‑text search across forms
get_crypto_price	Spot USD prices for major assets
get_stock_quote	Delayed equity/ETF quotes

05 · Response headers

x-parable-request-id	Correlates with usage and logs.
x-parable-model	Model that served the request, after fallbacks.
x-parable-cost-usd	Price charged, 6 dp. 1 USDG = 1 USD.
x-parable-latency-ms	Wall‑clock time at the gateway.

06 · Endpoints

POST	/v1/chat/completions	OpenAI‑compatible; SSE streaming; accepts parable extensions.
GET	/v1/models	Catalog with pricing, context, capabilities, live latency.
POST	/v1/parable/route	Dry run: ranked candidates and why others were rejected.
GET	/v1/parable/stats	Public network metrics.
GET	/v1/parable/chain	Live Robinhood Chain snapshot (block, gas, USDG supply).
GET	/v1/parable/pricing	Price sheet per model and route, in USDG.

07 · Access & payments

API keys	Bearer keys with per‑key rate limits and spend caps.	phase 1
USDG credits	Deposit USDG on Robinhood Chain; the balance draws down per request at the published price.	phase 3
x402 pay‑per‑request	No account: call the API, receive a 402 with a USDG quote, retry with a signed payment.	phase 3