API reference

API Docs

OpenAI-compatible inference with wallet auth and x402 credit billing. One endpoint, one key, every curated model.

OpenAI-compatible21 modelsx402 on Base

Get API key Quick Start View Pricing

No credit card required · free website chat

Quick Start

Get up and running with the OMA API in under a minute. Drop-in replacement for OpenAI with zero vendor lock-in.

Get API Key

Create an account and generate your API key from the dashboard.

Add to Headers

Include your key in the Authorization header: Bearer oma_sk_...

Start Building

Use the OpenAI SDK with base_url=https://www.oma-ai.com/api/v1

Your First Request

cURL

curl https://www.oma-ai.com/api/v1/chat/completions \
	-H "Authorization: Bearer oma_sk_your_key_here" \
	-H "Content-Type: application/json" \
	-d '{
	"model": "llama-3.3-70b",
	"messages": [{"role": "user", "content": "Hello!"}]
	}'

Base URL

https://www.oma-ai.com/api/v1

SDK Install

npm install openai

Chat · Anthropic messages · curl

Chat (OpenAI-compatible)

cURL

curl https://www.oma-ai.com/api/v1/chat/completions \
		-H "Authorization: Bearer oma_sk_your_key_here" \
		-H "Content-Type: application/json" \
		-d '{
	"model": "llama-3.3-70b",
	"messages": [{"role": "user", "content": "Hello!"}]
	}'

Messages (Anthropic-compatible)

cURL

curl https://www.oma-ai.com/api/v1/messages \
	-H "x-api-key: oma_sk_your_key_here" \
	-H "anthropic-version: 2023-06-01" \
	-H "Content-Type: application/json" \
	-d '{
	"model": "claude-sonnet-4-6",
	"max_tokens": 256,
	"messages": [{"role": "user", "content": "Hello!"}]
	}'

Authentication

Programmatic inference uses a Bearer API key. Website accounts use email or wallet sessions; eligible free website chat also works without sign-in.

Free OMA-AI models (website chat)

Eligible OMA-AI models can be used on /chat without an API key or wallet (20 msgs/day anonymous, 100/day signed-in), subject to live model availability. The chat model selector shows the current free options. Prepaid credits cover paid requests; rate limits still apply.

API-key calls (Authorization: Bearer oma_sk_…) always spend credits, including models eligible for free website chat. Free usage applies to browser website chat only, never to programmatic API traffic.

Authorization Header (paid API)

Authorization: Bearer oma_sk_...

curl https://www.oma-ai.com/api/v1/chat/completions \
	-H "Authorization: Bearer oma_sk_your_key_here" \
	-H "Content-Type: application/json" \
	-d '{"model":"llama-3.3-70b","messages":[{"role":"user","content":"Hello"}]}'

API Key Security

Keys use the oma_sk_ prefix. Only the SHA-256 hash is stored. Keep your key secret and never commit it to version control.

Email Magic Link

How it works

Enter your email address on the login page
Click the link sent to your inbox within 5 minutes
Create an API key in the dashboard and save it when shown
Credits can be added via USDC on Base (card when Polar is enabled)

Wallet Auth (SIWE)

Connect any EVM wallet using Sign-In with Ethereum (SIWE). No email, no password. Your wallet address is your identity.

How it works

Click "Connect Wallet" on the login page
Sign a SIWE message with your wallet (no gas cost)
Create an API key in the dashboard and save it when shown
Use the key with any OpenAI-compatible SDK

Agent & IDE integrations

Use any OpenAI-compatible client: set base URL to www.oma-ai.com/api/v1 and Authorization: Bearer $OMA_KEY

One key, one base URL

Base URL: https://www.oma-ai.com/api/v1
API key: create at /dashboard → API keys (oma_sk_…)
Default sample model: llama-3.3-70b (see /models)
Claude Code / Anthropic SDK: also use POST /v1/messages
System instructions: For custom agent setups, see our agent-skill.md file.

OpenCode

Point OpenCode at OMA as a custom OpenAI-compatible provider. Env vars are the portable path; project config can pin the model.

# ~/.bashrc / shell profile (or OpenCode env UI)
export OPENAI_API_KEY="oma_sk_your_key_here"
export OPENAI_BASE_URL="https://www.oma-ai.com/api/v1"

# Optional: default model for sessions
export OPENAI_MODEL="llama-3.3-70b"

# Then run OpenCode as usual — it will hit OMA for chat completions.

If your OpenCode build uses OPENAI_API_BASE instead of OPENAI_BASE_URL, set that to the same https://www.oma-ai.com/api/v1 value.

Project config (optional)

// opencode.json / provider settings (shape varies by OpenCode version)
{
	"provider": {
	"oma": {
	"npm": "@ai-sdk/openai-compatible",
	"name": "OMA",
	"options": {
	"baseURL": "https://www.oma-ai.com/api/v1",
	"apiKey": "{env:OMA_API_KEY}"
	},
	"models": {
	"llama-3.3-70b": { "name": "llama-3.3-70b" }
	}
	}
	}
}

Create the key at /dashboard → API keys. Never commit oma_sk_… to git — use env vars.

Claude Code

Claude Code talks the Anthropic Messages API. OMA exposes a compatible POST /v1/messages that rewrites to the same catalog. Configure a custom Anthropic base URL + key:

# Claude Code / Anthropic SDK env
export ANTHROPIC_API_KEY="oma_sk_your_key_here"
export ANTHROPIC_BASE_URL="https://www.oma-ai.com/api"

# Note: base URL is the origin (no /v1). The client appends /v1/messages.
# Model ids are OMA catalog ids (not anthropic/claude-3-5-sonnet):
# kimi-k3 · claude-opus-5 · llama-3.3-70b · …

# One-shot smoke test (Messages API — x-api-key style)
curl https://www.oma-ai.com/api/v1/messages \
	-H "x-api-key: oma_sk_your_key_here" \
	-H "anthropic-version: 2023-06-01" \
	-H "Content-Type: application/json" \
	-d '{
	"model": "kimi-k3",
	"max_tokens": 64,
	"messages": [{"role": "user", "content": "ping"}]
	}'

# Same call with Bearer (OpenAI-style auth also accepted)
curl https://www.oma-ai.com/api/v1/messages \
	-H "Authorization: Bearer oma_sk_your_key_here" \
	-H "anthropic-version: 2023-06-01" \
	-H "Content-Type: application/json" \
	-d '{"model":"kimi-k3","max_tokens":64,"messages":[{"role":"user","content":"ping"}]}'

Prefer one auth style consistently. Streaming: set stream: true — OMA returns Anthropic-shaped SSE. List models: GET /v1/models with the same key.

Cursor / Continue / Cline

Any IDE that accepts an OpenAI-compatible endpoint:

Open model / provider settings → Custom OpenAI
API Base: https://www.oma-ai.com/api/v1
API Key: your oma_sk_…
Model: any id from GET /v1/models

// Example Continue config snippet (~/.continue/config.json)
{
	"models": [
	{
	"title": "OMA llama-3.3-70b",
	"provider": "openai",
	"model": "llama-3.3-70b",
	"apiBase": "https://www.oma-ai.com/api/v1",
	"apiKey": "oma_sk_your_key_here"
	}
	]
}

openai-python / openai-node

Drop-in replacement — only change base URL and key:

from openai import OpenAI

client = OpenAI(
	api_key="oma_sk_your_key_here",
	base_url="https://www.oma-ai.com/api/v1",
)

r = client.chat.completions.create(
	model="llama-3.3-70b",
	messages=[{"role": "user", "content": "Hello from OMA"}],
	max_tokens=64,
)
print(r.choices[0].message.content)

Endpoints Reference

OpenAI-compatible REST API endpoints.

The public inference, model, balance, usage, and deposit endpoints.

OpenAI-compatible multi-turn chat. The endpoint you will hit 95% of the time. Supports SSE streaming.

Request Body

modelstring (required). Any model id from /v1/models, e.g. "llama-3.3-70b"

messagesarray of { role, content }. role is system, user, assistant, or tool

temperaturenumber 0 to 2 (optional, default from upstream)

max_tokensinteger. Cap on completion length

streamboolean. Server-Sent Events when true (text/event-stream)

toolsarray of tool definitions (optional)

response_formatobject (optional), e.g. { type: "json_object" }

Response

{
	"id": "chatcmpl-9f3k2a",
	"object": "chat.completion",
	"created": 1717194000,
	"model": "llama-3.3-70b",
	"choices": [
	{
	"index": 0,
	"message": { "role": "assistant", "content": "Hello. How can I help today?" },
	"finish_reason": "stop"
	}
	],
	"usage": { "prompt_tokens": 18, "completion_tokens": 12, "total_tokens": 30 }
}

Available Models

21 curated chat models across four tiers, plus web search in the playground. Check the live catalog before making a request.

Chat Completions

Text and reasoning models from Venice (frontier/premium/mid/budget) and OMA-AI (MiniMax M2.7, DeepSeek V4 Flash, GLM-5.3 Flash).

Vision-capable models

A subset of premium / frontier models accept image inputs alongside text in the OpenAI multimodal content array shape.

Budget7 models

Fast, cost-effective models for simple tasks and prototyping.

Model	Input / 1M	Output / 1M	Context	Capabilities
`xiaomi-mimo-v2-5` Xiaomi MiMo V2.5	$0.50	$2.50	1M	chatreasoningfast
`deepseek-v4-flash-0731` DeepSeek V4 Flash (Venice)	$0.21875	$0.4375	1M	chatreasoningfast
`mistral-small-2603` Mistral Small 4	$0.234375	$0.9375	256K	chatfastagents
`google-gemma-4-31b-it` Gemma 4 31B	$0.15	$0.45	256K	chatfast
`minimax-m2-7` MiniMax M2.7	$0.60	$0.60	200K	chatreasoningcodingagents
`deepseek-v4-flash-0731-openbroker` DeepSeek V4 Flash (OMA-AI)	$0.50	$0.50	400K	chatreasoningfastagents
`zai-org-glm-5-3` GLM-5.3 Flash (OMA-AI)	$0.30	$1.00	1M	chatreasoningmultimodalfastagents

Mid-Range5 models

Balanced performance and cost for everyday workloads.

Model	Input / 1M	Output / 1M	Context	Capabilities
`llama-3.3-70b` Llama 3.3 70B	$0.875	$3.50	128K	chatagents
`hermes-3-llama-3.1-405b` Hermes 3 405B	$1.375	$3.75	128K	chatreasoningagents
`zai-org-glm-5-1` GLM 5.1	$1.925	$6.05	200K	chatreasoningagents
`qwen-3-7-plus` Qwen 3.7 Plus	$0.625	$2.50	1M	chatreasoningagents
`venice-uncensored-1-2` Venice Uncensored 1.2	$0.25	$1.125	128K	chatuncensored

Premium7 models

High-performance models for demanding tasks.

Model	Input / 1M	Output / 1M	Context	Capabilities
`kimi-k3` Kimi K3	$4.6875	$23.4375	1M	chatreasoningagents
`zai-org-glm-5-2` GLM 5.2	$1.75	$5.50	1M	chatreasoningagents
`deepseek-v4-pro` DeepSeek V4 Pro	$2.0625	$4.12625	1M	chatreasoningcodingagents
`qwen-3-8-max` Qwen 3.8 Max	$3.125	$9.375	1M	chatreasoningcodingagents
`grok-4-5` Grok 4.5	$2.8375	$8.50	500K	chatreasoningcodingagents
`claude-sonnet-4-6` Claude Sonnet 4.6	$4.50	$22.50	1M	chatreasoningcodingagents
`gemini-3-6-flash` Gemini 3.6 Flash	$1.171875	$5.859375	1M	chatreasoningcodingvisionagents

Frontier2 models

Top-tier models for complex reasoning and generation tasks.

Model	Input / 1M	Output / 1M	Context	Capabilities
`openai-gpt-56-sol` GPT-5.6 Sol	$7.8125	$46.875	1M	chatreasoningcodingagents
`claude-opus-5` Claude Opus 5	$7.50	$37.50	1M	chatreasoningcodingagents

List All Models

Use the GET /v1/models endpoint to fetch the current model catalog with pricing and capabilities.

curl https://www.oma-ai.com/api/v1/models | jq '.data | length'

Provider Routing

OMA routes each request through the best available provider — automatically.

Intelligent Request Routing

Every model id resolves to an upstream provider. OMA routes each request to the model's catalog-mapped provider based on availability — with circuit-breaker protection against failing upstreams.

PrimaryVenice

Primary provider for the curated chat catalog. Handles chat and multimodal-input workloads with low-latency inference.

chatvisiontools

BudgetOMA-AI

Budget-tier models branded as OMA-AI — MiniMax M2.7, DeepSeek V4 Flash, and GLM-5.3 Flash on Gonka/OpenBroker. Same billing surface and API contract as the rest of the catalog.

chattoolsfree-tier

How Routing Works

Request arrives with a model id
OMA looks up the model→provider mapping in the catalog
The model's primary provider handles the request
On failure, the circuit breaker may retry a configured fallback for that mapped model
The public model id and customer price remain stable across that retry

Response Headers

Each response includes the public catalog provider label:

x-provider: venice | oma-ai

A mapped model may also have a compatible fallback provider configured. If a primary provider fails and that fallback is used, request content must be forwarded to the inference service that actually handles the retry. The public model id and listed customer price do not change. See Privacy for data-flow details.

Streaming

Real-time token-by-token responses via Server-Sent Events (SSE).

Why Stream?

Streaming provides real-time output as tokens are generated, reducing perceived latency from seconds to milliseconds. Essential for chat interfaces and interactive applications.

Enable Streaming

Add "stream": true to your request body. Responses arrive as Server-Sent Events (SSE) with Content-Type: text/event-stream.

Request with stream: true

curl https://www.oma-ai.com/api/v1/chat/completions \
	-H "Authorization: Bearer oma_sk_your_key_here" \
	-H "Content-Type: application/json" \
	-d '{
	"model": "llama-3.3-70b",
	"messages": [{"role": "user", "content": "Write a haiku about coding"}],
	"stream": true
	}'

Stream Response Format

Each event contains a chat.completion.chunk with a delta field containing partial content. The stream ends with data: [DONE].

SSE Response Stream

data: {"id":"chatcmpl-abc","object":"chat.completion.chunk","created":1717194000,"model":"llama-3.3-70b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

data: {"id":"chatcmpl-abc","object":"chat.completion.chunk","created":1717194000,"model":"llama-3.3-70b","choices":[{"index":0,"delta":{"content":"Code"},"finish_reason":null}]}

data: {"id":"chatcmpl-abc","object":"chat.completion.chunk","created":1717194000,"model":"llama-3.3-70b","choices":[{"index":0,"delta":{"content":" flows"},"finish_reason":null}]}

data: {"id":"chatcmpl-abc","object":"chat.completion.chunk","created":1717194000,"model":"llama-3.3-70b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

data: [DONE]

SDK Examples

from openai import OpenAI
client = OpenAI(
	api_key="oma_sk_your_key_here",
	base_url="https://www.oma-ai.com/api/v1"
)

stream = client.chat.completions.create(
	model="llama-3.3-70b",
	messages=[{"role": "user", "content": "Write a haiku about coding"}],
	stream=True
)

for chunk in stream:
	if chunk.choices[0].delta.content:
	print(chunk.choices[0].delta.content, end="")

Streaming Headers

Content-Type: text/event-stream

Cache-Control: no-cache

Connection: keep-alive

X-Accel-Buffering: no

Response Headers

x-provider: Provider handling the request

X-Correlation-ID: Request tracking ID

X-Key-Rotation-Recommended: Key age advisory

Rate Limits

One set of limits for every key — no plan tiers. Enforced per API key (or per IP for unauthenticated requests).

Limit	Value	Notes
Burst	60 requests	Token bucket capacity — the most requests you can fire at once.
Refill	20 req/s	How fast the bucket refills after a burst.
Sustained	60 RPM	Sliding-window requests-per-minute ceiling.

Per-Model Overrides

A few models carry their own request/token ceilings. Everything else uses the defaults above.

Model	RPM	TPM
`llama-3.3-70b`	60	120,000
`deepseek-v4-flash-0731-openbroker`	60	120,000
`kimi-k3`	60	120,000
`llama-3.2-3b`	120	240,000

Burst Handling

Token bucket algorithm allows short bursts above the sustained rate. The bucket holds 60 tokens and refills at 20 per second, so brief spikes succeed while sustained traffic settles at the RPM ceiling.

Rate Limit Headers

X-RateLimit-Limit - Requests per minute

X-RateLimit-Remaining - Requests left

X-RateLimit-Reset - Unix timestamp (seconds) when the limit resets

Retry-After - Seconds to wait before retrying (sent on 429/503)

Pricing

Per-model pricing. Input and output priced per million tokens.

Type	Input / 1M	Output / 1M	Context
GPT-5.6 Sol	$7.8125	$46.875	1M
Claude Opus 5	$7.50	$37.50	1M
Kimi K3	$4.6875	$23.4375	1M
GLM 5.2	$1.75	$5.50	1M
DeepSeek V4 Pro	$2.0625	$4.12625	1M
Qwen 3.8 Max	$3.125	$9.375	1M
Grok 4.5	$2.8375	$8.50	500K
Claude Sonnet 4.6	$4.50	$22.50	1M
Gemini 3.6 Flash	$1.171875	$5.859375	1M
Llama 3.3 70B	$0.875	$3.50	128K
Hermes 3 405B	$1.375	$3.75	128K
GLM 5.1	$1.925	$6.05	200K
Qwen 3.7 Plus	$0.625	$2.50	1M
Venice Uncensored 1.2	$0.25	$1.125	128K
Xiaomi MiMo V2.5	$0.50	$2.50	1M
DeepSeek V4 Flash (Venice)	$0.21875	$0.4375	1M
Mistral Small 4	$0.234375	$0.9375	256K
Gemma 4 31B	$0.15	$0.45	256K
MiniMax M2.7	$0.60	$0.60	200K
DeepSeek V4 Flash (OMA-AI)	$0.50	$0.50	400K
GLM-5.3 Flash (OMA-AI)	$0.30	$1.00	1M

Credit Deduction

Listed model prices are the final billed prices. $1 of credits buys 100 cents of usage at those rates, deducted per token in real-time.

No subscriptions. Funding methods have their own deposit limits.

Credits never expire while your account is active

x402 Payments

Fund prepaid credits with a signed USDC authorization on Base.

Verification moves funds

Calling /x402/verify validates the EIP-3009 authorization, broadcasts the USDC transfer, and waits for confirmation. It is not a dry run. Sign in with the paying wallet before starting; /x402/settle credits only the SIWE session whose linked wallet matches the USDC payer.

How funding works

Sign in via SIWE with the USDC payer wallet. Read the current treasury from /api/v1/deposit/usdc, then sign the intended amount and a fresh nonce.
POST the signed envelope to /x402/verify. The facilitator pays gas and transfers the authorized USDC; save the returned transaction hash.
POST the hash and network to /x402/settle with the paying wallet’s session cookie. This credits the confirmed payment once, preserving fractional credits. An API key does not authenticate this endpoint.
Use the account’s API key for inference against the prepaid balance. Ordinary chat requests do not issue a mid-request x402 challenge.

Payment signature and request sequence

Send base64-encoded JSON in PAYMENT-SIGNATURE. The v2 accepted fields specify the exact scheme, Base network, USDC asset, treasury, and atomic amount. The accepted amount and payTo must match authorization.value and authorization.to. The example below represents 1 USDC; replace every placeholder before signing.

{
  "x402Version": 2,
  "accepted": {
    "scheme": "exact",
    "network": "eip155:8453",
    "asset": "0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913",
    "amount": "1000000",
    "payTo": "<configured treasury address>"
  },
  "payload": {
    "signature": "<0x-prefixed 65-byte signature>",
    "authorization": {
      "from": "<payer wallet address>",
      "to": "<same configured treasury address>",
      "value": "1000000",
      "validAfter": "<Unix seconds>",
      "validBefore": "<Unix seconds>",
      "nonce": "<0x-prefixed 32-byte random nonce>"
    }
  }
}

Sign EIP-712 TransferWithAuthorization using the USDC domain: USD Coin, version 2, chain ID 8453, and the Base USDC contract. Times are Unix seconds, value is an integer string of atomic USDC units (6 decimals per token), and nonce is a fresh 32-byte value. Legacy raw-JSON PAYMENT-SIGNATURE envelopes and the older base64 X-PAYMENT header remain supported.

# Sign in via SIWE with the wallet that owns the USDC first.
# PAYMENT_SIGNATURE is a base64-encoded, signed v2 envelope.
# Calling verify broadcasts a transfer; it is not a dry run.
set -e
curl --fail-with-body -X POST https://www.oma-ai.com/x402/verify \
  -H "Content-Type: application/json" \
  -H "PAYMENT-SIGNATURE: $PAYMENT_SIGNATURE" > verification.json
jq -e '.valid == true' verification.json > /dev/null

# Credit the confirmed transfer to the same wallet's OMA account.
# Reuse the returned txHash and networkId; do not submit an amount.
jq '{txHash, networkId}' verification.json | \
  curl --fail-with-body https://www.oma-ai.com/x402/settle \
    -H "Content-Type: application/json" \
    --cookie "$OMA_SESSION_COOKIE" \
    --data-binary @-

# A successful response includes settled, receipt, and creditedCents.
# Now use that account's API key to spend the prepaid balance.

These shell examples require curl and jq. Keep private keys, session cookies, and signed payment envelopes out of logs and source control. The public funding URLs use /x402/*; the SDK base URL is https://www.oma-ai.com/api/v1.

Retry crediting with the same transaction

If funds moved but credits are pending, retry /x402/settle with the same hash and payer session. Crediting is idempotent and can recover the payer and amount from the confirmed USDC Transfer log after a server restart. A successful response includes a signed receipt string and creditedCents; PAYMENT-RESPONSE carries a base64 receipt summary.

A verification timeout can include a txHash even when valid is false. Check its Base receipt before authorizing another payment. Once confirmed, settle that hash. If no hash was returned, reconcile the signed nonce and wallet activity first.

Supported network

Base mainnet only (eip155:8453), using USDC. The service pins the USDC contract and configured treasury. Settlement accepts networkId base or eip155:8453 and derives the amount from verified payment evidence rather than a caller-supplied amount.

SDKs & Examples

Complete, working examples in every major language. OMA is fully OpenAI-compatible — just swap the base URL.

Python

pip install openai

Node.js

npm install openai

go get github.com/sashabaranov/go-openai

cURL

Built-in (no install)

Chat Completion

Send a multi-turn chat request with system prompt, temperature control, and token limits.

from openai import OpenAI

client = OpenAI(
	api_key="oma_sk_your_key_here",
	base_url="https://www.oma-ai.com/api/v1"
)

response = client.chat.completions.create(
	model="llama-3.3-70b",
	messages=[
	{"role": "system", "content": "You are a helpful assistant."},
	{"role": "user", "content": "Explain quantum computing in one sentence."}
	],
	temperature=0.7,
	max_tokens=256
)

print(response.choices[0].message.content)

Streaming

Real-time token-by-token responses via Server-Sent Events. Add "stream": true to your request.

from openai import OpenAI

client = OpenAI(
	api_key="oma_sk_your_key_here",
	base_url="https://www.oma-ai.com/api/v1"
)

stream = client.chat.completions.create(
	model="llama-3.3-70b",
	messages=[{"role": "user", "content": "Write a haiku about coding"}],
	stream=True
)

for chunk in stream:
	if chunk.choices[0].delta.content:
	print(chunk.choices[0].delta.content, end="")

List Models

Fetch the full model catalog with pricing and capabilities. No auth required for the public endpoint.

from openai import OpenAI

client = OpenAI(
	api_key="oma_sk_your_key_here",
	base_url="https://www.oma-ai.com/api/v1"
)

models = client.models.list()
for model in models.data:
	print(f"{model.id} ({model.get('category', 'chat')})")

Base URL

All examples use the same base URL. Just swap your API key and you're ready:

https://www.oma-ai.com/api/v1

Vercel AI SDK

For Next.js and React apps, use the Vercel AI SDK with OMA as the provider:

import { generateText } from 'ai';
import { createOpenAI } from '@ai-sdk/openai';

const oma = createOpenAI({
	baseURL: 'https://www.oma-ai.com/api/v1',
	apiKey: 'oma_sk_your_key_here',
});

const { text } = await generateText({
	model: oma('llama-3.3-70b'),
	prompt: 'Explain quantum computing in one sentence.',
});

Error Codes

Standard HTTP status codes returned by the API.

Response Format

Every error returns a JSON body with an error object carrying message, type, and code.

{
	"error": {
	"message": "Rate limit exceeded. Please slow down.",
	"type": "rate_limit_error",
	"code": "rate_limit_exceeded"
	}
}

Code	Status	Description
400	invalid_request_error	Malformed request body or invalid parameters. Schema failures return validation_error.
401	authentication_error	Missing or invalid API key.
402	payment_error	Insufficient credits. Deposit more USDC or settle via x402, then retry.
403	forbidden	Authenticated but not allowed. E.g. a resource/endpoint you lack permission for, or feature_disabled.
404	not_found	Route, model, key, or resource does not exist.
409	conflict	Request conflicts with current state (e.g. settle already applied to another account).
422	validation_error	Request is well-formed but semantically invalid; also returned as invalid_request_error for schema failures.
429	rate_limit_error	Rate limit exceeded. Wait Retry-After seconds, then retry. Specific codes: rate_limit_exceeded, daily_limit_exceeded.
500	internal_error	Unexpected server error (also server_error). Contact support if persistent.
503	service_unavailable	All upstream providers down. Try again shortly.

Handling Rate Limits

import time
from openai import OpenAI
client = OpenAI(api_key="oma_sk_...", base_url="https://www.oma-ai.com/api/v1")
for attempt in range(5):
	try:
	response = client.chat.completions.create(model="llama-3.3-70b", messages=[{"role": "user", "content": "Hi"}])
	break
	except Exception as e:
	if hasattr(e, 'status_code') and e.status_code == 429:
	time.sleep(2 ** attempt)
	else:
	raise

API reference

API Docs

OpenAI-compatible inference with wallet auth and x402 credit billing. One endpoint, one key, every curated model.

OpenAI-compatible21 modelsx402 on Base

Get API key Quick Start View Pricing

No credit card required · free website chat

Quick Start

Get up and running with the OMA API in under a minute. Drop-in replacement for OpenAI with zero vendor lock-in.

Get API Key

Create an account and generate your API key from the dashboard.

Add to Headers

Include your key in the Authorization header: Bearer oma_sk_...

Start Building

Use the OpenAI SDK with base_url=https://www.oma-ai.com/api/v1

Your First Request

cURL

curl https://www.oma-ai.com/api/v1/chat/completions \
	-H "Authorization: Bearer oma_sk_your_key_here" \
	-H "Content-Type: application/json" \
	-d '{
	"model": "llama-3.3-70b",
	"messages": [{"role": "user", "content": "Hello!"}]
	}'

Base URL

https://www.oma-ai.com/api/v1

SDK Install

npm install openai

Chat · Anthropic messages · curl

Chat (OpenAI-compatible)

cURL

curl https://www.oma-ai.com/api/v1/chat/completions \
		-H "Authorization: Bearer oma_sk_your_key_here" \
		-H "Content-Type: application/json" \
		-d '{
	"model": "llama-3.3-70b",
	"messages": [{"role": "user", "content": "Hello!"}]
	}'

Messages (Anthropic-compatible)

cURL

curl https://www.oma-ai.com/api/v1/messages \
	-H "x-api-key: oma_sk_your_key_here" \
	-H "anthropic-version: 2023-06-01" \
	-H "Content-Type: application/json" \
	-d '{
	"model": "claude-sonnet-4-6",
	"max_tokens": 256,
	"messages": [{"role": "user", "content": "Hello!"}]
	}'

Authentication

Programmatic inference uses a Bearer API key. Website accounts use email or wallet sessions; eligible free website chat also works without sign-in.

Free OMA-AI models (website chat)

Authorization Header (paid API)

Authorization: Bearer oma_sk_...

curl https://www.oma-ai.com/api/v1/chat/completions \
	-H "Authorization: Bearer oma_sk_your_key_here" \
	-H "Content-Type: application/json" \
	-d '{"model":"llama-3.3-70b","messages":[{"role":"user","content":"Hello"}]}'

API Key Security

Keys use the oma_sk_ prefix. Only the SHA-256 hash is stored. Keep your key secret and never commit it to version control.

Email Magic Link

How it works

Enter your email address on the login page
Click the link sent to your inbox within 5 minutes
Create an API key in the dashboard and save it when shown
Credits can be added via USDC on Base (card when Polar is enabled)

Wallet Auth (SIWE)

Connect any EVM wallet using Sign-In with Ethereum (SIWE). No email, no password. Your wallet address is your identity.

How it works

Click "Connect Wallet" on the login page
Sign a SIWE message with your wallet (no gas cost)
Create an API key in the dashboard and save it when shown
Use the key with any OpenAI-compatible SDK

Agent & IDE integrations

Use any OpenAI-compatible client: set base URL to www.oma-ai.com/api/v1 and Authorization: Bearer $OMA_KEY

One key, one base URL

Base URL: https://www.oma-ai.com/api/v1
API key: create at /dashboard → API keys (oma_sk_…)
Default sample model: llama-3.3-70b (see /models)
Claude Code / Anthropic SDK: also use POST /v1/messages
System instructions: For custom agent setups, see our agent-skill.md file.

OpenCode

Point OpenCode at OMA as a custom OpenAI-compatible provider. Env vars are the portable path; project config can pin the model.

# ~/.bashrc / shell profile (or OpenCode env UI)
export OPENAI_API_KEY="oma_sk_your_key_here"
export OPENAI_BASE_URL="https://www.oma-ai.com/api/v1"

# Optional: default model for sessions
export OPENAI_MODEL="llama-3.3-70b"

# Then run OpenCode as usual — it will hit OMA for chat completions.

If your OpenCode build uses OPENAI_API_BASE instead of OPENAI_BASE_URL, set that to the same https://www.oma-ai.com/api/v1 value.

Project config (optional)

// opencode.json / provider settings (shape varies by OpenCode version)
{
	"provider": {
	"oma": {
	"npm": "@ai-sdk/openai-compatible",
	"name": "OMA",
	"options": {
	"baseURL": "https://www.oma-ai.com/api/v1",
	"apiKey": "{env:OMA_API_KEY}"
	},
	"models": {
	"llama-3.3-70b": { "name": "llama-3.3-70b" }
	}
	}
	}
}

Create the key at /dashboard → API keys. Never commit oma_sk_… to git — use env vars.

Claude Code

Claude Code talks the Anthropic Messages API. OMA exposes a compatible POST /v1/messages that rewrites to the same catalog. Configure a custom Anthropic base URL + key:

# Claude Code / Anthropic SDK env
export ANTHROPIC_API_KEY="oma_sk_your_key_here"
export ANTHROPIC_BASE_URL="https://www.oma-ai.com/api"

# Note: base URL is the origin (no /v1). The client appends /v1/messages.
# Model ids are OMA catalog ids (not anthropic/claude-3-5-sonnet):
# kimi-k3 · claude-opus-5 · llama-3.3-70b · …

# One-shot smoke test (Messages API — x-api-key style)
curl https://www.oma-ai.com/api/v1/messages \
	-H "x-api-key: oma_sk_your_key_here" \
	-H "anthropic-version: 2023-06-01" \
	-H "Content-Type: application/json" \
	-d '{
	"model": "kimi-k3",
	"max_tokens": 64,
	"messages": [{"role": "user", "content": "ping"}]
	}'

# Same call with Bearer (OpenAI-style auth also accepted)
curl https://www.oma-ai.com/api/v1/messages \
	-H "Authorization: Bearer oma_sk_your_key_here" \
	-H "anthropic-version: 2023-06-01" \
	-H "Content-Type: application/json" \
	-d '{"model":"kimi-k3","max_tokens":64,"messages":[{"role":"user","content":"ping"}]}'

Prefer one auth style consistently. Streaming: set stream: true — OMA returns Anthropic-shaped SSE. List models: GET /v1/models with the same key.

Cursor / Continue / Cline

Any IDE that accepts an OpenAI-compatible endpoint:

Open model / provider settings → Custom OpenAI
API Base: https://www.oma-ai.com/api/v1
API Key: your oma_sk_…
Model: any id from GET /v1/models

// Example Continue config snippet (~/.continue/config.json)
{
	"models": [
	{
	"title": "OMA llama-3.3-70b",
	"provider": "openai",
	"model": "llama-3.3-70b",
	"apiBase": "https://www.oma-ai.com/api/v1",
	"apiKey": "oma_sk_your_key_here"
	}
	]
}

openai-python / openai-node

Drop-in replacement — only change base URL and key:

from openai import OpenAI

client = OpenAI(
	api_key="oma_sk_your_key_here",
	base_url="https://www.oma-ai.com/api/v1",
)

r = client.chat.completions.create(
	model="llama-3.3-70b",
	messages=[{"role": "user", "content": "Hello from OMA"}],
	max_tokens=64,
)
print(r.choices[0].message.content)

Endpoints Reference

OpenAI-compatible REST API endpoints.

The public inference, model, balance, usage, and deposit endpoints.

OpenAI-compatible multi-turn chat. The endpoint you will hit 95% of the time. Supports SSE streaming.

Request Body

modelstring (required). Any model id from /v1/models, e.g. "llama-3.3-70b"

messagesarray of { role, content }. role is system, user, assistant, or tool

temperaturenumber 0 to 2 (optional, default from upstream)

max_tokensinteger. Cap on completion length

streamboolean. Server-Sent Events when true (text/event-stream)

toolsarray of tool definitions (optional)

response_formatobject (optional), e.g. { type: "json_object" }

Response

{
	"id": "chatcmpl-9f3k2a",
	"object": "chat.completion",
	"created": 1717194000,
	"model": "llama-3.3-70b",
	"choices": [
	{
	"index": 0,
	"message": { "role": "assistant", "content": "Hello. How can I help today?" },
	"finish_reason": "stop"
	}
	],
	"usage": { "prompt_tokens": 18, "completion_tokens": 12, "total_tokens": 30 }
}

Available Models

21 curated chat models across four tiers, plus web search in the playground. Check the live catalog before making a request.

Chat Completions

Text and reasoning models from Venice (frontier/premium/mid/budget) and OMA-AI (MiniMax M2.7, DeepSeek V4 Flash, GLM-5.3 Flash).

Vision-capable models

A subset of premium / frontier models accept image inputs alongside text in the OpenAI multimodal content array shape.

Budget7 models

Fast, cost-effective models for simple tasks and prototyping.

Model	Input / 1M	Output / 1M	Context	Capabilities
`xiaomi-mimo-v2-5` Xiaomi MiMo V2.5	$0.50	$2.50	1M	chatreasoningfast
`deepseek-v4-flash-0731` DeepSeek V4 Flash (Venice)	$0.21875	$0.4375	1M	chatreasoningfast
`mistral-small-2603` Mistral Small 4	$0.234375	$0.9375	256K	chatfastagents
`google-gemma-4-31b-it` Gemma 4 31B	$0.15	$0.45	256K	chatfast
`minimax-m2-7` MiniMax M2.7	$0.60	$0.60	200K	chatreasoningcodingagents
`deepseek-v4-flash-0731-openbroker` DeepSeek V4 Flash (OMA-AI)	$0.50	$0.50	400K	chatreasoningfastagents
`zai-org-glm-5-3` GLM-5.3 Flash (OMA-AI)	$0.30	$1.00	1M	chatreasoningmultimodalfastagents

Mid-Range5 models

Balanced performance and cost for everyday workloads.

Model	Input / 1M	Output / 1M	Context	Capabilities
`llama-3.3-70b` Llama 3.3 70B	$0.875	$3.50	128K	chatagents
`hermes-3-llama-3.1-405b` Hermes 3 405B	$1.375	$3.75	128K	chatreasoningagents
`zai-org-glm-5-1` GLM 5.1	$1.925	$6.05	200K	chatreasoningagents
`qwen-3-7-plus` Qwen 3.7 Plus	$0.625	$2.50	1M	chatreasoningagents
`venice-uncensored-1-2` Venice Uncensored 1.2	$0.25	$1.125	128K	chatuncensored

Premium7 models

High-performance models for demanding tasks.

Model	Input / 1M	Output / 1M	Context	Capabilities
`kimi-k3` Kimi K3	$4.6875	$23.4375	1M	chatreasoningagents
`zai-org-glm-5-2` GLM 5.2	$1.75	$5.50	1M	chatreasoningagents
`deepseek-v4-pro` DeepSeek V4 Pro	$2.0625	$4.12625	1M	chatreasoningcodingagents
`qwen-3-8-max` Qwen 3.8 Max	$3.125	$9.375	1M	chatreasoningcodingagents
`grok-4-5` Grok 4.5	$2.8375	$8.50	500K	chatreasoningcodingagents
`claude-sonnet-4-6` Claude Sonnet 4.6	$4.50	$22.50	1M	chatreasoningcodingagents
`gemini-3-6-flash` Gemini 3.6 Flash	$1.171875	$5.859375	1M	chatreasoningcodingvisionagents

Frontier2 models

Top-tier models for complex reasoning and generation tasks.

Model	Input / 1M	Output / 1M	Context	Capabilities
`openai-gpt-56-sol` GPT-5.6 Sol	$7.8125	$46.875	1M	chatreasoningcodingagents
`claude-opus-5` Claude Opus 5	$7.50	$37.50	1M	chatreasoningcodingagents

List All Models

Use the GET /v1/models endpoint to fetch the current model catalog with pricing and capabilities.

curl https://www.oma-ai.com/api/v1/models | jq '.data | length'

Provider Routing

OMA routes each request through the best available provider — automatically.

Intelligent Request Routing

Every model id resolves to an upstream provider. OMA routes each request to the model's catalog-mapped provider based on availability — with circuit-breaker protection against failing upstreams.

PrimaryVenice

Primary provider for the curated chat catalog. Handles chat and multimodal-input workloads with low-latency inference.

chatvisiontools

BudgetOMA-AI

Budget-tier models branded as OMA-AI — MiniMax M2.7, DeepSeek V4 Flash, and GLM-5.3 Flash on Gonka/OpenBroker. Same billing surface and API contract as the rest of the catalog.

chattoolsfree-tier

How Routing Works

Request arrives with a model id
OMA looks up the model→provider mapping in the catalog
The model's primary provider handles the request
On failure, the circuit breaker may retry a configured fallback for that mapped model
The public model id and customer price remain stable across that retry

Response Headers

Each response includes the public catalog provider label:

x-provider: venice | oma-ai

Streaming

Real-time token-by-token responses via Server-Sent Events (SSE).

Why Stream?

Streaming provides real-time output as tokens are generated, reducing perceived latency from seconds to milliseconds. Essential for chat interfaces and interactive applications.

Enable Streaming

Add "stream": true to your request body. Responses arrive as Server-Sent Events (SSE) with Content-Type: text/event-stream.

Request with stream: true

curl https://www.oma-ai.com/api/v1/chat/completions \
	-H "Authorization: Bearer oma_sk_your_key_here" \
	-H "Content-Type: application/json" \
	-d '{
	"model": "llama-3.3-70b",
	"messages": [{"role": "user", "content": "Write a haiku about coding"}],
	"stream": true
	}'

Stream Response Format

Each event contains a chat.completion.chunk with a delta field containing partial content. The stream ends with data: [DONE].

SSE Response Stream

data: {"id":"chatcmpl-abc","object":"chat.completion.chunk","created":1717194000,"model":"llama-3.3-70b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

data: {"id":"chatcmpl-abc","object":"chat.completion.chunk","created":1717194000,"model":"llama-3.3-70b","choices":[{"index":0,"delta":{"content":"Code"},"finish_reason":null}]}

data: {"id":"chatcmpl-abc","object":"chat.completion.chunk","created":1717194000,"model":"llama-3.3-70b","choices":[{"index":0,"delta":{"content":" flows"},"finish_reason":null}]}

data: {"id":"chatcmpl-abc","object":"chat.completion.chunk","created":1717194000,"model":"llama-3.3-70b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

data: [DONE]

SDK Examples

from openai import OpenAI
client = OpenAI(
	api_key="oma_sk_your_key_here",
	base_url="https://www.oma-ai.com/api/v1"
)

stream = client.chat.completions.create(
	model="llama-3.3-70b",
	messages=[{"role": "user", "content": "Write a haiku about coding"}],
	stream=True
)

for chunk in stream:
	if chunk.choices[0].delta.content:
	print(chunk.choices[0].delta.content, end="")

Streaming Headers

Content-Type: text/event-stream

Cache-Control: no-cache

Connection: keep-alive

X-Accel-Buffering: no

Response Headers

x-provider: Provider handling the request

X-Correlation-ID: Request tracking ID

X-Key-Rotation-Recommended: Key age advisory

Rate Limits

One set of limits for every key — no plan tiers. Enforced per API key (or per IP for unauthenticated requests).

Limit	Value	Notes
Burst	60 requests	Token bucket capacity — the most requests you can fire at once.
Refill	20 req/s	How fast the bucket refills after a burst.
Sustained	60 RPM	Sliding-window requests-per-minute ceiling.

Per-Model Overrides

A few models carry their own request/token ceilings. Everything else uses the defaults above.

Model	RPM	TPM
`llama-3.3-70b`	60	120,000
`deepseek-v4-flash-0731-openbroker`	60	120,000
`kimi-k3`	60	120,000
`llama-3.2-3b`	120	240,000

Burst Handling

Rate Limit Headers

X-RateLimit-Limit - Requests per minute

X-RateLimit-Remaining - Requests left

X-RateLimit-Reset - Unix timestamp (seconds) when the limit resets

Retry-After - Seconds to wait before retrying (sent on 429/503)

Pricing

Per-model pricing. Input and output priced per million tokens.

Type	Input / 1M	Output / 1M	Context
GPT-5.6 Sol	$7.8125	$46.875	1M
Claude Opus 5	$7.50	$37.50	1M
Kimi K3	$4.6875	$23.4375	1M
GLM 5.2	$1.75	$5.50	1M
DeepSeek V4 Pro	$2.0625	$4.12625	1M
Qwen 3.8 Max	$3.125	$9.375	1M
Grok 4.5	$2.8375	$8.50	500K
Claude Sonnet 4.6	$4.50	$22.50	1M
Gemini 3.6 Flash	$1.171875	$5.859375	1M
Llama 3.3 70B	$0.875	$3.50	128K
Hermes 3 405B	$1.375	$3.75	128K
GLM 5.1	$1.925	$6.05	200K
Qwen 3.7 Plus	$0.625	$2.50	1M
Venice Uncensored 1.2	$0.25	$1.125	128K
Xiaomi MiMo V2.5	$0.50	$2.50	1M
DeepSeek V4 Flash (Venice)	$0.21875	$0.4375	1M
Mistral Small 4	$0.234375	$0.9375	256K
Gemma 4 31B	$0.15	$0.45	256K
MiniMax M2.7	$0.60	$0.60	200K
DeepSeek V4 Flash (OMA-AI)	$0.50	$0.50	400K
GLM-5.3 Flash (OMA-AI)	$0.30	$1.00	1M

Credit Deduction

Listed model prices are the final billed prices. $1 of credits buys 100 cents of usage at those rates, deducted per token in real-time.

No subscriptions. Funding methods have their own deposit limits.

Credits never expire while your account is active

x402 Payments

Fund prepaid credits with a signed USDC authorization on Base.

Verification moves funds

How funding works

Sign in via SIWE with the USDC payer wallet. Read the current treasury from /api/v1/deposit/usdc, then sign the intended amount and a fresh nonce.
POST the signed envelope to /x402/verify. The facilitator pays gas and transfers the authorized USDC; save the returned transaction hash.
POST the hash and network to /x402/settle with the paying wallet’s session cookie. This credits the confirmed payment once, preserving fractional credits. An API key does not authenticate this endpoint.
Use the account’s API key for inference against the prepaid balance. Ordinary chat requests do not issue a mid-request x402 challenge.

Payment signature and request sequence

{
  "x402Version": 2,
  "accepted": {
    "scheme": "exact",
    "network": "eip155:8453",
    "asset": "0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913",
    "amount": "1000000",
    "payTo": "<configured treasury address>"
  },
  "payload": {
    "signature": "<0x-prefixed 65-byte signature>",
    "authorization": {
      "from": "<payer wallet address>",
      "to": "<same configured treasury address>",
      "value": "1000000",
      "validAfter": "<Unix seconds>",
      "validBefore": "<Unix seconds>",
      "nonce": "<0x-prefixed 32-byte random nonce>"
    }
  }
}

# Sign in via SIWE with the wallet that owns the USDC first.
# PAYMENT_SIGNATURE is a base64-encoded, signed v2 envelope.
# Calling verify broadcasts a transfer; it is not a dry run.
set -e
curl --fail-with-body -X POST https://www.oma-ai.com/x402/verify \
  -H "Content-Type: application/json" \
  -H "PAYMENT-SIGNATURE: $PAYMENT_SIGNATURE" > verification.json
jq -e '.valid == true' verification.json > /dev/null

# Credit the confirmed transfer to the same wallet's OMA account.
# Reuse the returned txHash and networkId; do not submit an amount.
jq '{txHash, networkId}' verification.json | \
  curl --fail-with-body https://www.oma-ai.com/x402/settle \
    -H "Content-Type: application/json" \
    --cookie "$OMA_SESSION_COOKIE" \
    --data-binary @-

# A successful response includes settled, receipt, and creditedCents.
# Now use that account's API key to spend the prepaid balance.

Retry crediting with the same transaction

Supported network

SDKs & Examples

Complete, working examples in every major language. OMA is fully OpenAI-compatible — just swap the base URL.

Python

pip install openai

Node.js

npm install openai

go get github.com/sashabaranov/go-openai

cURL

Built-in (no install)

Chat Completion

Send a multi-turn chat request with system prompt, temperature control, and token limits.

from openai import OpenAI

client = OpenAI(
	api_key="oma_sk_your_key_here",
	base_url="https://www.oma-ai.com/api/v1"
)

response = client.chat.completions.create(
	model="llama-3.3-70b",
	messages=[
	{"role": "system", "content": "You are a helpful assistant."},
	{"role": "user", "content": "Explain quantum computing in one sentence."}
	],
	temperature=0.7,
	max_tokens=256
)

print(response.choices[0].message.content)

Streaming

Real-time token-by-token responses via Server-Sent Events. Add "stream": true to your request.

from openai import OpenAI

client = OpenAI(
	api_key="oma_sk_your_key_here",
	base_url="https://www.oma-ai.com/api/v1"
)

stream = client.chat.completions.create(
	model="llama-3.3-70b",
	messages=[{"role": "user", "content": "Write a haiku about coding"}],
	stream=True
)

for chunk in stream:
	if chunk.choices[0].delta.content:
	print(chunk.choices[0].delta.content, end="")

List Models

Fetch the full model catalog with pricing and capabilities. No auth required for the public endpoint.

from openai import OpenAI

client = OpenAI(
	api_key="oma_sk_your_key_here",
	base_url="https://www.oma-ai.com/api/v1"
)

models = client.models.list()
for model in models.data:
	print(f"{model.id} ({model.get('category', 'chat')})")

Base URL

All examples use the same base URL. Just swap your API key and you're ready:

https://www.oma-ai.com/api/v1

Vercel AI SDK

For Next.js and React apps, use the Vercel AI SDK with OMA as the provider:

import { generateText } from 'ai';
import { createOpenAI } from '@ai-sdk/openai';

const oma = createOpenAI({
	baseURL: 'https://www.oma-ai.com/api/v1',
	apiKey: 'oma_sk_your_key_here',
});

const { text } = await generateText({
	model: oma('llama-3.3-70b'),
	prompt: 'Explain quantum computing in one sentence.',
});

Error Codes

Standard HTTP status codes returned by the API.

Response Format

Every error returns a JSON body with an error object carrying message, type, and code.

{
	"error": {
	"message": "Rate limit exceeded. Please slow down.",
	"type": "rate_limit_error",
	"code": "rate_limit_exceeded"
	}
}

Code	Status	Description
400	invalid_request_error	Malformed request body or invalid parameters. Schema failures return validation_error.
401	authentication_error	Missing or invalid API key.
402	payment_error	Insufficient credits. Deposit more USDC or settle via x402, then retry.
403	forbidden	Authenticated but not allowed. E.g. a resource/endpoint you lack permission for, or feature_disabled.
404	not_found	Route, model, key, or resource does not exist.
409	conflict	Request conflicts with current state (e.g. settle already applied to another account).
422	validation_error	Request is well-formed but semantically invalid; also returned as invalid_request_error for schema failures.
429	rate_limit_error	Rate limit exceeded. Wait Retry-After seconds, then retry. Specific codes: rate_limit_exceeded, daily_limit_exceeded.
500	internal_error	Unexpected server error (also server_error). Contact support if persistent.
503	service_unavailable	All upstream providers down. Try again shortly.

Handling Rate Limits

import time
from openai import OpenAI
client = OpenAI(api_key="oma_sk_...", base_url="https://www.oma-ai.com/api/v1")
for attempt in range(5):
	try:
	response = client.chat.completions.create(model="llama-3.3-70b", messages=[{"role": "user", "content": "Hi"}])
	break
	except Exception as e:
	if hasattr(e, 'status_code') and e.status_code == 429:
	time.sleep(2 ** attempt)
	else:
	raise