Migrate from OpenAI

Venice AI is a drop-in replacement for OpenAI. Same SDK, same code — just change two lines. Get privacy-first inference, uncensored models, and competitive pricing.

The 2-Line Migration

# Before (OpenAI)
from openai import OpenAI
client = OpenAI()

# After (Venice) — change api_key and base_url
from openai import OpenAI
client = OpenAI(
    api_key="your-venice-api-key",           # ← Change 1
    base_url="https://api.venice.ai/api/v1",  # ← Change 2
)

// Before (OpenAI)
import OpenAI from "openai";
const client = new OpenAI();

// After (Venice)
import OpenAI from "openai";
const client = new OpenAI({
  apiKey: "your-venice-api-key",
  baseURL: "https://api.venice.ai/api/v1",
});

# Before
curl https://api.openai.com/v1/chat/completions ...

# After — just change the URL and key
curl https://api.venice.ai/api/v1/chat/completions ...

Generate a key from venice.ai/settings/api or follow the API key guide.

Environment variables

# Before
OPENAI_API_KEY=sk-...
OPENAI_BASE_URL=https://api.openai.com/v1

# After
OPENAI_API_KEY=your-venice-api-key
OPENAI_BASE_URL=https://api.venice.ai/api/v1

Many libraries and tools read OPENAI_API_KEY and OPENAI_BASE_URL automatically. Updating these env vars may be all you need. Some tools still use OPENAI_API_BASE for the same URL.

Models

Pass a Venice model ID, a trait such as default or most_uncensored, or, for chat and embedding models, a familiar OpenAI-style name. Venice maps OpenAI chat and embedding model names through compatibility mapping; other endpoints, including audio, need Venice model IDs. Traits always resolve to the current model for that role — see model traits. Browse the live catalog on Text models and Pricing. For a first-request walkthrough, see the Quickstart.

Feature compatibility

Feature	OpenAI	Venice	Notes
Chat Completions	✅	✅	Compatible. Unknown top-level fields, including OpenAI-only ones such as `logit_bias` and `service_tier`, return HTTP 400
Streaming	✅	✅	Same SSE format, plus a final usage chunk, see Streaming format
Function Calling	✅	✅	Same `tools` parameter
Structured Output	✅	✅	Same `response_format`
Vision	✅	✅	Same content array format
Embeddings	✅	✅	Text only; token arrays return HTTP 400
Image Generation	✅	✅	OpenAI-compatible via `/images/generations`
Image Edits	✅	❌	No `/images/edits` endpoint. Use the native `/image/edit` endpoint
TTS	✅	✅	Compatible request shape; use Venice model IDs (OpenAI audio model names return HTTP 400)
STT	✅	✅	Compatible request shape; use Venice model IDs. `response_format` is `json` or `text` only. No `/audio/translations`
Responses API	✅	✅	Beta. Native for OpenAI models, translated for others, see limitations
Assistants API	✅	❌	Use Characters or function calling
Batch API	✅	❌	Not yet available
Fine-tuning	✅	❌	Not available

For native image options beyond the OpenAI-compatible endpoint, see Image generation.

Streaming format

Chat Completions streams use the OpenAI chunk format and end with data: [DONE]. Two details differ:

Usage chunk. Every stream ends with a usage chunk before [DONE], even when stream_options.include_usage is false. Its choices array is empty and it carries usage and a cost object. Clients that read choices[0] on every chunk should skip chunks with no choices.
Errors after the stream starts. The HTTP status is already 200, so the error arrives as a data: event with an error object (message, type, code), followed by data: [DONE]. When the model is overloaded, the object also includes retry_after in seconds.

data: {"error":{"message":"...","type":"server_error","code":"...","param":null}}

data: [DONE]

Venice-only features

Pass Venice-specific options through extra_body as venice_parameters. Built-in web search is the usual first extra:

response = client.chat.completions.create(
    model="default",
    messages=[{"role": "user", "content": "Latest AI news today"}],
    extra_body={
        "venice_parameters": {
            "enable_web_search": "auto"
        }
    },
)

Unknown keys inside venice_parameters are ignored rather than rejected, so check the spelling of each key against the API reference. The same pattern covers web scraping, citations, and Characters. If a client cannot change the request body, append a model feature suffix such as default:enable_web_search=auto. Venice also has native video, music, web retrieval, and x402 APIs alongside the OpenAI-compatible surface.

Frameworks

Most AI frameworks work with Venice by changing the base URL:

LangChain

base_url in ChatOpenAI; check_embedding_ctx_length=False on OpenAIEmbeddings

Vercel AI SDK

baseURL in createOpenAI

LlamaIndex

api_base on the OpenAI-compatible client

CrewAI

OPENAI_API_BASE env var

PydanticAI

OpenAI-compatible model with Venice base URL

Cursor

Custom API endpoint in settings

Claude Code

Route Claude Code through Venice

Codex CLI

config.toml model provider

Aider

OPENAI_API_BASE env var

More coding agents and tools are listed in AI Agents.

Uncensored models

Venice’s private uncensored models have no content filtering, making them suitable for:

Creative writing without guardrails
Security research and red teaming
Honest analysis without refusal patterns
Medical and legal information without extra disclaimers

Use the most_uncensored trait or a current uncensored model ID from Text models.

Get Your API Key

Generate a Venice API key and start migrating in minutes

Docs

Getting Started

Text & Chat

Image

Video

Audio

Search & RAG

Agents & Integrations

Coding Tools

Agent Tooling

SDKs & Frameworks

The 2-Line Migration

Environment variables

Models

Feature compatibility

Streaming format

Venice-only features

Frameworks

LangChain

Vercel AI SDK

LlamaIndex

CrewAI

PydanticAI

Cursor

Claude Code

Codex CLI

Aider

Uncensored models

Get Your API Key

​The 2-Line Migration

​Environment variables

​Models

​Feature compatibility

​Streaming format

​Venice-only features

​Frameworks

LangChain

Vercel AI SDK

LlamaIndex

CrewAI

PydanticAI

Cursor

Claude Code

Codex CLI

Aider

​Uncensored models

Get Your API Key

The 2-Line Migration

Environment variables

Models

Feature compatibility

Streaming format

Venice-only features

Frameworks

Uncensored models