Venice API Overview

Venice.ai is an OpenAI-compatible inference platform for text, image, audio, video, embeddings, and typed decisions. One API - two ways to pay: a traditional API key (Venice account), or a wallet (x402, USDC on Base or Solana, no account required).

Use when

• You're writing code against api.venice.ai for the first time.
• You need to decide between API-key and x402/wallet authentication.
• You want a quick map of which endpoint to call for which task.
• You need to understand the common response headers (x-ratelimit-*, PAYMENT-REQUIRED, deprecation headers, etc.).

Base URL

All endpoints live under:

https://api.venice.ai/api/v1

The OpenAPI spec is served at https://api.venice.ai/api/v1/swagger.yaml (info.version is a YYYYMMDD.HHMMSS timestamp; read it from the live spec).

Authentication

Scheme	Header	Best for
BearerAuth	Authorization: Bearer $VENICE_API_KEY	Server-side apps, account management, usage analytics, DIEM / bundled credits
siwx (x402)	SIGN-IN-WITH-X: <base64 SIWX JSON> (legacy X-Sign-In-With-X also accepted)	No account, pay-as-you-go with USDC on Base or Solana, serverless / agents

Not every endpoint accepts both:

Endpoints	Accepts
All inference: chat, responses, embeddings, decisions, image, audio (speech, transcriptions, voices, queue/retrieve/complete, voice-changer queue/retrieve/complete), video queue/retrieve/complete, augment, POST /crypto/rpc/{network}	Bearer or SIWX
/api_keys/* (except generate_web3_key), /billing/* (except the retired /billing/usage), /characters/*	Bearer only (a SIGN-IN-WITH-X header alone gets 401 Authentication failed)
/x402/balance/{wallet}, /x402/transactions/{wallet}	SIWX only (signer must own the wallet)
/models*, /image/styles, /video/quote (except upscale models such as topaz-video-upscale: Bearer key required, so wallets can't quote them), /audio/quote, /audio/voice-changer/quote, /crypto/rpc/networks, /tee/attestation, /tee/signature, /api_keys/generate_web3_key, POST /x402/top-up	No auth needed

On any route that needs credentials - Bearer-only ones included - a request with no Authorization and no SIGN-IN-WITH-X header gets 402 with an x402 discovery body (payment options + SIWX challenge + authOptions), not 401. See venice-auth.

# Bearer
curl https://api.venice.ai/api/v1/chat/completions \
  -H "Authorization: Bearer $VENICE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"zai-org-glm-5-2","messages":[{"role":"user","content":"hi"}]}'

// x402 wallet via the SDK (EVM / Base wallets)
import { VeniceClient } from 'venice-x402-client'
const v = new VeniceClient(process.env.WALLET_KEY!)
await v.models()

Endpoint map

Inference

Category	Endpoints	Skill
Chat	POST /chat/completions	venice-chat
Responses (Alpha)	POST /responses	venice-responses
Embeddings	POST /embeddings	venice-embeddings
Decisions (Beta)	POST /decisions (alias POST /systemone)	venice-decisions
Image gen	POST /image/generate, POST /images/generations, GET /image/styles	venice-image-generate
Image edit	POST /image/edit, POST /image/multi-edit, POST /image/upscale, POST /image/background-remove	venice-image-edit
TTS	POST /audio/speech, POST /audio/voices (voice cloning)	venice-audio-speech
STT	POST /audio/transcriptions	venice-audio-transcription
Music / audio (async)	POST /audio/quote, /audio/queue, /audio/retrieve, /audio/complete	venice-audio-music
Voice changer (async)	POST /audio/voice-changer/quote, /queue, /retrieve, /complete	venice-audio-voice-changer
Video (async)	POST /video/quote, /video/queue, /video/retrieve, /video/complete	venice-video

Voice changer: the endpoints are in the spec, but no voice-changer model is publicly listed today - check GET /models?type=music for a model with voice_changer: true first.

Catalog

Category	Endpoints	Skill
Models	GET /models, /models/traits, /models/compatibility_mapping	venice-models
Characters	GET /characters, /characters/{slug}, /characters/{slug}/reviews	venice-characters

GET /models?type= accepts text (default), image, video, music, tts, asr, embedding, upscale, inpaint, decision, plus the filters all and code.

Account, billing, wallet

Category	Endpoints	Skill
API keys	GET/POST/PATCH/DELETE /api_keys, GET /api_keys/{id}, /api_keys/rate_limits, /api_keys/rate_limits/log, /api_keys/generate_web3_key	venice-api-keys
Billing (Beta)	GET /billing/balance, /billing/usage-history, /billing/usage-analytics	venice-billing
x402 wallet	POST /x402/top-up, GET /x402/balance/{wallet}, GET /x402/transactions/{wallet}	venice-x402

Utility

Category	Endpoints	Skill
Crypto RPC proxy	GET /crypto/rpc/networks, POST /crypto/rpc/{network}	venice-crypto-rpc
Augment	POST /augment/text-parser, /augment/scrape, /augment/search	venice-augment
TEE verification	GET /tee/attestation, GET /tee/signature (public, 10 req/min per IP)	venice-text-routing

Retired (return 410 Gone)

Endpoint	Replacement
GET /billing/usage - sunset 2026-09-16	GET /billing/usage-history (cursor pagination: pageSize + nextCursor, startTimestamp / endTimestamp)
POST /video/transcriptions	POST /chat/completions with a video_url part on a model whose capabilities.supportsVideoInput is true

Both answer with 410 plus Deprecation and Link: <…>; rel="successor-version" headers (/billing/usage also sends Sunset). They need no auth; a per-IP limit of 60 requests/minute returns 429 to clients that keep polling.

Response headers to watch

Header	When	Meaning
x-ratelimit-limit-requests / -remaining-requests / -reset-requests	Most inference responses	Request window for your account - per model (the tighter of per-minute and per-day), or per endpoint on video / audio-generation routes and /augment/scrape / /augment/search. Reset is a Unix timestamp in milliseconds.
x-ratelimit-limit-tokens / -remaining-tokens / -reset-tokens	Token-limited text models	Tokens-per-minute window (reset in ms).
x-ratelimit-remaining / x-ratelimit-resets	Most responses on routes that need credentials	The error budget (failed requests allowed in the 30 s window), not your request quota. Read before the current response is counted, so a failed response showing 1 means none are left. Reset in ms.
x-venice-balance-usd / x-venice-balance-diem	Inference responses	Spendable balance when the request started (x402 callers see their wallet credit here). Omitted when that balance is zero; the USD figure excludes bundled and earned credits.
x-venice-version	Inference responses	Server revision - handy in bug reports.
x-venice-deprecated, x-venice-model-deprecation-date, x-venice-model-deprecation-warning, x-venice-deprecated-replacement	Requests to a model scheduled for retirement (chat, image, video queue)	Retirement date and suggested replacement.
PAYMENT-REQUIRED	Every x402 402: no credentials, wallet below the minimum balance, /x402/top-up discovery, /x402/balance / /x402/transactions without SIWX	Base64 JSON of the x402 v2 payment-required object (accepts[], plus the sign-in-with-x challenge except on /x402/top-up).
Retry-After	429 "model overloaded"	Seconds to wait (default 30).
Deprecation / Sunset / Link	Retired endpoints	See table above.
Content-Encoding	When you send Accept-Encoding (gzip, br, deflate)	Compressed responses on any route, once the body is large enough to be worth compressing. The spec documents it on chat, embeddings, decisions and image generation.

The spec also documents an X-Balance-Remaining header on x402 responses, but current server code does not set it - read x-venice-balance-usd or call GET /x402/balance/{wallet} instead.

Pricing model at a glance

• Pricing is dynamic per request, metered in USD. Paid endpoints in the spec carry an x-payment-info block (price.mode: dynamic, min: "0.001", max: "10.00" USD); POST /x402/top-up is 5-10000. Read-only routes (/models, quotes) have none.
• API-key accounts charge each request to a single currency, picked in the order DIEM → earned credits → bundled credits → USD (see venice-billing). Per-key consumptionLimits can cap USD / DIEM spend.
• x402 wallets spend a prepaid USDC credit balance topped up on Base or Solana (minimum top-up $5; a request needs at least $0.10 of balance to start). An EVM wallet linked to a Venice account with staked DIEM spends DIEM first.
• The per-model price is on GET /models → model_spec.pricing, already including any promotion active for your account. Video has no price there - use POST /video/quote; music / voice changer have exact quotes via /audio/quote and /audio/voice-changer/quote. See venice-models.

Standard error shape

Most errors are:

{ "error": "Human-readable message" }

Schema validation failures (400) add a details tree and an issues array:

{ "error": "Invalid request parameters", "details": { "_errors": [], "type": { "_errors": ["Invalid enum value…"] } }, "issues": [ … ] }

Exceptions worth knowing: context-length overflows on chat return an OpenAI-style object ({ "error": { "message", "type": "invalid_request_error", "param": "messages", "code": "context_length_exceeded" } }), upstream provider rejections return the provider's message (plus request_id on routes that track one, such as chat), and 402 on x402 returns structured top-up data. A POST whose Content-Type is neither application/json nor multipart/form-data gets 400 "'Content-Type' must be 'application/json'" before auth runs. See venice-errors for the full table and retry strategy.

OpenAI compatibility - what works and what doesn't

• Drop-in: /chat/completions, /responses, /embeddings, /images/generations, /audio/speech, /audio/transcriptions, /models.
• Accepted but ignored for compat: user, store (on chat). user is not an alias of Venice's anon_user_id (it does partition the error budget - see venice-errors).
• Venice-only extensions live under venice_parameters - the full set on /chat/completions, a seven-field subset (character_slug, enable_e2ee, enable_web_search, enable_web_scraping, enable_web_citations, include_venice_system_prompt, include_search_results_in_stream) on /responses.
• Model feature suffixes (e.g. zai-org-glm-5-1:enable_web_search=on, kimi-k2-6:strip_thinking_response=true&disable_thinking=true) flip venice_parameters via the model ID - see venice-chat.
• A trait name (default_reasoning) or a legacy alias (gpt-4o) can be sent as model; Venice resolves it. See venice-models.

Versioning

• info.version in swagger.yaml is a timestamp (YYYYMMDD.HHMMSS). There is no /v2; features roll forward on the single /api/v1 surface and are guarded by:
  • Alpha/Beta labels in endpoint descriptions (Responses is Alpha; Decisions and Billing are Beta).
  • betaModel / deprecation metadata and capability flags on /models.
• Retired endpoints answer 410 with Deprecation / Link headers rather than disappearing silently.
• Always check the model's model_spec.capabilities (supportsWebSearch, supportsReasoning, supportsE2EE, supportsXSearch, supportsMultipleImages, supportsFunctionCalling, supportsAudioInput, supportsVideoInput, …) before relying on a feature.

Fast start checklist

1. Read venice-auth and choose Bearer vs x402.
2. GET /models?type=… - pick a model and note its model_spec.constraints and model_spec.pricing.
3. Wire up one happy-path call from the matching skill.
4. Add error handling using venice-errors (402, 422, 429).
5. Hook up observability via x-ratelimit-* / x-venice-balance-* headers, /billing/usage-history (ADMIN key only), or /x402/transactions/{wallet} (wallet callers).