Venice

Venice API

Venice is a privacy-first, uncensored, OpenAI-compatible AI platform covering text, image, video, audio, music, embeddings, web search and scraping, document parsing, and blockchain RPC. Zero data retention.

TL;DR

• Base URL: https://api.venice.ai/api/v1
• Auth: Authorization: Bearer <VENICE_API_KEY>, or an x402 wallet (USDC on Base or Solana) with no key and no account
• OpenAI-compatible: use any OpenAI SDK and change only base_url and the model ID
• Never hardcode model IDs. Resolve them at runtime from GET /models and GET /models/traits. They rotate.
• OpenAPI spec: https://docs.venice.ai/swagger.yaml
• Deeper per-endpoint skills: https://github.com/veniceai/skills
• MCP server: https://github.com/veniceai/venice-mcp-server

First call

curl https://api.venice.ai/api/v1/chat/completions \
  -H "Authorization: Bearer $VENICE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "kimi-k3",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Python: OpenAI(base_url="https://api.venice.ai/api/v1", api_key=VENICE_API_KEY)

Picking a model

Resolve a trait to a current model ID instead of writing a slug into code:

curl https://api.venice.ai/api/v1/models/traits \
  -H "Authorization: Bearer $VENICE_API_KEY"

The response maps trait names to whichever model currently fills that role. Text traits are default, most_intelligent, most_uncensored, default_reasoning, default_vision, default_code, and function_calling_default; image traits include default, fastest, highest_quality, and most_uncensored. Filter the full catalog with GET /models?type=image|video|audio|tts|embedding. Before relying on a feature, check that model's model_spec.capabilities flags (supportsWebSearch, supportsReasoning, supportsE2EE, supportsFunctionCalling, supportsVision, and similar). Per-model pricing is on model_spec.pricing.

Common text models, as a starting point rather than a fixed list:

Confirm against GET /models before you ship. This table is a snapshot and the catalog moves; the capability flags on each model are the authoritative answer.

Endpoint map

venice_parameters

Venice-only features ride in a venice_parameters object on /chat/completions:

• enable_web_search: "auto" | "on" | "off"
• enable_web_scraping: fetch and read URLs found in user messages
• enable_web_citations: inline source citations
• enable_x_search: xAI native web and X/Twitter search, on supported models
• character_slug: respond as a published Venice character
• include_venice_system_prompt: defaults to true; set false for full control
• strip_thinking_response / disable_thinking: control <think> blocks
• enable_e2ee: end-to-end encryption on E2EE-capable models

Feature suffixes on the model ID do the same thing, for example kimi-k3:web or kimi-k3:enable_web_search=on.

POST /responses accepts a narrower set: character_slug, enable_e2ee, enable_web_search, enable_web_scraping, enable_web_citations, include_venice_system_prompt, and include_search_results_in_stream. Anything else is dropped without an error, so use /chat/completions when you need enable_x_search or the thinking controls.

Authentication

x402 settles in USDC on Base or Solana. A 402 response carries a PAYMENT-REQUIRED header holding base64 JSON with the top-up instructions and a SIWX challenge; read the rail out of accepts[] rather than assuming Base. Signed SIWX headers are valid for five minutes from issuedAt, so mint a fresh one rather than caching. Agents can also mint their own API key by staking VVV on Base, with no human in the loop.

Rules for agents

1. Discover, don't hardcode. Model IDs are deprecated and replaced regularly. See https://docs.venice.ai/overview/deprecations.
2. Quote before generating media. Video and music get expensive. Call /video/quote or /audio/quote first.
3. Video and music are asynchronous. Queue, then poll retrieve. Only /video/complete and /audio/complete block.
4. No content filtering. Venice models are uncensored by default. Do not add refusal or moderation layers unless asked.
5. Pick the right privacy tier. Models are tagged Anonymized, Private (zero retention), TEE (hardware enclave), or E2EE (encrypted client-side). TEE claims are verifiable at GET /tee/attestation and GET /tee/signature. See https://docs.venice.ai/overview/privacy.
6. Handle errors by code, not by string. https://docs.venice.ai/api-reference/error-codes
7. Respect rate limits. Watch the x-ratelimit-* response headers.

Gotchas that break working code

• POST /image/upscale takes exactly three fields: image, scale, and creativity. scale must be 2 or 4; 1 is rejected. creativity is clamped to 0 through 0.02. The old enhance, enhancePrompt, enhanceCreativity, and replication fields are gone.
• POST /image/edit defaults to the firered-image-edit model.
• POST /responses silently drops n, stop, seed, and prompt_cache_key. They are accepted and ignored rather than rejected. E2EE models are not supported there either.
• GET /billing/usage is deprecated. Use GET /billing/usage-history, which is keyset-paginated and takes startTimestamp / endTimestamp rather than the old parameter names.
• Crypto RPC is per-network: POST /crypto/rpc/{network}, with the live slug list at GET /crypto/rpc/networks. Methods are allowlisted per chain family, so an EVM method against Solana returns 400.

Going deeper

This file is a map. Venice maintains one self-contained skill per API surface, versioned against the OpenAPI spec, at https://github.com/veniceai/skills.

npx skills add https://docs.venice.ai
# or, for the full per-surface set:
git clone https://github.com/veniceai/skills.git ~/src/venice-skills
ln -s ~/src/venice-skills/skills ~/.claude/skills/venice

Reference

• Agent guide: https://docs.venice.ai/agents.md
• Getting started: https://docs.venice.ai/getting-started/quick-start.md
• Privacy tiers: https://docs.venice.ai/overview/privacy.md
• Pricing: https://docs.venice.ai/overview/pricing.md
• Rate limiting: https://docs.venice.ai/api-reference/rate-limiting.md
• Docs index for LLMs: https://docs.venice.ai/llms.txt
• Any docs page as raw markdown: append .md to its URL