Skip to main content

Venice AI Provider Guide

Venice AI is a Tier-2 catalog provider: its integration is one JSON file (src/lib/providers/catalog/venice-ai.json) rather than hand-written code. That file is the source of truth for everything on this page.

Verification status: this entry is docs- and roster-verified, not yet live-verified. The catalog fields below were taken from Venice's own public pages and unauthenticated GET requests to /models, /models/traits and /chat/completions — no account was created and no API key was used to build it. evidence.liveMatrix is null until someone runs the live capability matrix with a real key (see Verification status below).

Key Facts​

  • Provider id: venice-ai (alias venice)
  • Protocol: OpenAI-compatible (/chat/completions) — the chat-completions reference says "This endpoint is OpenAI-compatible"
  • Base URL: https://api.venice.ai/api/v1
  • Default model: zai-org-glm-5-2
  • Models in catalog: 10 (curated from a 126-model live roster)
  • Streaming: supported — the chat-completions reference documents a stream parameter: "Whether to stream back partial progress. Defaults to false."
  • Tool calling: model-dependent — the function-calling guide says "Function calling support is model-specific." and points to the supportsFunctionCalling field of the models API
  • Tools while streaming: not declared (false) — no streamed tool-call request was made
  • Structured output: supported — the structured-responses guide documents response_format with json_schema; it says "This functionality is not natively available for all models." and points to the supportsResponseSchema field of the models API
  • Structured output + tools together: not declared (false) — the structured-responses guide says "Structured Outputs via response_format are not compatible with parallel function calls", and no combined request was made
  • Embeddings: not declared on this catalog entry. Venice's API index (https://docs.venice.ai/llms.txt) lists an Embeddings endpoint, but NeuroLink's generic ConfiguredOpenAICompatProvider (the class Tier-2 catalog entries use) does not implement embed()/embedMany(), so this flag tracks that, not the vendor's own API surface
  • Thinking: declared (true) — the reasoning-models guide documents a reasoning_content output field and a reasoning_effort parameter, and the roster carries a supportsReasoning flag per model
  • Billing: no-free-tier — see Billing below
  • Key format: none declared by Venice

Quick Start​

1. Get an API key​

  1. Visit: https://venice.ai/settings/api and select Generate New API Key; the API key guide lists "Sign in to your Venice account." under Before you start (https://docs.venice.ai/guides/getting-started/generating-api-key)
  2. Copy the key when it is shown — the API key guide states "Venice will show the full API key one time." (https://docs.venice.ai/guides/getting-started/generating-api-key) — and paste it without a Bearer prefix or quotes (https://docs.venice.ai/getting-started/quick-start)
  3. Fund the account before calling models: the docs state "model requests will not succeed until the account can consume DIEM, bundled credits, or USD" (https://docs.venice.ai/guides/getting-started/generating-api-key); payment options are listed at https://docs.venice.ai/overview/pricing
  4. Set VENICE_AI_API_KEY in your .env file (VENICE_API_KEY, the variable Venice's own docs use, is also read)

2. Configure​

export VENICE_AI_API_KEY=your-api-key
export VENICE_AI_MODEL=zai-org-glm-5-2   # optional — overrides the default model
export VENICE_AI_BASE_URL=https://api.venice.ai/api/v1   # optional — proxy or self-hosted gateway

3. Use it​

import { NeuroLink } from "@juspay/neurolink";

const neurolink = new NeuroLink();

const result = await neurolink.generate({
  input: { text: "Explain context windows in one paragraph." },
  provider: "venice-ai",
  model: "zai-org-glm-5-2",
});

console.log(result.content);
# CLI
npx @juspay/neurolink generate "Hello" --provider venice-ai

Per-request credentials work the same way as on the other catalog providers:

await neurolink.generate({
  input: { text: "Hello" },
  provider: "venice-ai",
  credentials: { veniceAi: { apiKey: process.env.VENICE_AI_API_KEY } },
});

Billing​

What the pages opened on 2026-09-29 say:

  • The API key guide says "You can create a key before funding the account, but model requests will not succeed until the account can consume DIEM, bundled credits, or USD."
  • The pricing page lists three payment options. The USD card says "Buy Venice credits with credit card. Credits never expire." The Stake DIEM card says "Each Diem = $1/day of credits that refresh daily."
  • Venice's plan page lists "API Access (pay with credits)" under the Free plan.
  • Venice's API page shows the tag "Free tier, no card" in its hero section, and a Pay as you go card that lists "Add funds only when needed".

The entry records no-free-tier, following the documented balance requirement in the first bullet.

Models​

ModelContextMax outputVision$/M in · out (cache read)Notes
zai-org-glm-5-2 ⭐1,000,000131,072no$1.4 / $4.4 ($0.26)NeuroLink default. Roster traits "default" and "function_calling_default"; the /models/traits GET returned "default":"zai-org-glm-5-2".
zai-org-glm-5198,00032,000no$1 / $3.2 ($0.2)Fallback. The quick-start page lists it with the text "Default model for most use cases".
zai-org-glm-5-1200,00080,000no$1.54 / $4.84 ($0.286)Fallback. The model in the reasoning-models examples.
deepseek-v4-pro-08131,000,00032,768no$1.65 / $4.95 ($0.165)Fallback. Roster trait "default_code".
kimi-k31,000,000131,072yes$3.75 / $18.75 ($0.375)Roster trait "default_reasoning".
kimi-k2-6256,00065,536yes$0.75 / $3.5 ($0.16)The quick-start page lists it with the text "Strong reasoning for more complex tasks".
qwen3-vl-235b-a22b128,00016,384yes$0.21 / $1.9 ($0.1)The model in the vision guide's examples; models.visionModel.
venice-uncensored-1-2128,0008,192yes$0.2 / $0.9Roster trait "most_uncensored"; the quick-start page lists it with the text "Venice's uncensored model".
grok-4-7500,000200,000yes$2.27 / $6.8 ($0.57)Roster trait "most_intelligent". The pricing page has a further row labelled "↳ >200K Context" at $4.53 input and $13.60 output, which the flat prices here do not carry.
claude-opus-4-81,000,000128,000yes$6 / $30 ($0.6)The quick-start page lists it with the text "High-intelligence model for complex tasks". Roster privacy "anonymized".

Context, max output, vision and prices come from an unauthenticated GET https://api.venice.ai/api/v1/models call made 2026-09-29 (126 text models total; 10 curated into this catalog; the roster's e2ee- ids are not in it). The input and output prices and the rounded context also appear on the pricing page and the text-models page; the pricing page shows cache-read prices rounded to two decimals. "Roster trait" means the id appears in the model's traits list on the roster; the deprecations page lists the current trait mapping. The text-models page lists each model's Capabilities, for example "Function calling, Reasoning, Vision, Code" for kimi-k3.

models.defaultContextWindow (128,000) and models.defaultMaxOutputTokens (4,096) are placeholders the vendor does not publish: Venice publishes context and max-output figures per model on the roster, not as a general default.

Fallback order when the default is unavailable: zai-org-glm-5 → zai-org-glm-5-1 → deepseek-v4-pro-0813. The runtime fallback model name the loader derives (fallbacks[1]) is zai-org-glm-5-1.

Verification status​

Tier-2 onboarding requires evidence before a provider is accepted, and pnpm run verify:provider-onboarding gates it in CI. This is what the catalog currently records for Venice AI — docs- and roster-verified, not live-verified:

ProbeResult
Rosterunauthenticated GET /models, HTTP 200, 126 text models, 2026-09-29 — no key used
Default modelunauthenticated GET /models/traits, HTTP 200, 2026-09-29 — returned "default":"zai-org-glm-5-2"
Billingpublic pages quoted in Billing above, 2026-09-29 — recorded as no-free-tier
Auth-failure shapean unauthenticated GET /chat/completions returned HTTP 404 with body {"error":"Not found"} on 2026-09-29; no request that fails authentication was made. errorRules use the statuses and messages listed on the error-codes page: 401, 402, 429, and the model-not-found messages ("Specified model not found", "Invalid model specified")
Tools / structured outputdocumented on the function-calling and structured-responses guides linked above; neither was exercised live, and no combined tools+schema request was sent — structuredOutputWithTools stays false
Live capability sweepnot run. evidence.liveMatrix is null. Before treating this provider as production-ready, run npx tsx test/continuous-test-suite-provider-matrix.ts --provider=venice-ai with a real key and record the result.

Do not treat this entry as equivalent to a live-verified Tier-2 provider (e.g. FriendliAI, Novita AI) until that live matrix has been run and evidence.liveMatrix is filled in.

Troubleshooting​

Each row is what a Venice page states; none was reproduced live.

SymptomCauseFix
HTTP 401The API key guide says a 401 with AUTHENTICATION_FAILED means the key itself was rejected, and lists a partial copy, a Bearer prefix or quotes captured into the value, or a stray space or newline as the usual causesThe quick-start page says to paste the key exactly as Venice issued it, with no Bearer prefix, on a single line with no quotes
HTTP 402The error-codes page lists INSUFFICIENT_BALANCE and two per-key spend-limit codes under 402; the API key guide says that with a spend-limit code "the cause is the key's own limit rather than your account balance"For INSUFFICIENT_BALANCE the error-codes page points to https://venice.ai/settings/api to add credits; for a spend-limit code the API key guide says to raise or remove the key's Epoch Consumption Limits at the same page
HTTP 429RATE_LIMIT_EXCEEDED or MODEL_OVERLOADED on the error-codes page; the rate-limiting page also lists two 30-second error budgets, 50 failed requests and 200 unsupported-feature requests, each returning 429The rate-limiting page says "Failed requests (500, 503, 429) should be retried with exponential backoff." and to check the x-ratelimit-reset-requests header for the time you can retry
HTTP 504 on a long non-streaming requestREQUEST_TIMEOUT on the error-codes pageThe error-codes page says to use the streaming API by setting stream=true for long-running non-streaming inference requests
Model not foundMODEL_NOT_FOUND (404, "Specified model not found") or INVALID_MODEL (400, "Invalid model specified") on the error-codes pageModel ids are served by the unauthenticated GET https://api.venice.ai/api/v1/models (List Models)
Structured output ignored with tools attachedstructuredOutputWithTools is false on this entry; the structured-responses guide says structured outputs are not compatible with parallel function callsThe generic provider suppresses response_format when tools are attached (the default for entries without structuredOutputWithTools: true)

See also​