Providers and provider keys
Connect your own OpenAI, Anthropic, Mistral, MiniMax and other provider keys to Venom Cloud, choose the models each key serves, rotate keys and watch how each provider serves you.
Venom Cloud serves a request from your own daemons first, then from provider keys. On the Providers page you bring your own keys (BYOK): the cloud stores each one sealed in its key vault, lists the models the key can reach, and serves only the models you turn on. Nothing on these pages calls a provider until you paste a key or press a check button.
| Page | What it is for |
|---|---|
/providers |
Connected providers as cards (keys, models enabled, last check, last error), then every other provider the cloud supports, and the "use Venom's keys" switch. |
/providers/new |
The provider picker. /providers/new?provider=mistral opens one provider's connect flow directly. |
/providers/{id} |
One provider: its keys in serving order, the models each key serves, "Priced as", rotation, and the last 24 hours of traffic. |
/providers/status |
How each provider, model and key served your organisation, measured from your own requests. |
Connect a key
- Open
/providersand press Add provider, or pick a provider from the list. - Paste the key. A key that matches the provider's usual format is checked at once; the cloud (never your browser) calls the provider's models endpoint and answers "Key works" with the account and the models it found.
- Chat and embedding models start switched on. Turn off any you don't want served, set the provider's options (a region, for example), and press Save.
The key travels from your browser to the cloud once. No page, URL, log or answer ever shows it again: the console keeps only a hint such as sk-…a1b2. A key that fails the check shows the provider's status in words, never the key.
After saving, test it with any client key of the organisation:
curl https://cloud.usevenom.dev/v1/chat/completions \
-H "Authorization: Bearer vck_…" \
-H "Content-Type: application/json" \
-d '{"model": "mistral-small-latest", "messages": [{"role": "user", "content": "Reply with: ready"}]}'The request shows up on /providers/{id} under "Requests, 24 h", and in Requests filtered by served:cloud provider:mistral.
Supported providers
The ids are what the console, the admin API and Requests filters (provider:<id>) use. Every provider is reached over its OpenAI-compatible endpoint unless noted.
| Id | Provider | Endpoint | Key looks like | Options |
|---|---|---|---|---|
openai |
OpenAI | https://api.openai.com/v1 |
sk-… |
organization, project |
anthropic |
Anthropic | https://api.anthropic.com/v1 |
sk-ant-… |
|
gemini |
Google Gemini | https://generativelanguage.googleapis.com/v1beta/openai |
AIza… |
|
mistral |
Mistral | https://api.mistral.ai/v1 |
32 letters and digits | |
groq |
Groq | https://api.groq.com/openai/v1 |
gsk_… |
|
together |
Together AI | https://api.together.xyz/v1 |
64 hex characters or tgp_v1_… |
|
fireworks |
Fireworks AI | https://api.fireworks.ai/inference/v1 |
fw_… |
|
deepseek |
DeepSeek | https://api.deepseek.com |
sk-… |
|
xai |
xAI | https://api.x.ai/v1 |
xai-… |
|
openrouter |
OpenRouter | https://openrouter.ai/api/v1 |
sk-or-v1-… |
|
qwen |
Alibaba Cloud Model Studio (Qwen) | https://dashscope-intl.aliyuncs.com/compatible-mode/v1 |
sk-… |
region, workspace |
zai |
Z.ai | https://api.z.ai/api/paas/v4 |
id.secret |
endpoint: global, GLM Coding Plan or China |
typesafe |
TypeSafe | https://api.typesafe.ai/v1 |
your TypeSafe key | |
cerebras |
Cerebras | https://api.cerebras.ai/v1 |
csk-… |
|
deepinfra |
DeepInfra | https://api.deepinfra.com/v1/openai |
API token | |
nebius |
Nebius Token Factory | https://api.tokenfactory.nebius.com/v1 |
API key | |
moonshot |
Moonshot AI (Kimi) | https://api.moonshot.ai/v1 |
sk-… |
endpoint: global or China |
minimax |
MiniMax | https://api.minimax.io/v1 (global) or https://api.minimax.cn/v1 (China) |
sk-api-… or sk-cp-… |
region |
elevenlabs |
ElevenLabs (voice only) | https://api.elevenlabs.io/v1 |
sk_… |
|
fishaudio |
Fish Audio (voice only) | https://api.fish.audio |
API key | |
byteplus-speech |
BytePlus Seed Speech (voice only) | https://voice.ap-southeast-1.bytepluses.com |
console API key, or App ID and Access Token | |
siliconflow |
SiliconFlow | https://api.siliconflow.com/v1 |
sk-… |
region |
novita |
Novita AI | https://api.novita.ai/openai/v1 |
sk_… |
|
chutes |
Chutes | https://llm.chutes.ai/v1 |
cpk_… |
|
baseten |
Baseten | https://inference.baseten.co/v1 |
API key | |
azure |
Azure OpenAI (Foundry) | https://{resource}.openai.azure.com/openai/v1 |
32 or 84 characters | resource, endpoint (Azure OpenAI or Foundry), deployment names typed in |
vercel |
Vercel AI Gateway | https://ai-gateway.vercel.sh/v1 |
Vercel AI Gateway key | |
cloudflare |
Cloudflare AI | https://api.cloudflare.com/client/v4/accounts/{account_id}/ai/v1 |
API token with Workers AI read | account ID, gateway ID (default default) |
huggingface |
Hugging Face | https://router.huggingface.co/v1 |
hf_… |
|
portkey |
Portkey | https://api.portkey.ai/v1 |
API key | |
requesty |
Requesty | https://router.requesty.ai/v1 |
API key | region |
martian |
Martian | https://api.withmartian.com/v1 |
API key | |
aimlapi |
AI/ML API | https://api.aimlapi.com/v1 |
API key | |
edenai |
Eden AI | https://api.edenai.run/v3/llm |
API key | |
bedrock |
Amazon Bedrock (API key) | https://bedrock-mantle.us-east-1.api.aws/v1 |
ABSK… or bedrock-api-key-… |
endpoint (mantle or runtime), region |
bedrock-iam |
Amazon Bedrock (IAM) | Bedrock Runtime Converse API, SigV4-signed | access key ID and secret | region, session token |
sagemaker |
Amazon SageMaker | SageMaker real-time endpoints, SigV4-signed | access key ID and secret | region, payload format, price per token in and out |
litellm |
LiteLLM | your proxy's URL | sk-… |
base URL |
custom |
Other (OpenAI-compatible) | any URL you type | the server's key | name, base URL, headers, auth style |
The key format is a hint, not a gate: a key that doesn't match is flagged before any call, and you can still try it.
Mistral
Make a key under API Keys in the Mistral console (La Plateforme). A Codestral-only key, issued for codestral.mistral.ai, does not work here. The listing includes Mistral Large 4, Small 4, Ministral 3, Codestral and the embedding models (Mistral Embed, Codestral Embed), checked against Mistral's model list on 2026-11-03. Retired models such as Magistral or Devstral may still appear in a key's listing for a while.
MiniMax
Use a pay-as-you-go key (sk-api-…, from API Keys, Create new secret key). An M Plan subscription key (sk-cp-…) works once the account has a seat or credits. Keys from the China site (minimax.cn) only work with Region: China; the two sites' keys are not interchangeable. MiniMax-M3 and M3.1 Flash have a 1M-token context and return reasoning. MiniMax has no OpenAI-compatible embeddings endpoint, so the cloud never sends embeddings to it.
Voice providers
ElevenLabs, Fish Audio and BytePlus Seed Speech serve speech only. OpenAI, Model Studio (Qwen TTS), MiniMax (T2A speech), Mistral (Voxtral transcription) and your own OpenAI-compatible speech server serve speech next to their chat models. A voice model serves /v1/audio/* only, never chat. A fresh check turns on voice models whose role is known (text to speech or speech to text), with the speech models a provider's listing leaves out added to it: MiniMax's speech-2.8-hd and speech-2.8-turbo, Model Studio's qwen3-tts-flash, Mistral's voxtral-mini-latest, ElevenLabs' scribe_v2. Checked against each provider's docs on 2026-11-03.
- ElevenLabs: create a key under Developers, API Keys, with Text to Speech, Speech to Text, Voices (read) and Models (read) access. The key is sent as
xi-api-key. The listing is ElevenLabs' speech models (Eleven v3, Multilingual v2, Flash v2.5) plus Scribe v2. - Fish Audio: create a key on the API Keys page of your Fish Audio account; it is shown once. Fish Audio has no model list, so the check reads your account's own voices page and offers
s2.1-pro,s1andtranscribe-1. - BytePlus Seed Speech: use a Seed Speech console API key (sent as
X-Api-Key), not a ModelArk key. An older app's App ID and Access Token also work: fill in both fields and they are sent asX-Api-App-IdandX-Api-Access-Key. BytePlus has no endpoint that lists models or checks a key, so the key is first tried when it serves. Models:seed-tts-2.0,seed-tts-1.0,seed-icl-2.0(cloned voices),seed-asr-fast.
AWS
Amazon Bedrock with an IAM access key and Amazon SageMaker ask for the access key ID, the secret and an optional session token in separate fields, stored together as one sealed key. The IAM key needs bedrock:InvokeModel, bedrock:InvokeModelWithResponseStream, bedrock:ListFoundationModels and bedrock:ListInferenceProfiles (Bedrock), or sagemaker:InvokeEndpoint, sagemaker:InvokeEndpointWithResponseStream and sagemaker:ListEndpoints (SageMaker). SageMaker endpoints bill by the instance hour, so their requests cost $0 until you set prices for them (see the Pricing page, Pricing and invoice reconciliation).
Your own server
Other (OpenAI-compatible) connects any server that speaks the OpenAI API: give it a name, the base URL, the key, optional headers and the auth header style, then Discover models. When the server has no models listing, type the model ids in by hand. LiteLLM works the same way and also reads each model's context window and price from the proxy's /model/info when it answers.
One provider's page
/providers/{id} shows four figures (keys, models enabled, requests and errors in the last 24 hours, each linking into Requests) and then:
- Keys, in serving order. The first key serves first. A
401,429or5xxfrom one key falls through to the next. Move keys up and down to change the order. - Models. One key at a time, its last listing with a toggle per model, search, bulk actions, and model ids typed in by hand. Saving writes the key's enabled list; serving follows within 10 seconds.
- Priced as. For each model a key serves, the model it is priced as (typed in, or what the provider's listing says) and the price that gives under your pricing source. Use it for a model the cloud doesn't know by name, such as an Azure deployment.
- New models and prices. Check asks the provider what it lists now with your keys, and shows new models, retired ones, prices that differ from what the cloud books and unpriced models. Nothing changes until you accept a row.
- Provider options for the key, such as the region.
- Last 24 hours of cloud traffic for this provider.
Each key's menu:
| Action | What it does |
|---|---|
| Refresh models | Lists the key's models again; new ones are shown, not turned on. |
| Rotate key | Paste a new key. It serves first at once; the old key keeps serving behind it until the grace period ends, so nothing fails while you switch. Name, order, models and history stay. |
| Rotation schedule | When to remind you to rotate this key (the bell and webhooks that take key_rotation events), counted from when the current key was added. |
| Rename | Up to 64 characters. |
| Pause / Resume | A paused key stops serving within 10 seconds; resuming brings it back within 10 seconds. |
| Delete | Removes the key from the vault. |
| Rotation option | Choices | Default |
|---|---|---|
| Keep the old key serving | Retire it at once, 1 hour, 24 hours, 3 days, 7 days | the organisation's default (Settings, Key rotation) |
| Check the new key first | on or off | on |
| Remind | organisation default, never, 30, 60, 90, 180 or 365 days | organisation default |
Use Venom's keys
Above the provider list, the "use Venom's keys" switch decides whether Venom's own provider keys serve your requests after your keys and daemons. When it is on, a provider you haven't connected shows "Served with Venom's key until you add yours"; that traffic is metered on your plan. When it is off, only your keys and daemons serve. On a deployment without keys of its own, the switch is absent and the page says so. Only an owner or admin can change it.
Provider status
/providers/status reads your organisation's own request log; it never asks a provider's status page.
| Control or figure | Meaning |
|---|---|
| Window | 24 hours, 7 days or 30 days. |
| API key, Daemon | Narrow the figures to one client key or one daemon. |
| Now | The worst provider over the last 15 minutes. |
| Uptime | Share of requests that did not fail on the provider's side, over the window. |
| Requests | Requests in the window, and how many failed on the provider's side. |
| First token p50 | Median time to first token of streamed requests. |
| Incidents | Hours with errors or slow first tokens, shown as bands on each provider's timeline. |
Each provider row expands to its models and targets (your keys, Venom's key, daemons) with errors by class, first-token p50 and p95, and tokens per second.
Who can do what
| Action | Permission | Roles |
|---|---|---|
See /providers, /providers/{id} and the pricing |
providers:read |
owner, admin, developer, viewer |
See /providers/status |
usage:read |
owner, admin, developer, billing, viewer |
| Add, rotate, pause, reorder, delete keys; choose models; accept catalog rows | providers:write |
owner, admin |
| The "use Venom's keys" switch | providers:write |
owner, admin |
A platform operator acts on the organisation picked in the org switcher; with none picked, the pages are read-only.
Troubleshooting
| Message | Cause and fix |
|---|---|
| Paste the API key. | The key field was empty. |
| Your role cannot change provider keys. | Your role lacks providers:write. Ask an owner or admin. |
| Your session ended. Sign in again. | The console session expired. |
| Key works, but a model you expect is missing | The provider's listing doesn't include it. Type the id in by hand under Models, then test it. |
| A model books $0 | The cloud has no price for it. Set "Priced as" on the provider page, or a custom price on the Pricing page. |
| Requests still use a paused or deleted key | Serving follows key changes within 10 seconds. |
401 from the provider after rotation |
The new key was saved without the check. Rotate again with Check the new key first on, or retire the bad key from its row. |
Related
- Gears: pick which model each role of a gear uses.
- Routing: where requests go between daemons, trusted nodes and provider keys.
- Pricing and invoice reconciliation: what each model books, and checking a provider's bill.
- Venom Cloud: the section overview.