AI keys and models
Bring your own AI provider keys, choose which models your agents think with, and see what each model costs.
Agents think with AI models from the providers you choose, using your own keys. Your provider bills you directly for what agents use; Everdeck's plan doesn't include model usage. Everdeck works with OpenAI, Anthropic, Google Gemini, xAI, Mistral, Groq, DeepSeek, OpenRouter, Azure OpenAI and OpenAI-compatible.
Add a key
Account owners and admins manage keys on the AI keys page. Adding, testing and deleting a key each ask you to confirm it's you first, so a key never waits in your browser while you find your phone.
- Select Add key (or Add your first key).
- Choose the Provider. The form links to the provider's console, where you create the key.
- Choose who it's Used by: Every company in the account, or one company. A company with its own key for a provider uses it instead of the account's, for example to bill its usage separately.
- Paste the API key and fill in anything else the provider needs: Azure OpenAI asks for its Endpoint and Deployment names; an OpenAI-compatible server (Together, Fireworks, your own vLLM) asks for its Base URL and Model IDs.
- Optionally give it a Name, like "Production", and select Check and save.
Everdeck checks the key with the provider before saving it. Once saved, it's never shown again: the page shows only its last four characters and a fingerprint, so you can tell keys apart.
Key status
Each key shows its status and when it was last checked:
- Working: the provider accepted it.
- Not checked: Everdeck couldn't reach the provider when it was saved. Select Test to check again.
- Refused: the provider turned it down. Agents can't use it until it's replaced.
To replace a key, add a new one for the same provider and the same company (or the whole account): the form says which key it replaces. Delete removes a key at once; a company that had its own key then falls back to the account's key for that provider, if there is one.
When no key reaches an agent's model, the agent can't be activated or resumed, and an active agent pauses at its next session. Its page says what's missing and links to AI keys.
Choose a model for an agent
On an agent's Model tab, company admins choose:
- Model: the model it thinks with. Only models your keys cover are listed. New agents start on
anthropic:claude-sonnet-5-5when your keys cover it, or the first model they do. - Fallback model: used straight away when the main model fails or its provider is down. A model from a different provider gives the best cover.
- Temperature: lower is more predictable. Leave it empty for the model's default.
- Longest reply (output tokens): leave it empty for the model's default.
The list shows the models Everdeck has verified for agents, which pass its tool-calling checks. Tick Show models not verified for agents to see the rest. The changes save with the charter, as a new version.
Verified models and price estimates
Everdeck estimates what agents spend from each model's published list price per million tokens, for the spending limits and the usage reports. The amount your provider bills is the real one.
| Provider | Model | Good for | Estimate per million tokens |
|---|---|---|---|
| OpenAI | GPT-6.1 Sol (openai:gpt-6.1-sol) |
The hard parts | US$2 in, US$10 out |
| OpenAI | GPT-6 Astra (openai:gpt-6-astra) |
The hard parts | US$10 in, US$50 out |
| Anthropic | Claude Opus 5.5 (anthropic:claude-opus-5-5) |
The hard parts | US$4 in, US$20 out |
| Anthropic | Claude Sonnet 5.5 (anthropic:claude-sonnet-5-5) |
The hard parts | US$2 in, US$10 out |
| Anthropic | Claude Haiku 4.5 (anthropic:claude-haiku-4-5) |
Quick, inexpensive work | US$1 in, US$5 out |
| Google Gemini | Gemini 3.8 Flash (google:gemini-3.8-flash) |
Quick, inexpensive work | US$0.75 in, US$3.75 out |
| xAI | Grok 4.7 (xai:grok-4.7) |
The hard parts | US$2 in, US$6 out |
| xAI | Grok 4.3 (xai:grok-4.3) |
Quick, inexpensive work | US$1.25 in, US$2.50 out |
| Mistral | Mistral Medium 3.5 (mistral:mistral-medium-3-5) |
The hard parts | US$1.50 in, US$7.50 out |
| Groq | GPT-OSS 120B on Groq (groq:openai/gpt-oss-120b) |
The hard parts | US$0.15 in, US$0.60 out |
| DeepSeek | DeepSeek V4 Pro (deepseek:deepseek-v4-pro) |
The hard parts | US$1.32 in, US$3.96 out |
| OpenRouter | Claude Sonnet 5.5 via OpenRouter (openrouter:anthropic/claude-sonnet-5.5) |
The hard parts | US$2 in, US$10 out |
| OpenRouter | GPT-6.1 Sol via OpenRouter (openrouter:openai/gpt-6.1-sol) |
The hard parts | US$2 in, US$10 out |
| OpenRouter | Gemini 3.8 Flash via OpenRouter (openrouter:google/gemini-3.8-flash) |
Quick, inexpensive work | US$0.75 in, US$3.75 out |
Azure OpenAI deployments and OpenAI-compatible models aren't in Everdeck's price list, so their usage is estimated at the price of the most expensive model listed. Spending limits still stop work instead of never noticing it; the estimates are on the high side.
See Spending caps and alerts for the limits that use these estimates.
More in Your account and team
Still stuck?
Email [email protected] and say what you were trying to do. Or browse every guide.