Anvil is an alpha testing release. It may have bugs. If something here doesn't match what you see, tell us.
Anvil docs · Your AI keys
Is it safe to give an AI tool your API keys?
It can be, if the tool keeps your keys on your own computer, encrypted, and never sends them to its maker's servers. The risk is a service that holds people's keys on its own servers: if it is breached, attackers can run up usage on every linked account. Vitkilabs Anvil keeps your provider keys and assistant sign-ins in an encrypted vault on your computer and uses them from there. They are never sent to Vitkilabs. You may not need a key at all: Anvil can use your existing Claude or ChatGPT sign-in, or a model on your own machine.
Where is the key stored? On your computer, or on the company's servers? A key on someone else's server is only as safe as that server.
Is it encrypted, and who can unlock it? Look for encryption at rest with a passphrase or system keychain that you control.
Can it end up in a prompt? Anything in a chat is sent to the model provider with the rest of the prompt. Keys belong in a tool's settings, not in the conversation.
What does the tool send home? A trustworthy tool can tell you, in a short plain list.
Do you need a key at all? A sign-in you already have, or a local model, may do the job. Where you do use keys, give each tool its own, so you can revoke one without touching the rest.
Your AI keys stay home
In a vault on your machine. Never sent to us.
Anvil keeps your AI provider keys and assistant sign-ins in an encrypted vault on your own computer, and uses them from there. They are never sent to Vitkilabs, so even a breach of our servers couldn't hand out your AI accounts.
The only things Anvil sends to Vitkilabs are your licence check-in, your account sign-in, bug reports you choose to send (scrubbed of anything that looks like a secret), and update checks. Bug reports never include your memories, prompts or code.
Encrypted with your passphrase. You create the vault on first run and save a recovery PDF. Anvil won't save a key until the vault exists.
Unlocked by your system. After you unlock it, Anvil keeps an unlock key in your system's credential store: Keychain on macOS, Credential Manager on Windows, Secret Service on Linux.
Sign-ins stay with their apps. You sign in to Claude or ChatGPT in their own apps. Anvil never handles that login.
Keys and prompts
Kept out of what goes to the model.
Keys in the vault stay out of the model's context. Anvil also scrubs API keys and tokens from text before anything leaves your machine, and shows you a preview of what it redacted. Both are best-effort. They are a safety net, not a guarantee, so still keep keys out of your chats.
Add a key, or skip it
Three ways to give Anvil a model.
Use a sign-in you already have. Connect Claude Code or Codex, and Anvil uses that Claude or ChatGPT sign-in for its own work. No key needed.
Add an API key. Groq, Google AI, Cerebras, OpenRouter, OpenAI or Anthropic. In Anvil's Settings, open Vault & secrets, click SET… beside the service, paste your key and click Save.
Run a local model. Anvil talks to Ollama, llama.cpp, MLX (on Apple silicon) and vLLM. Install one yourself, start it and download a model. Anvil looks for Ollama at 127.0.0.1:11434. With a local model, Anvil can run with no internet at all.
Avoid it. Anything in a chat is sent to the model provider with the rest of the prompt. Put keys in a tool's encrypted settings, never in the conversation.
Does Anvil send my API keys to Vitkilabs?
No. Your keys and assistant sign-ins stay in the vault on your computer and are used from there. Even a breach of Vitkilabs' servers couldn't hand them out.
Where does Anvil keep my API keys?
In Anvil's encrypted vault on your own computer, protected by a passphrase you choose. Anvil won't save a key until the vault exists.
Do I need an API key to use Anvil?
No. Anvil can use your existing Claude or ChatGPT sign-in, or a model running on your own machine through Ollama, llama.cpp, MLX or vLLM.