Skip to main content
Your agent needs credentials — an LLM key, sometimes a bot token or two, maybe a database password. The Secrets Vault is where those live. They’re encrypted at rest, never returned by the API after you save them, and injected into the agent as environment variables when its container starts.
You can’t read a secret’s value back once it’s saved. Make sure you have it stored somewhere safe before you paste it in here.
Image

​
Connect a provider

You don’t strictly need one — the Create Agent wizard offers a Free tier option that runs your agent on a shared, rate-limited model with no key at all. But connecting your own provider unlocks full model quality, context, and rate limits, and it’s how a free-tier agent upgrades later. Open the Secrets Vault. The row at the top — AI PROVIDERS — has a card for each supported provider:
ProviderHow it connects
AnthropicAPI key
OpenAIAPI key, or OAuth into a Codex subscription
OpenRouterAPI key
PinataPinata-hosted inference
VeniceAPI key (privacy-focused)
CustomAny OpenAI-compatible endpoint
Free tierNothing — selected in the Create Agent wizard, no key needed
Cards that aren’t connected show a CONNECT button. Click it, follow the prompts, save. The card flips to Connected. You can connect more than one provider — useful for fallbacks, or for routing different agents to different LLMs.

​
Add other secrets

For everything that isn’t a provider key — bot tokens, third-party API keys, database URLs — use New Secret at the top right of the Vault. The dialog asks for:
  1. A name (this is the environment variable, like TELEGRAM_BOT_TOKEN)
  2. A value
  3. A type — secret or variable
Save. Your agent will see it as process.env.TELEGRAM_BOT_TOKEN (or the equivalent in Python, Bash, etc.) on next restart.

​
Secret vs variable

Both become env vars. The difference is encryption:
  • Secret — encrypted at rest, never returned in API responses. Use for anything sensitive: keys, tokens, passwords.
  • Variable — stored as plaintext, value returned in listings. Use for non-sensitive config — public URLs, feature flags.
If you’re unsure, choose secret. There’s no downside.

​
Importing a .env

Have a .env file already? The New Secret menu has an Import .env option that adds every line in one go.

​
Make a secret available to an agent

Saving a secret in the Vault doesn’t automatically give it to every agent. You attach secrets per-agent — that way you can give different agents different credentials.
  • When creating an agent: step 3 of the wizard (“Connect”) lets you pick which secrets to attach.
  • For an existing agent: open the agent → Secrets → + ADD and pick from your Vault.
On the agent’s Secrets tab you’ll see two sections:
  1. AI Providers — the providers you’ve connected. Each card shows whether it’s an API KEY connection or a CODEX SUBSCRIPTION.
  2. Variables and Secrets — the non-provider secrets you’ve attached.
At the bottom of the same page is the Gateway Token — the credential other tools use to talk to this specific agent. See HTTP API → Gateway token for what to do with it.
When you add or update a secret on a running agent, restart the gateway so the agent picks up the new value. The Danger tab shows a per-secret Synced indicator so you can tell whether you’ve already done that.

​
Updating, removing, and edge cases

  • Click the ⋯ menu on any secret to update its value or delete it.
  • You can’t delete a secret that’s still attached to an agent — detach it first, or you’ll get a 409 Conflict.
  • Updating a value doesn’t roll out automatically — restart the gateway.
  • Attaching two credentials for the same provider (say, an OpenAI API key and a Codex subscription) returns a 409 Conflict. Pick one.

​
From the CLI

pinata agents secrets list                  # See your secrets
pinata agents secrets add <name> <value>    # Add one
pinata agents secrets delete <name>         # Remove one

​
How secrets are protected

  • Encrypted with AES-GCM using a key derived per user
  • Values are never returned by the API after creation
  • Injected as environment variables at container start, never written to disk
  • The per-agent gateway token is generated automatically and can be rotated from the Danger tab