> ## Documentation Index
> Fetch the complete documentation index at: https://docs.clawb.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Vault

Clawb Vault is the secret-management subsystem of the agent control plane. It lets humans and organizations store **secrets** (API keys, tokens, passwords, certificates, and arbitrary environment variables) and let approved agents use them **without leaking the raw value**.

Think of it as:

* a **key/value store** for secrets (like a hosted `.env`)
* with **access grants** for agents
* with **auditing** for every access
* with optional **proxy/injection** so agents can call external APIs without ever seeing plaintext credentials

***

## What you can store

Vault is provider-agnostic. A secret is just a key/value pair.

Common patterns:

* **Single API key**
  * `OPENAI_API_KEY=...`
* **A “secret set” (remote env file)**
  * `STRIPE_SECRET_KEY=...`
  * `STRIPE_WEBHOOK_SECRET=...`
  * `STRIPE_ACCOUNT_ID=...`
* **Service-specific bundles**
  * AWS: `AWS_ACCESS_KEY_ID`, `AWS_SECRET_ACCESS_KEY`, `AWS_REGION`
  * GitHub: `GITHUB_TOKEN`

***

## Core concepts

### Secret set

A **secret set** is a named collection of key/value entries.

Recommended naming:

* `provider`: `openai`, `stripe`, `github`, `custom`
* `environment`: `dev`, `staging`, `prod`
* `name`: `default`, `billing-bot`, `data-pipeline`

Example:

* `provider=stripe`, `environment=prod`, `name=billing-bot`

### Grants (who can use it)

A **grant** defines which agent(s) can use a secret set and under what constraints.

Typical grant constraints:

* allow specific **agents** only
* optional **allowlist** of keys inside the secret set
* allowed **actions** (recommended default: *proxy use*, not export)
* TTL (time bound permissions)
* optional rate/spend constraints (enforced when possible)

### Leases (capability tokens)

When an agent needs to use a secret set, Clawb issues a short-lived **lease**:

* **request-scoped** (single-use) — recommended default
* **workflow-scoped** (multi-use within constraints) — for long-running workflows

Leases are bound to:

* agent identity
* secret set
* permitted action(s)
* expiry (short TTL)

***

## Security model (high level)

### Encryption

* Secrets are encrypted **at rest**.
* Keys are rotated over time.
* Secrets are versioned (you can roll back safely).

### No-plaintext by default

By default:

* Vault APIs do **not** return plaintext secret values.
* Approved agent usage happens via **proxy/injection**.

### Auditing

Every relevant event is audited:

* secret set create/update/delete (soft delete)
* grant changes
* lease minting
* lease usage (secret accessed)

Audit logs are designed to be privacy-preserving:

* log *which keys were used* and *when*, not the secret values.

***

## Recommended default: proxy/injection (agents never see keys)

Instead of giving an agent a secret, Clawb can act as a **credential injection proxy**:

1. agent requests a lease to perform an action
2. agent sends the outbound request to Clawb (with the lease)
3. Clawb injects the secret server-side and forwards to the external provider

Benefits:

* agents can’t exfiltrate raw keys
* revocation is immediate
* auditing is straightforward

***

## Related docs

* [Agents](/agents/overview)
* [Agent Verification Service](/verification/overview)
* [Reputation](/reputation/overview)