AgentValet vs HashiCorp Vault
Vault is the secrets platform many security teams already run, and it now has features aimed at AI agents. AgentValet is much narrower. The useful question is whether your agents should receive secrets at all, or only the results of calls made on their behalf.
Updated
The short answer
Pick AgentValet if you want AI agents to act on SaaS platforms and MCP servers without ever receiving a credential, with per-action grants and a person approving risky calls, as a managed service.
Pick HashiCorp Vault if you need a secrets platform for your whole estate that you can run yourself, with dynamic secrets, leases you can revoke, and an agent registry in Vault Enterprise.
Two approaches, in plain words
What HashiCorp Vault is
Vault is HashiCorp’s secrets and identity platform. Clients authenticate through an auth method, Vault maps them to an identity and its policies, and they read secrets, including dynamic secrets that Vault creates on request with a lease that can be renewed or revoked. Audit devices can record every request and response. You run it yourself as Community or Enterprise, or use HCP Vault Dedicated.
For agents, Vault Enterprise 2.1 adds agentic IAM: an agent registry, policies where an agent’s access is the intersection of its owner’s policies, its own baseline and a ceiling, and per-request authorisation where the token exists only for that request. HashiCorp also publishes a Vault MCP server, labelled beta.
What AgentValet is
AgentValet is a credential broker for AI agents. It doesn’t hand secrets out: the broker holds the credential for each platform, checks the agent’s grant for that platform and action, and attaches the credential in memory for the one call.
A risky action can wait for a person, and each decision is logged with the policy that made it. AgentValet isn’t a general secrets manager, and you can’t self-host it.
The question that separates them
Does the agent end up holding the secret?
Vault’s model is to give a client the secret it’s entitled to, ideally short-lived, and to revoke it when the lease ends. For an agent, that puts the secret in the agent’s process for the lease, or in an MCP server the agent calls, as in HashiCorp’s validated pattern for agent identity. Short leases and narrow policies limit what a leaked secret can do.
AgentValet never releases the credential. The agent sends the call, and the broker makes it with the credential attached in memory, so no secret outlives the task in the agent’s hands.
- Agent or MCP server authenticates
- Vault checks policy and issues a lease
- Secret returned to the client (the key sits with the agent)
- Client calls the platform
- Real service
The secret is short-lived and revocable, and the client holds it until the lease ends or is revoked.
- Your agent
- Broker checks identity, grant and approval
- Credential attached in memory
- SaaS API or MCP server
The agent holds no platform key, and a risky call can wait for a person before it runs.
The standard comparison
One line per cell. Open the detail underneath for the longer version.
| AgentValet | HashiCorp Vault | |
|---|---|---|
| Per-agent identity | Each agent gets its own RS256 signing keypair and signs every request. | Vault Enterprise 2.1 adds an agent registry; otherwise agents authenticate like any other client. |
| Credential exposure to the agent | The agent never holds or sees the credential. | The client that reads a secret receives it, along with a lease ID to renew or revoke it. |
| Token lifetime | Each request carries a short-lived token signed by the agent; upstream credentials stay in the broker. | Dynamic secrets and tokens carry leases with a TTL, and revoking a lease invalidates the secret immediately. |
| Scoping granularity | Per agent, per platform and per action, deny by default. | Path-based policies, and for agents the intersection of owner, baseline and ceiling policies plus per-request authorisation details. |
| Human approval | Risky actions pause at the broker until a person approves them. | Control groups can hold a request until named approvers authorise it, on Vault Enterprise or HCP Vault Dedicated. |
| Audit evidence | Every decision is logged with the policy, version, result and reason, and exports as CSV or JSON. | Audit devices record nearly every API request and response, with most string values stored as HMAC hashes. |
| MCP support | A native MCP server, and it brokers calls to upstream MCP servers as an OAuth client. | A Vault MCP server, in beta, exposes Vault operations as tools over stdio or streamable HTTP. |
| Self-host option | No. It is a managed service hosted in Australia. | Yes: Community and Enterprise run on your infrastructure, and HCP Vault Dedicated is the managed option. |
| Open source | The parts that run on your machines (MCP server, CLI, SDKs and adapters) are MIT licensed. The broker is not. | Source-available under the Business Source License 1.1 since version 1.15; the Vault MCP server is MPL 2.0. |
| Setup time | One command, npx @agentvalet/register, registers an agent. | A development server starts with one command; production means running and configuring a cluster, or using HCP. |
| Starting price | Free plan, with paid plans from US$99 a month. | Community is free to run under its licence; HCP Vault Dedicated is billed per cluster hour plus per active client; Enterprise is quoted. |
The detail behind 8 of these rows
- Per-agent identity
AgentValet: Every agent publishes an identity document and key set, so each call is attributable to one agent. Child agents can be issued their own identities with scopes cut down to a subset of the parent’s.
HashiCorp Vault: A request resolves to an identity entity, Vault checks for an agent registry record, and on-behalf-of workflows are capped by the agent’s authorisation ceiling.
- Credential exposure to the agent
AgentValet: Each credential is envelope-encrypted and decrypted in memory only for the call it serves. It is never logged or returned to the agent.
HashiCorp Vault: HashiCorp’s validated pattern for AI agents puts an MCP server between the agent and Vault, so the MCP server holds the dynamic credentials and the agent can’t reach the resource directly.
- Token lifetime
HashiCorp Vault: Revoking a token revokes every lease created with it. With agentic IAM, the per-request token exists only for the request and isn’t written to storage.
- Scoping granularity
AgentValet: The method and endpoint of each call are matched against the scope granted to that agent before any credential is decrypted.
- Human approval
AgentValet: Approve with a passkey, an email link or a mobile push. The approval is recorded with the action it released.
HashiCorp Vault: The requester gets a time-limited wrapping token, approvers authorise it, and the original request is processed once every approval is in. The agentic IAM docs don’t describe an approval step of their own.
- Audit evidence
AgentValet: The log is append-only and policy-stamped, and each decision carries a signed receipt.
HashiCorp Vault: Audit devices are off by default on a new cluster. If no enabled audit device can be written to, Vault refuses the request, and HashiCorp recommends enabling at least two.
- MCP support
HashiCorp Vault: HashiCorp strongly discourages beta features in production and advises a dedicated, narrowly scoped token rather than a root or shared one.
- Starting price
HashiCorp Vault: HCP clients are counted once per billing month, from their first authentication. We couldn’t read current HCP rates from HashiCorp’s pricing page, so check them there.
Where HashiCorp Vault genuinely wins
Vault is the more complete secrets platform by a distance. Dynamic secrets, leases you can revoke along with every secret a token created, control groups for human sign-off, and audit devices that refuse a request rather than skip the log are all mature, and you can run the whole thing yourself.
If your agents are one consumer among hundreds of services, keeping them on the platform that already holds your secrets has real advantages, and the agent registry and per-request authorisation in Vault Enterprise 2.1 bring agent-specific controls to it.
Where AgentValet wins
AgentValet is narrower and built for one job: agents acting on outside platforms without holding the key. The broker makes the call with the credential attached in memory, so the agent never holds a secret for any length of time, and grants are per agent, per platform and per action.
A person can approve a risky action with a passkey, an email link or a push notification, and each decision carries a signed receipt. One command registers an agent, with no cluster to run.
They stack well. Vault can keep serving secrets to your services and pipelines while AgentValet holds the platform credentials your AI agents use, so the agents don’t need Vault tokens at all.
The first step, side by side
Each command is the documented starting point for that product. HashiCorp Vault’s comes from HashiCorp’s Vault setup tutorial.
npx @agentvalet/registervault server -dev -dev-root-token-id root -dev-tlsWhich one fits
Choose HashiCorp Vault if
- You need one secrets platform for services, pipelines and people, not only agents.
- You must self-host, or keep everything in your own network.
- You rely on dynamic secrets for databases or cloud accounts.
Choose AgentValet if
- Your agents act on SaaS platforms or MCP servers and shouldn’t hold any secret.
- You want per-action grants and approvals without writing Vault policies.
- You want to start with one command and nothing to operate.
Run both if
- Vault already serves your infrastructure secrets and you’re adding agents that act on outside platforms.
Checked against Vault and agentic AI, HashiCorp’s announcement of AI agent support in Vault, the validated pattern for AI agent identity, Vault MCP server best practices, Vault leases, Vault audit devices, Vault control groups, HCP pricing definitions, the Vault setup tutorial, the Vault licence and the Vault MCP server repository on 8 November 2026. If something has changed, report a correction and we’ll fix the page.
Sources
- Vault and agentic AI, accessed 8 November 2026
- HashiCorp’s announcement of AI agent support in Vault, accessed 8 November 2026
- the validated pattern for AI agent identity, accessed 8 November 2026
- Vault MCP server best practices, accessed 8 November 2026
- Vault leases, accessed 8 November 2026
- Vault audit devices, accessed 8 November 2026
- Vault control groups, accessed 8 November 2026
- HCP pricing definitions, accessed 8 November 2026
- the Vault setup tutorial, accessed 8 November 2026
- the Vault licence, accessed 8 November 2026
- the Vault MCP server repository, accessed 8 November 2026
Govern one agent first.
Register an agent on the free plan with npx @agentvalet/register, grant it one scope, then try a call outside it. You’ll see the refusal, and the audit row it leaves behind.
npx @agentvalet/register