AgentValet vs Infisical Agent Vault

Agent Vault and AgentValet start from the same idea: the agent should never hold the real key. Both put something between the agent and the API that attaches the credential on the way through. They differ on who runs that layer, what the unit of access is, and whether a person can step in on a single call.

Updated

The short answer

Pick AgentValet if you want each agent to have its own signing identity and a person able to approve risky actions at the broker, as a managed service with nothing to run.

Pick Infisical Agent Vault if you want an open-source credential proxy on your own machines, with access scoped down to hosts, methods and paths and priced per brokered request.

Two approaches, in plain words

What Infisical Agent Vault is

Agent Vault is Infisical’s access layer for AI agents. You put each real credential in a service (a host, the methods and paths it allows, and the key), group services into an access bundle, and start the agent with a session scoped to one bundle. The agent sends its traffic through a forward proxy, set with HTTPS_PROXY, and the proxy attaches the credential or swaps it in for a placeholder.

The proxy runs on your infrastructure, on a separate machine from the agent in Infisical’s recommended setup, and intercepts TLS with its own certificate authority, which the agent is set up to trust. Agent Vault began as an open-source research preview and is now an Infisical product.

What AgentValet is

AgentValet is a managed credential broker for AI agents. Each agent has its own signing keypair, the broker holds the credential, and every call is matched against that agent’s grant for the platform and action before the credential is decrypted.

A risky action can pause at the broker until a person approves it, and each decision is logged with the policy that made it and a signed receipt. You don’t run any part of the broker yourself.

The question that separates them

Who is making this call: a session, or a named agent?

In Agent Vault the unit of access is a session. It is created from an access bundle, the agent holds its token, and the session log records each request. Infisical notes this already beats agents using their own keys, where logs name only the credential.

AgentValet makes the agent itself the unit. Each agent signs every request with its own key, so the record names one agent across all its sessions, and the grant follows that agent. Both approaches keep the real key out of the agent’s reach.

Agent Vault enforces at a proxy you run
  1. Agent with a session token
  2. Proxy checks bundle, host, method and path
  3. Credential attached by the proxy
  4. Real API or MCP server

The agent never holds the real key. A request that breaks a service’s method or path rules is refused at the proxy.

AgentValet enforces at the platform call
  1. Your agent
  2. Broker checks identity, grant and approval
  3. Credential attached in memory
  4. 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 compared with Infisical Agent Vault
AgentValetInfisical Agent Vault
Per-agent identityEach agent gets its own RS256 signing keypair and signs every request.Sessions, each a token scoped to one access bundle, rather than a standing identity for each agent.
Credential exposure to the agentThe agent never holds or sees the credential.The agent never sees the real credential; it holds only a session token.
Token lifetimeEach request carries a short-lived token signed by the agent; upstream credentials stay in the broker.You set each session’s TTL; the API default is seven days, and the CLI revokes its session when the agent exits.
Scoping granularityPer agent, per platform and per action, deny by default.Per access bundle, down to the hosts, HTTP methods and paths each service allows.
Human approvalRisky actions pause at the broker until a person approves them.Not described in the Agent Vault docs; the bundle is fixed when the session is created.
Audit evidenceEvery decision is logged with the policy, version, result and reason, and exports as CSV or JSON.Session logs record each request the agent made, with 30-day retention on the Pro plan.
MCP supportA native MCP server, and it brokers calls to upstream MCP servers as an OAuth client.MCP servers are a service type, with OAuth sign-in (and token refresh) for servers that support it.
Self-host optionNo. It is a managed service hosted in Australia.Yes: the proxy always runs on your infrastructure, and the Infisical platform itself can be self-hosted.
Open sourceThe parts that run on your machines (MCP server, CLI, SDKs and adapters) are MIT licensed. The broker is not.Yes: the Agent Vault repository is MIT licensed outside its ee/ directory.
Setup timeOne command, npx @agentvalet/register, registers an agent.Create a project, add services to an access bundle, run the proxy, then launch the agent through the CLI.
Starting priceFree plan, with paid plans from US$99 a month.Pro includes 5,000 brokered requests a month, then US$2 per 1,000; Enterprise is custom.
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.

Infisical Agent Vault: Sessions are created by a person or a machine identity. When code creates sessions, Infisical has you use a separate machine identity that can reach only one access bundle.

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.

Infisical Agent Vault: The proxy attaches the credential as a header, or replaces a placeholder in a header, path or query string. Running the proxy on a separate machine keeps its token and CA private key out of the agent’s reach.

Token lifetime

Infisical Agent Vault: The CLI’s --ttl default is never. Once a session expires or is revoked, the proxy stops attaching its credentials when it next checks with Infisical, every 60 seconds by default.

Scoping granularity

AgentValet: The method and endpoint of each call are matched against the scope granted to that agent before any credential is decrypted.

Infisical Agent Vault: MCP tool filtering is listed on the Pro plan.

Human approval

AgentValet: Approve with a passkey, an email link or a mobile push. The approval is recorded with the action it released.

Infisical Agent Vault: Infisical has approval workflows in other products: access requests in Secrets Management on Enterprise, and in Privileged Access on every plan.

Audit evidence

AgentValet: The log is append-only and policy-stamped, and each decision carries a signed receipt.

Infisical Agent Vault: For MCP calls the log records the tool name. Enterprise adds custom retention, your own S3 storage and SIEM streaming.

Self-host option

Infisical Agent Vault: Infisical’s core platform is MIT licensed and free to self-host on any plan; supported self-hosting, including air-gapped, comes with Enterprise.

Setup time

Infisical Agent Vault: Guides cover Modal, Daytona, Kubernetes, Docker, E2B and Cursor Cloud Agents; no time estimate is published.

Where Infisical Agent Vault genuinely wins

Agent Vault is open source and its proxy runs where you choose, so if the call path must stay on your own infrastructure, or you need to read the code that handles your keys, Infisical fits and AgentValet doesn’t. Method and path rules, placeholders in headers, paths or query strings, and guides for sandboxes such as Modal, E2B and Daytona are practical strengths.

Pricing per brokered request, with the first 5,000 each month included, is easy to start on, and it sits beside Infisical’s secrets manager if you already use it.

Where AgentValet wins

AgentValet gives each agent a standing identity, a signing keypair it uses on every request, so the record names the agent rather than the session. Child agents can be issued their own identities with a subset of the parent’s scopes. A risky action can pause until a person approves it with a passkey, an email link or a push notification.

Every decision is logged with the policy, version, result and reason, and carries a signed receipt you can check. There’s no proxy or certificate authority to run, and the parts that do run on your machines are MIT licensed too.

Some teams will run both: Agent Vault in code sandboxes calling a fixed set of APIs, AgentValet for agents that need per-action grants and approval.

The first step, side by side

Each command is the documented starting point for that product. Infisical Agent Vault’s comes from the Infisical CLI reference for agents.

AgentValet
npx @agentvalet/register
Infisical Agent Vault (from its CLI reference)
infisical agent-vault run --project-id=<project-id> --access-bundle=code-review --ttl=8h --label="review PR 123" --proxy=10.0.1.5:17323 -- claude

Which one fits

Choose Infisical Agent Vault if

  • You want the call path on infrastructure you run, with code you can read.
  • Your agents run in sandboxes such as Modal, E2B or Daytona, and you want ready-made guides.
  • Session-scoped access with method and path rules covers what you need.

Choose AgentValet if

  • You want each agent to carry its own identity across sessions.
  • You want a person to approve individual risky actions before they run.
  • You’d rather not run a proxy or a certificate authority yourself.

Run both if

  • Different teams run sandboxed coding agents and business agents, with different approval needs.

Checked against the Agent Vault quickstart, How Agent Vault works, Why use Agent Vault?, the Agent Vault sessions API, the Infisical CLI reference for agents, Infisical pricing, Infisical’s Agent Vault announcement and the Agent Vault repository and licence on 8 November 2026. If something has changed, report a correction and we’ll fix the page.

Sources

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