Agent Proxy gives an AI agent a fake credential and substitutes the real one at the network boundary, so an agent that is compromised or manipulated has no secret to leak.
SAN FRANCISCO, July 30, 2026 /PRNewswire-PRWeb/ -- Infisical today announced the general availability of Infisical Agent Proxy, which lets an AI agent use a credential without ever receiving its real value. It is built for the developers, platform engineers, and security teams moving AI agents into production.
The traditional way of doing secrets management worked: an application fetched the secret and used it. That was safe because an application does only what it was programmed to do. AI agents need those same credentials to call an API, query a database, or reach an internal service, but agents don't come with a safety guarantee. A prompt injection, a compromised environment, or a single request to a model provider is enough to put a live credential somewhere it was never meant to go. Security teams know this, which is why agent projects sit in review instead of production.
How Agent Proxy Works
With Agent Proxy, the agent is configured with a placeholder value rather than a real secret, and its outbound requests route through the proxy. Infisical catches the placeholder, fetches the real secret, and substitutes it on the outbound request, or replaces the auth headers entirely, before forwarding the authenticated request to its destination.
Two settings control the swap: a host pattern that matches the service being called, and an injection point that tells the proxy where in the request the credential belongs.
"Agents should never see your secrets," said Vlad Matsiiako, CEO and co-founder of Infisical. "Agent Proxy gives teams a way to do secrets management for agents and untrusted code execution, so credential risk stops being the reason a company cannot put agents into its infrastructure."
What Agent Proxy Delivers for Engineering Teams
- The agent never holds a real secret. Because the swap happens outside the agent, there is nothing inside it for a prompt injection, a compromised environment, or a model provider to take.
- No changes to how teams connect. Anything making HTTP requests through the proxy works unchanged, including API calls, SDKs, CLIs, and Model Context Protocol (MCP) servers. There is no agent code to rewrite and no prompts to adjust.
- More than 30 pre-built integrations. Host patterns and injection points arrive preconfigured for the services agents call most often, removing a setup step that is error prone to build by hand.
From Agent Vault to Agent Proxy
Infisical first took on credential exfiltration in April 2026 with Agent Vault, an open source project built on the same substitution approach. It reached 2,000 GitHub stars and 1,000 daily users within three months.
That response showed the problem was widespread enough to belong in the platform itself, and Agent Proxy is the result, built into Infisical Secrets Management.
Availability
Infisical Agent Proxy is generally available and is included on every plan, including the free plan. Existing Infisical Secrets Management customers can start using it immediately, with nothing to enable, upgrade, or purchase.
Read the Agent Proxy documentation or learn more about Infisical Secrets Management.
About Infisical
Infisical is the security infrastructure for developer and agents. Infisical's platform helps secure secrets, certificates, and access to sensitive infrastructure for companies like LG, Hugging Face, Airbus, and Wise.
Media Contact
Kate Neuens, Infisical, 1 9206500810, [email protected], https://infisical.com/
SOURCE Infisical
Share this article