Launch HN: Alter (YC S25) – Access control for multi-user agents

Alter provides a privileged access management platform specifically for AI agents, replacing scattered, over‑privileged keys with per‑agent identities, scoped credentials, and a policy engine that enforces least privilege. Instead of embedding API keys in prompts or repos, credentials live in a vault and are injected per request via an SDK; every request is authenticated, authorized against default‑deny rules (including parameter‑level checks), and logged. Sensitive operations can be routed to a named human approver, and revocation or onboarding cascades through IdP webhooks so access can be removed instantly. The core argument is operational security: giving agents their own keys and auditable traces prevents one master key from being reused across agents and provides an evidentiary trail for incident response and customer security reviews.

The platform bundles a credential vault, policy engine, agent identity management, and a human‑in‑the‑loop workflow with integrations to about 140 providers and first‑class SDKs for common agent frameworks. Implementation details stress hardened practices: a dedicated encrypted secret store, plaintext credentials never in logs, proxy‑only keys that cannot extract secrets, and fail‑closed policy evaluation; SOC 2 is in progress and continuous monitoring and penetration testing are published in a trust center. Practical uses include running OAuth flows on behalf of users, giving agents scoped access, provisioning backend secrets, partitioning agent memory by identity, and building multi‑user servers without exposing credentials.