Skip to the collection

agent-native · Oversight & Approval · Verified 2026-09-26

Sallyport

Listed on The Agent-Native Index

Let your agent touch prod. Keep the keys

What it does

Sallyport is a local Mac credential vault and execution gate for AI agents. An agent asks for a structured HTTP request, SSH command, credential-provisioning request, or upstream MCP tool call; Sallyport resolves the bound credential, applies its fixed authorization ladder, writes a durable audit intent, executes the operation, and returns a bounded result without providing a credential-reveal command.

The boundary is intentionally precise. Stored keys remain in the app for HTTP and SSH operations, but configured local stdio MCP servers necessarily receive credentials in their child environments. Target/upstream responses are returned without content inspection and may echo sensitive values. Sallyport protects its own execution path; it cannot govern traffic the agent sends elsewhere.


Agent-facing interfaces

How agents onboard

brew install --cask olegsotnikov/tap/sallyport

See the full onboarding section in the dossier.

Protocol surface

Interface Detail
MCP stdio sp mcp exposes Sallyport tools to a local MCP client
Private Unix socket Connects the minimal shim to Sallyport.app with kernel peer identity
HTTP executor Structured GET/POST/PUT/PATCH/DELETE/HEAD with bearer, basic, custom-header, AWS SigV4, or OAuth2 client-credentials adapters
SSH executor Configured-host command execution with challenge signing and encrypted asciicast recording
Upstream MCP Local stdio or remote Streamable HTTP with API key or OAuth 2.1 support
Audit JSONL ECIES-sealed rows, hash chain, ECDSA signatures, and an integrity anchor under ~/.sallyport/audit

Verification

Last verified 2026-09-26.

Sources: