Sesame: Secrets Stay Secret. Even from Your AI Agent.

Sesame is a user-controlled credential broker that lets your AI agents make authenticated API calls without ever seeing your API keys. Instead of injecting secrets into agent prompts or environment variables, Sesame intercepts every request, attaches the right Authorization header server-side, and forwards only the response back to the agent. You approve every new access, see every request in real time, and can revoke any agent in one click.

Introduction

Learn what Sesame is, why it exists, and how the zero-trust brokering model protects your credentials.

Quickstart

Install the CLI and make your first brokered request in under five minutes.

How It Works

Understand Sesame’s zero-trust architecture and credential-injection model.

CLI Reference

Full reference for every sesame command, flag, and option.

Agent Skills

Install the Sesame skill so your AI agent automatically routes calls through the broker.

Integrations

Browse 70+ pre-built provider integrations and MCP server support.

Get Up and Running

Install the CLI

Run the installer on macOS (arm64/x86_64) or Linux (x86_64):

curl -fsSL https://getsesame.dev/install.sh | sh

Generate a cryptographic device identity and link it to your Sesame account:

sesame login

This opens a one-click claim URL in your browser. After approval, your device is registered and ready to broker requests.

Make an authenticated request

Call any API without an API key in sight:

sesame request POST https://api.stripe.com/v1/payment_intents \
  -H "Content-Type: application/x-www-form-urlencoded" \
  -d "amount=2000&currency=usd"

Sesame looks up the secret configured for api.stripe.com, injects the Authorization header, and returns the response.

Add the agent skill (optional)

Teach your AI agent (Claude Code, Codex, Cursor, and more) to use Sesame automatically:

npx skills add getsesame/skills

Why Sesame?

Secrets Never Exposed

Credentials are injected at the broker — they never appear in agent prompts, logs, tool arguments, or memory.

Human Approval

Every first access to a new hostname requires your explicit approval via the app or Telegram.

Instant Revocation

Cut off an agent, a single grant, or the entire door in one click — no credential rotation needed.

70+ Providers

Pre-built integrations for Anthropic, Stripe, GitHub, Slack, and dozens more out of the box.

Audit Trail

Immutable logs capture every proxied request, approval, and revocation — with credentials redacted.

MCP Support

Connect MCP servers alongside API keys. Your agent gets the tools, never the tokens.