Corsair

About Corsair

Corsair is an open source integration layer for AI agents and applications, providing a software development kit that connects products to third-party apps such as Gmail, Slack, Notion, HubSpot, Google Calendar, Stripe, Airtable, Linear, GitHub, PostHog, Resend, Sheets, and Discord. It handles the underlying integration work of authentication flows, token refresh, webhook signature verification, API deprecations, rate limit handling, and multi-tenant credential isolation, exposing method calls and results to an agent without the agent seeing API keys, since credentials are resolved internally at call time. It supports permission modes per integration, requiring explicit approval for destructive actions, and caches data via webhooks and polling so repeated queries hit a database rather than the third-party API on every request.

The product can be self-hosted at no cost, running the full SDK on a customer's own infrastructure so credentials and data stay in that infrastructure rather than with Corsair. It is intended for developers building AI agent products or applications that need to call third-party services, whether through an agent or through direct calls to methods and actions like syncing data or posting messages, without maintaining that integration infrastructure themselves. Being open source, it allows developers to add missing integrations themselves via pull request or by forking the repository.

It offers a Hobby plan at $0/month for small projects with up to 50 connections, 100k webhook events, and up to 3 team members; a Pro plan at $200/month for teams in production with unlimited connections, webhooks, and team members; and a custom-priced Enterprise plan. All plans include unlimited tool calls and managed permissions/auth pages. Support is provided via Discord community on Hobby and Slack on Pro and Enterprise plans.

Key features

• Pre-built integrations for apps like Gmail, Slack, Notion, HubSpot, GitHub, and Stripe
• Per-integration permission modes requiring explicit approval for destructive actions
• Self-hostable open source SDK with no per-seat pricing or vendor markup on API calls
• Credential resolution handled internally so agents never see API keys
• Data caching kept fresh via webhooks and polling, partitioned per tenant
• Usable as a typed library without requiring an AI agent

Frequently asked questions