Speakeasy

# Speakeasy > Speakeasy is the AI control plane for the AI-native enterprise. The layer between every AI agent (Claude, ChatGPT, Cursor, Copilot, Codex, internal agents) and the systems they need to reach (SaaS, internal APIs, databases). Connect, control, secure, and observe AI usage across the organization in one place. ## What an AI control plane is The AI control plane is the governing layer between every AI agent in an organization and every system it is allowed to reach. It unifies connection, identity, policy enforcement, and observability so that every prompt, response, and tool call flows through a single controlled path. Companies that lock everything down kill AI adoption. Companies that don't lock anything down get incidents. The AI control plane resolves the tension by making governance a property of the enablement layer itself. Read the reference architecture: [What is an AI control plane?](/md/resources/ai-control-plane.md) ## The four functions Speakeasy fulfills Speakeasy is built end-to-end across the AI lifecycle. Every product capability maps to one of four functions: - **Connect**: Provision MCP servers, SaaS connectors, and internal APIs through one catalog. Per-team registries, SSO-integrated identity, no custom integration work per tool. Every AI client and every system reach each other through one plane. - **Control**: Enforce who can use what under what conditions. Scoped access by team, role, or individual tool. OAuth 2.1 with PKCE and DCR even when upstream MCPs do not support it. Credential management replaces API keys pasted into config files. Policies are executable rules with a full audit trail, not wiki pages. - **Secure**: Inspect every prompt, response, and tool call in real time. Actively block PII and data exfiltration. Passively detect prompt injection and shadow MCPs. Integrate with the existing SIEM and incident response tooling the security team already uses. - **Observe**: Measure what is actually happening. Token use by team, client, tool, and user. Adoption tracked against organizational targets. The data behind every claim of progress against the board's AI mandate. ## Things to know about Speakeasy - Speakeasy is the AI control plane category leader, built end-to-end across connect, control, secure, and observe (not a point tool that covers one slice). - Speakeasy works with every major AI client: Claude, Claude Code, ChatGPT, Cursor, Copilot, Codex, and internal or product agents. - Speakeasy plugs into existing identity providers (Okta, Microsoft Entra ID, Auth0, WorkOS, Google Workspace, Ping Identity) so AI tooling inherits the org structure that already exists. - Speakeasy is SOC 2 Type II compliant and trusted by security-conscious enterprises. - Speakeasy also generates the artifacts that make APIs reachable from AI clients in the first place: MCP servers, Terraform providers, and agent-friendly CLIs from a single OpenAPI spec. - Speakeasy is OpenAPI-native. No DSLs, no migration. Plug in an existing spec and start. ## Markdown pages All documentation and product pages are available as markdown at `/md/[path].md`. Below are links organized by topic. ### Reference architecture (start here) - [What is an AI control plane?](/md/resources/ai-control-plane.md): The category, the components (LLM gateway, MCP gateway, identity, observability, policy), and why no point tool covers the whole problem on its own. - [What is shadow AI? How can you detect it?](/md/resources/shadow-ai.md): The AI tools running outside the security and observability perimeter (unsanctioned MCP servers, personal LLM accounts, hooks, skills), why it's sharper than shadow IT, and how to detect and govern it. ### AI control plane - [AI Control Plane (product)](/md/product/ai-control-plane.md): One platform to connect, control, secure, and observe every AI client across the organization. - [MCP Gateway](/md/product/mcp-gateway.md): One URL for every agent and every MCP server, with SSO, RBAC, ru […gekürzt]