Tilde vs AI SDK

The short answer

The AI SDK and Tilde work at different levels. The AI SDK is a library: it standardises how your code talks to models, runs agent loops and calls tools. Tilde is a platform: it gives those agents authenticated tools, screens what goes in and out, and decides who can use what.

In fact, Tilde's harness SDK is designed to sit on top of the AI SDK. You write your agent with the AI SDK and use Tilde for the parts that need a platform.

At a glance

Where AI SDK is strong

The AI SDK is one of the most widely used ways to build AI features in TypeScript. It gives you a single API across dozens of model providers, agent classes including durable workflow agents, an MCP client, tool approval and polished chat UI hooks for React, Vue and Svelte.

It's open source under Apache 2.0, runs anywhere Node.js does, and has built-in OpenTelemetry. Around it, Vercel offers an AI Gateway, sandboxes for running code and a Chat SDK for channel adapters.

Where Tilde is different

The AI SDK leaves integrations, credentials, guardrails and access control to your code. Tilde provides them as a platform, so each new agent doesn't rebuild them.

• Integrations without the plumbing. Create MCP tool servers from our catalog of SaaS providers, proxy an existing MCP server, or forward requests to your own API.
• Credentials you don't have to build. Tilde handles OAuth, API keys and custom authentication, including encryption and rotation.
• Security middleware on every agent. Tilde blocks jailbreak attempts and redacts personal information (PII) in agent inputs and outputs.
• Access scoped to the work. Workspaces and grants limit which people and agents can reach which resources, with single sign-on through your OIDC provider.
• Control on sensitive actions. Govern checks policy, asks a person to approve and records the decision before an action runs.
• Channels in a single click. Add WhatsApp, Slack, Discord, Telegram or GitHub to an agent without writing an adapter. ChatKit adds message history, sessions, webhooks, schedules and agent-to-agent delegation.
• Secrets kept out of the model. The secure browser signs in to websites with injected credentials the agent never sees.
• Voice when you need it. Build realtime voice agents on the same tools and permissions.
• One inventory. The Registry records every agent, who owns it and who can use it, whichever framework built it.
• Runs where your data lives. Use Tilde cloud, or deploy Tilde on-prem in your own infrastructure. Embedded engineers can help you set it up.

Using AI SDK and Tilde together

This is the setup we recommend. Build the agent with the AI SDK, then add Tilde's harness SDK for tools, credentials, memory and channels. Your AI SDK agent calls Tilde MCP servers as tools, and Tilde applies security middleware and permissions on every call.

Register the deployed agent in the Tilde Registry so it has an owner and an access scope, and connect it to users through Tilde channels.

flowchart TB
  People["People on Slack, WhatsApp or web"] --> Channels
  subgraph App["Your app"]
    SDK["AI SDK agent"]
    Harness["Tilde harness SDK"]
  end
  subgraph Tilde["Tilde"]
    Channels["Channels"]
    Guard["Security middleware"]
    MCP["MCP tools and managed credentials"]
    Registry["Registry and IAM"]
  end
  Channels --> SDK
  SDK --> Harness
  Harness --> Guard
  Guard --> MCP
  MCP --> Systems["SaaS APIs and your systems"]
  Registry -.-> SDK

Which should you choose?

• Use the AI SDK on its own for AI features and simple agents where you control every integration yourself.
• Add Tilde when agents need many authenticated integrations, managed credentials, runtime guardrails, workspace-level access, approvals or channels, in Tilde cloud or on-prem.
• Most production teams use both. Talk to our team if you'd like help planning it.