Tyk AI Studio Plugin Overview - Tyk Documentation

Availability

Edition	Deployment Type
Community & Enterprise	Self-Managed, Hybrid

Tyk AI Studio’s plugin system enables powerful extensibility across the entire platform through a Unified Plugin SDK. Built on HashiCorp’s go-plugin framework, plugins run as isolated processes with gRPC communication, providing security and fault tolerance.

Unified Plugin SDK

All plugins now use a single SDK (pkg/plugin_sdk) that works seamlessly in both AI Studio and Edge Gateway contexts. The SDK automatically detects the runtime environment and provides appropriate capabilities.

Key Features

Single Import: One SDK works everywhere
10 Plugin Capabilities: Mix and match to build exactly what you need
Runtime Detection: Automatic AI Studio vs Edge Gateway detection
Service API Access: Built-in KV storage, logging, and management APIs
Type-Safe: Clean Go interfaces, no manual proto handling

Plugin Capabilities

Plugins implement one or more of these 12 capabilities:

Capability	Where It Works	Purpose	Common Use Cases
Pre-Auth	Studio + Gateway	Process before authentication	IP filtering, request validation
Auth	Studio + Gateway	Custom authentication	OAuth, API keys, JWT validation
Post-Auth	Studio + Gateway	Process after authentication	Request enrichment, policy enforcement
Response	Studio + Gateway	Modify responses	Content filtering, header injection
Data Collection	Studio + Gateway	Collect telemetry	Export to Elasticsearch, ClickHouse
Custom Endpoints	Gateway	Serve custom HTTP endpoints	MCP proxy, OAuth provider, webhooks
Object Hooks	Studio only	Intercept CRUD operations	Validation, approval workflows
Agent	Studio only	Conversational AI	Chat-based agents, LLM wrapping
UI Provider	Studio only	Dashboard extensions	Custom dashboards, admin tools
AI Portal UI	Studio only	AI Portal extensions	User-facing forms, pages, dashboards
Config Provider	Studio + Gateway	Provide JSON Schema config	Dynamic configuration
Manifest Provider	Gateway only	Plugin manifest	Gateway-only plugins

Multi-Capability Plugins

A single plugin can implement multiple capabilities. For example, a rate limiter might:

Implement Post-Auth to check limits before request
Implement Response to update counters after response
Implement UI Provider to show rate limit dashboard

Plugin Types Overview

While all plugins use the unified SDK, they generally fall into three categories based on their primary use case:

1. Edge Gateway Plugins

Edge Gateway plugins provide middleware hooks in the LLM proxy request/response pipeline using the unified SDK. Learn more →

2. AI Studio UI Plugins

AI Studio UI plugins extend the dashboard with custom WebComponents, adding new pages, sidebars, and interactive features to the admin interface. These also use the unified SDK and can combine UI capabilities with middleware hooks. Learn more →

3. AI Studio Agent Plugins

Experimental Feature: Agent plugins are currently experimental. The API and behavior may change in future releases.

Agent plugins enable conversational AI experiences in the Chat Interface using the unified SDK. These plugins can wrap LLMs, add custom logic, integrate external services, and create sophisticated multi-turn conversations. Learn more →

4. Object Hooks Plugins

Object Hooks are a powerful AI Studio-only capability that allows plugins to intercept and control CRUD operations on key objects before they reach the database. This is particularly useful for validation, approval workflows, and policy enforcement. Learn more →

Plugin Architecture

Monolithic Plugin Architecture

A single plugin binary can run in both AI Studio and the Edge Gateway, so long as the requisite interfaces are implemented and the requirements are clear in the manifest file. This means one plugin can provide UI extensions in Studio, middleware hooks in the gateway, and even use the event bus to communicate between its Studio and gateway components in near-real-time.

Scheduled Tasks

Studio plugins can register scheduled tasks — periodic calls from the host to the plugin on a configurable interval. This enables long-running background jobs such as analytics analysis, compliance checks, log scanning, or data synchronization.

Process Isolation

Plugins run as separate processes, communicating with the main platform via gRPC. This provides:

Security: Plugin crashes don’t affect the main platform
Language Flexibility: While the gRPC protocol theoretically supports any language, only Go plugins have been tested in production. If you plan to write plugins in another language, expect to do additional integration work.
Resource Management: Plugins can be restarted independently
Version Independence: Update plugins without platform restarts

Plugins can also run as standalone gRPC services in a sidecar or elsewhere on the network, rather than as local sub-processes. See Plugin Deployment for details.

Communication Flow

Service API (AI Studio Plugins Only)

AI Studio UI and Agent plugins can access the Service API via a reverse gRPC broker connection: The Service API provides 100+ gRPC operations for managing LLMs, apps, tools, datasources, analytics, and more. Access is controlled via permission scopes declared in the plugin manifest. Learn more about Service API →

Deployment Options

Plugins support three deployment methods:

file://

Local filesystem plugins for development and testing:

file:///path/to/plugin-binary

grpc://

Remote gRPC plugins running as network services:

grpc://plugin-host:50051

oci://

Container registry plugins (OCI artifacts):

oci://registry.example.com/plugins/my-plugin:v1.0.0

Learn more about deployment →

Upgrade Installed Plugins

To publish a plugin to a marketplace and release new versions, refer to Publish Plugins. From v2.2.0, AI Studio tracks the marketplace version of each plugin that you installed from the marketplace. After each marketplace sync, the Plugins page shows the installed version. It also shows an Update chip on each plugin that has a newer published version. AI Studio recognizes plugins that you installed before this version from their OCI reference, so you do not need to install them again. To upgrade a plugin, select Upgrade on the chip, the row menu, the plugin detail page, or the marketplace card of the plugin. Before you apply the upgrade, AI Studio shows:

The target version. You can also select a different published version, also an older one.
The changelog of that version
The new permissions that the version asks for. You must approve them to continue. AI Studio removes the permissions that the new version does not use.
A warning if your current configuration does not match the settings of the new version. This warning does not block the upgrade.

An upgrade replaces only the plugin binary. The plugin keeps its configuration and data: KV data, agents, LLM associations, schedules, the Apps that use its resources, and the permissions granted on it. If the new version does not start, AI Studio restores the previous version and shows the reason. If the plugin also runs on Edge Gateways, AI Studio marks them as pending after the upgrade. They load the new version with the next configuration push. Upgrades need the plugins:write permission. AI Studio does not track plugins that you added from a local file or from a private registry that is not a marketplace source. To upgrade these plugins, edit their command.

Known Issues

Reactivating a plugin does not restart its AI Studio part. After you turn a plugin off and on again, click Reload on the plugin, or call POST /api/v1/plugins/{id}/reload.
Running Edge Gateways do not reload an in-place upgrade that uses a file:// command. Restart the Edge Gateways after the upgrade. Marketplace (OCI) upgrades change the checksum and reload as usual.
AI Studio does not enforce min_studio_version. The marketplace shows the value, but you can install a plugin on an older AI Studio version. Check the minimum version before you install.

Permissions and Scopes

AI Studio plugins declare required permissions in their manifest:

{
  "permissions": {
    "services": [
      "llms.proxy",      // Call LLMs via proxy
      "llms.read",       // List and read LLM configs
      "tools.execute",   // Execute tools
      "datasources.query", // Query datasources
      "kv.readwrite",    // Key-value storage
      "analytics.read"   // Read analytics data
    ]
  }
}

Permissions are validated when plugins call the Service API. The platform enforces least-privilege access based on declared scopes. Learn more about manifests →

Getting Started

Choose Your Plugin Type

Need to intercept/modify LLM requests? → Edge Gateway Plugin
Building dashboard UI features? → AI Studio UI Plugin
Creating conversational AI experiences? → AI Studio Agent Plugin

Development Workflow

Choose your plugin type
Read the specific plugin guide
Review example plugins in examples/plugins/ and community/plugins/
Use the SDK to implement required interfaces
Build and test with file:// deployment (see Development Workflow Guide for fast iteration)
Deploy with grpc:// or oci:// for production

Pro tip: Use the reload API (POST /api/v1/plugins/{id}/reload) to test changes instantly without reinstalling. See the Development Workflow Guide for the fastest iteration loop.

SDK Installation

All plugins use the unified SDK:

go get github.com/TykTechnologies/midsommar/v2/pkg/plugin_sdk

import "github.com/TykTechnologies/midsommar/v2/pkg/plugin_sdk"

type MyPlugin struct {
    plugin_sdk.BasePlugin
}

func main() {
    plugin_sdk.Serve(NewMyPlugin())
}

Note: If you have existing plugins using the old SDKs (microgateway/plugins/sdk or pkg/ai_studio_sdk), see the Migration Guide for upgrade instructions. From v2.2.1, the gateway plugin interfaces are in pkg/gatewayplugin, and the old microgateway/plugins paths forward to them.

​Availability

​Unified Plugin SDK

​Key Features

​Plugin Capabilities

​Multi-Capability Plugins

​Plugin Types Overview

​1. Edge Gateway Plugins

​2. AI Studio UI Plugins

​3. AI Studio Agent Plugins

​4. Object Hooks Plugins

​Plugin Architecture

​Monolithic Plugin Architecture

​Scheduled Tasks

​Process Isolation

​Communication Flow

​Service API (AI Studio Plugins Only)

​Deployment Options

​file://

​grpc://

​oci://

​Upgrade Installed Plugins

​Known Issues

​Permissions and Scopes

​Getting Started

​Choose Your Plugin Type

​Development Workflow

​SDK Installation