@inkeep/ai-sdk-provider weekly downloads statistics

@inkeep/ai-sdk-provider

AI SDK provider for Inkeep Agent Framework. This package allows you to use Inkeep agents through the Vercel AI SDK.

Installation

Usage

Basic Usage

Text Generation

Streaming Responses

Provider Options

Pass options when creating a provider instance:

Additional Options

Configuration

Provider Settings

When creating a custom provider with createInkeep(), you can configure:

• baseURL - Required. The base URL of your Inkeep agents API (can also be set via INKEEP_AGENTS_RUN_API_URL environment variable) • apiKey - Your Inkeep API key (can also be set via INKEEP_API_KEY environment variable) • headers - Additional headers to include in requests

Model Options

When creating a model instance, you can configure:

• conversationId - Conversation ID for multi-turn conversations • headers - Additional headers for context (validated against agent's context config)

Environment Variables

• INKEEP_AGENTS_RUN_API_URL - Base URL for the Inkeep agents API (unless provided via baseURL option)

Features

• ✅ Text generation (generateText) • ✅ Streaming responses (streamText) • ✅ Multi-turn conversations • ✅ Custom headers for context • ✅ Authentication with Bearer tokens • ✅ Tool call observability (with x-emit-operations header)

API Endpoint

This provider communicates with the /run/api/chat endpoint of your Inkeep Agents Run API.

• Non-streaming (generateText): Sends stream: false parameter - returns complete JSON response • Streaming (streamText): Sends stream: true parameter - returns Vercel AI SDK data stream

The endpoint supports both streaming and non-streaming modes and uses Bearer token authentication.

Tool Call Observability

To receive tool call and tool result events in your stream, include the x-emit-operations: true header:

Note: Tool events are only emitted when the x-emit-operations: true header is set. Without this header, you'll only receive text lifecycle events (text-start, text-delta, text-end) and the final response.

Supported Stream Events

The provider emits the following AI SDK v2 stream events:

Text Events (always emitted): • text-start - Marks the beginning of a text stream • text-delta - Text content chunks as they arrive • text-end - Marks the end of a text stream

Tool Events (requires x-emit-operations: true header): • tool-call - Agent is calling a tool • tool-result - Tool execution completed

Control Events (always emitted): • finish - Stream completion with usage statistics • error - Stream error occurred

Description from the npm listing.