Migration
Python and TypeScript applications can adopt Agent RT incrementally through compatibility entry points, then move toward native runtime contracts when they need deeper control over tools, state, policy, or execution.
Compatibility surfaces
For common imports, migration can be as small as changing the module prefix. Keep the class or symbol you already use and point the import at Agent RT's compatibility layer.
Python
from langchain_openai import ChatOpenAIfrom agent_rt.langchain_openai import ChatOpenAIfrom llama_index.llms.openai import OpenAIfrom agent_rt.llama_index.llms.openai import OpenAIfrom langchain_qdrant import QdrantVectorStorefrom agent_rt.langchain_qdrant import QdrantVectorStorefrom llama_index.vector_stores.qdrant import QdrantVectorStorefrom agent_rt.llama_index.vector_stores.qdrant import QdrantVectorStorefrom openai import AsyncOpenAIfrom agent_rt.openai import AsyncOpenAIfrom anthropic import AsyncAnthropicfrom agent_rt.anthropic import AsyncAnthropicTypeScript / JavaScript
import { ChatOpenAI } from "@langchain/openai";import { ChatOpenAI } from "agent-rt/@langchain/openai";import { OpenAI } from "@llamaindex/openai";import { OpenAI } from "agent-rt/@llamaindex/openai";import { QdrantVectorStore } from "@langchain/qdrant";import { QdrantVectorStore } from "agent-rt/@langchain/qdrant";import { QdrantVectorStore } from "@llamaindex/qdrant";import { QdrantVectorStore } from "agent-rt/@llamaindex/qdrant";import OpenAI from "openai";import OpenAI from "agent-rt/openai";import Anthropic from "@anthropic-ai/sdk";import Anthropic from "agent-rt/@anthropic-ai/sdk";Adopt in layers
Start at the compatibility entry point closest to the code you already have. Once behavior is stable, move runtime-sensitive concerns—tools, run limits, sandboxing, retrieval, memory, policy, and checkpoints—onto Agent RT-native contracts where the control boundary is explicit.
Supported compatibility themes
The migration layer covers common chat/completion entry points, streaming, tool history, OpenAI Responses continuation, file-search integration through Agent RT retrieval providers, web retrieval integration, reasoning/thinking controls, batch lifecycle compatibility, and read/query vector-store adapters for Chroma, Pinecone, Qdrant, Milvus, and Weaviate.
LangChain compatibility exposes vendor-shaped imports such as agent_rt.langchain_qdrant and agent-rt/@langchain/qdrant. LlamaIndex compatibility exposes agent_rt.llama_index.vector_stores.qdrant and agent-rt/@llamaindex/qdrant, including integration with VectorStoreIndex.from_vector_store() / fromVectorStore(). These adapters query existing collections; vendor write/admin lifecycle APIs remain outside the compatibility layer. See the vector database guide for the complete mapping.