Installation

Agent RT ships as a Python package and a TypeScript/Node package. Choose the runtime your application already uses; the core contracts intentionally follow the same mental model in both implementations.

Requirements

Runtime	Supported versions	Install
Python	3.10–3.14	`pip install agent-rt`
Node.js	20, 22, 24, or 26	`npm install agent-rt`

Python

python -m venv .venv
source .venv/bin/activate
pip install agent-rt

TypeScript / Node.js

npm install agent-rt

Docker Hub images

Prebuilt images are also available on Docker Hub: Agent RT Python and Agent RT TypeScript. Pull the latest tag, or use a published version tag such as 0.1.2:

docker pull praneeth1v/agent-rt-python:latest
docker pull praneeth1v/agent-rt-typescript:latest

The Python image starts the agent-rt CLI; the TypeScript image starts Node.js and has the agent-rt npm package preinstalled:

docker run --rm praneeth1v/agent-rt-python:latest --help
docker run --rm praneeth1v/agent-rt-typescript:latest node -e "console.log(typeof require('agent-rt').AgentLoop)"

The Python image includes the API and web extras; the TypeScript image includes the OpenAI and Anthropic SDK peers. Docker Hub displays each image's pull count; the GitHub README also includes live pull-count badges. Provide provider credentials/configuration at runtime, not in your image.

Configure models

Agent RT is configured with one YAML file: ~/.agent-rt/config.yaml in your home folder, or a project file explicitly passed with --config / load_config(path). Declare as many models as you need; secrets are referenced as os.environ/NAME and are the only thing read from the environment.

model_list:
  - model_name: my-model
    agent_rt_params:
      model: openai/your-model          # or anthropic/...; add api_base for compatible endpoints
      api_key: os.environ/OPENAI_API_KEY

Repeating a model_name adds an ordered fallback. See the configuration reference and the example file.

Provider-neutral by designAgent code is built against Agent RT contracts. OpenAI, Anthropic, compatible endpoints, Decision models, and custom providers can sit behind that boundary.

Vector databases

No Chroma, Pinecone, Qdrant, Milvus, Weaviate, LangChain, or LlamaIndex package is required for Agent RT's built-in vector retrieval and compatibility adapters. Declare services in vector_store_registry (several are allowed), then see the vector database guide for backend credentials and framework migration imports.

PreviousDocumentation overview NextQuickstart →