Agent RT Documentation
Documentation menu
Overview Installation Quickstart Runtime model Tools & policy State & memory Migration Production patternsVector databasesRuntime guide Features & defaults Comparison Architecture DevelopmentPrivacy & terms
Agent RT documentation

Agent RT
Documentation

Start with a runnable agent, then move into tools, state, policy, production operations, and extension contracts. The website documentation is organized by task instead of repository layout.

Solutions journal Building around safety, reliability, performance, or framework selection? Read the Agent RT blog for problem-first articles with implementation checklists.

Start here

01
Install Agent RTInstall from PyPI, npm, or Docker Hub, then configure a model provider.
02
Run your first agentMove from a minimal call to tools, streaming, structured output, and limits.
03
Understand the runtime loopSee how model requests, tool execution, state, budgets, and termination fit together.
Section 1

Getting started

From install to first run.

↗SetupInstallation

Supported runtimes, Docker Hub images, package installation, and provider environment.

 ↗First runQuickstart

Incremental examples for agents, tools, streaming, schemas, and limits.

 ↗Complete theoryIntroduction to agent harnesses

The complete reference for agent loops, tools, context, state, memory, permissions, safety, and deployment.

 ↗Detailed handbookComplete runtime guide

Full Python and TypeScript examples for providers, CLI, memory, skills, sandboxes, chat bots, and more.

 ↗API compatibilityComplete migration reference

Detailed migration recipes for frameworks, model SDKs, memory, tool APIs, and provider compatibility.

 ↗YAML schemaConfiguration reference

Model lists, fallbacks, vector store settings, runtime options, and secrets.

 ↗Complete contractsExtension contracts

Module boundaries, security invariants, provider-state contracts, and conformance testing.

 ↗Contributor workflowComplete development handbook

Python and TypeScript tests, security scans, packaging, evaluation, and CI.

 ↗Launch reviewDeployment privacy checklist

Practical review points for hosting, accounts, marketing email, user content, and billing.
Section 2

Core concepts

The mental model behind Agent RT.

↗ExecutionRuntime model

The model → tool → model loop, routing, limits, and cancellation.

 ↗CapabilitiesTools & policy

Registries, validation, visibility, permissions, approvals, and guardrails.

 ↗PersistenceState & memory

Messages, workflow state, memory scopes, checkpoints, artifacts, and resumption.
Section 3

Guides

Common adoption and deployment paths.

↗AdoptionMigration

Compatibility surfaces for LangChain, LlamaIndex, OpenAI, and Anthropic applications, with 65+ third-party repositories passing before and after migration in ongoing field testing.

 ↗OperationsProduction patterns

Boundaries, durable execution, sandboxing, observability, evaluation, and deployment.

 ↗RetrievalVector databases

Switch Chroma, Milvus, Pinecone, Qdrant, or Weaviate with environment variables, or use LangChain/LlamaIndex-compatible vector-store imports.

 ↗Usage detailsRuntime guide

Providers, memory packages, API server, CLI, web UI, chat bots, sandboxes, skills, and migration entry points.
Section 4

Reference

Architecture and contributor workflows.

↗Complete referenceFeatures & defaults

Every major capability, what is enabled by default, what is opt-in, execution order, and how to configure it.

 ↗Framework comparisonAgent RT vs alternatives

Compare runtime features, latency, memory, guardrails, APIs, sandboxing, and portability.

 ↗ExtensionsArchitecture

Module boundaries and the extension contracts that keep the runtime provider-neutral.

 ↗ContributingDevelopment

Repository setup, test matrices, and language-specific workflows.