Can Ruby on Rails run AI agent workflows?

Yes. Rails is a good host for AI agent workflows because it already has durable models, authorization, background jobs, admin screens, and audit trails. Calling a model is the small piece; the reliability work is tool design, state, retries, approvals, observability, and cost control.

Should AI agents run inside web requests?

No. Multi-step agents should run as background jobs because they can call models several times, execute tools, wait for human approval, stream progress, and continue across deploys. Controllers should create a run record, enqueue work, and return a status endpoint or stream target.

What makes an AI agent reliable in a Rails app?

A reliable agent has narrow tools, authorization inside every tool, approval gates for writes, stored step traces, bounded retries, usage tracking, timeouts, and tests around provider response shapes. Prompt quality matters, but reliability comes from ordinary application boundaries.

Content Hub

Rails AI Agents: Tools, Approvals, and Background Jobs

Guides for building AI agent workflows in Ruby and Rails, written for developers wiring Claude or Gemini into an app that already has users, jobs, and a database. If the steps are predictable, use a workflow instead. Reach for an agent when the model has to choose the next step from intermediate results.

NEED CLAUDE IN RAILS

Start with the Anthropic walkthrough when you need the full loop: tools, approvals, background jobs, streaming, and tests.

ONLY NEED SDK SETUP

Use the Anthropic Ruby SDK reference for client setup, messages, streaming, tools, retries, and when another gem fits better.

NO RUBY SDK

Use the Gemini guide when the provider boundary is plain HTTP and your app owns request shapes, parsing, streaming, and retries.

Start here

Rails AI Agents with the Anthropic SDK

The Rails shape: agent loop, tool classes, streaming, MCP, authorization, approval gating, and tests.

Choose the client boundary

SDK REFERENCE

Anthropic Ruby SDK Reference

The official anthropic gem, not the renamed ruby-anthropic: client lifecycle, messages, streaming, tools, retries, and when another gem fits better.

GEM COMPARISON

Anthropic Ruby SDK vs ruby_llm

The official typed SDK against the multi-provider framework: feature coverage, real gaps, and when each wins.

AGENT SDK

Claude Agent SDK in Ruby

There is no official Ruby Agent SDK. The three real paths compared, with a working agent loop on the anthropic gem.

CLAUDE CODE

Claude Code for Rails

Set up Claude Code in a Rails project: CLAUDE.md, allowed commands, denied files, Skills, hooks, MCP, and editor diagnostics.

MCP SERVERS

Ruby MCP Servers

Build a Model Context Protocol server in Ruby, expose Rails data as tools, and connect it to Claude.

Run the loop

GEMINI

Gemini API in Ruby Without an SDK

Faraday transport, Interactions API state, function calls, background execution, streaming, and parser fixtures.

BACKGROUND JOBS

Running AI Agents with Solid Queue

Queue isolation, worker throttling, resumable runs, human-in-the-loop pauses, and per-account cost tracking.

SCHEDULED AGENTS

Recurring Agent Work

How Solid Queue recurring jobs affect polling agents, scheduled inference, and at-least-once execution safety.

Expose tools

BUSINESS SYSTEMS

API Integrations for Agent Tools

Agents become useful when their tools read vendor systems with scoped credentials and request approval before external writes.

Building an AI workflow inside Rails?

With agents, calling the model is the small part; the real work is deciding where the agent is allowed to read, write, retry, and stop. Map that permission boundary across your tools and tenant scope first, and the SDK choice, tool design, and execution hardening follow from it.

View AI agent services