Show HN: Rails Agent – Build Autonomous AI Agents Natively in Ruby on Rails

Rails Agent, a new fullstack framework for building autonomous AI agents natively in Ruby on Rails, was announced on Hacker News. The framework allows developers to build, test, deploy, and monitor agents using Rails skills without needing to understand LLMs, embeddings, or prompt engineering. It supports Rails 7+ and Ruby 3.2+, includes a hosted cloud runtime, and offers features like encrypted provider credentials, vector store, queue workers, and tracing. Pricing includes a transparent 5% service fee on metered managed infrastructure.

The fullstack framework for building agents in Rails. Build, test, deploy, and monitor in one platform. A Rails-native DSL, coding-agent docs, Playbooks, Connectors, Slack channels via Open-Wire, Packages from open registries, and the cloud to test, deploy, and monitor. Works with Rails 7+ · Ruby 3.2+ · Cloud runtime included You shouldn't have to become an AI engineer to add agentic features to your app. Rails Agent turns the Rails skills you already have into production-ready agents — one install, docs + Playbooks, one deploy. You don't need to understand LLMs, embeddings, or prompt engineering. If you can write Rails, you can ship agents. Follow embedded Dashboard Docs and your external coding agent. Local app/agents/ files are the source of truth — scaffold blank or from a Playbook. Attach encrypted provider credentials per workspace and agent. Model spend bills your OpenAI, Anthropic, or other account directly. One platform covers the whole lifecycle: draft, test, ship, and watch — no separate tools to wire up. Hosted runtime, autoscaling, retries, and tracing are included. We handle the hard parts so you don't have to. Every trace, log, error, and cost is searchable in /agents. One place to understand every run. Every agent maps to a typed Ruby base class. Pick a type at create time or when cloning a Playbook — the scaffold includes metadata and the right gem base. Knowledge Agent Eg: Answer “Where is order ORD-1001?” from your Rails database and support docs. RailsAgents::KnowledgeAgent Workflow Agent Eg: New order → update order info in Google Sheets and send an email notification. RailsAgents::WorkflowAgent No vector database or Redis to provision. Attach provider keys under Settings → Providers. Rails Agent Cloud runs the production harness — model spend bills your provider; managed infra is pass-through at cost plus a transparent 5% service fee when metered. Your job is agent logic in app/agents/. We handle deploy, scale, traces, and monitoring. Models BYOK — attach OpenAI, Anthropic, Google, or OpenRouter keys. Optional :auto fallback when configured. Vector store Embed and retrieve knowledge. No pgvector, no Pinecone. Queue & workers Long-running tools, streaming, retries — all handled. Deploys & scaling Push once, we autoscale. Rollbacks in one click. Traces & logs Every run, every tool call, every token — searchable. Platform keys rak API keys for cloud auth — distinct from provider BYOK. No YAML soup. No hidden config. Every agent is one folder inside your existing Rails app — and you can read it top to bottom in a minute. generated by rails agent new support app/agents/support/ One directory per agent. Everything it needs lives inside it — nothing hidden in config, nothing scattered across the repo. agent.rb A plain Ruby class. Pick a model, write the instructions in English. That's the whole agent. tools/ Each file is one action — look up an order, send an email, refund a charge. Just Ruby methods on your existing models. connectors/ Connect Google Sheets, Notion, HubSpot, and 3,000+ apps — OAuth managed for you. Install from the dashboard Integrate tab. memory.rb Per-user, per-thread or per-org. We store it, back it up and expire it. You just say what to remember. knowledge/ Drop in PDFs, markdown, or point at a database. We chunk, embed and retrieve — no vector DB to run. evals/ Example conversations with the answers you expect. Every change runs through them before shipping. prompt.md The system prompt in plain markdown. Edit it in the dashboard, commit it in git — same file, same source of truth. Connect Google Sheets, Notion, HubSpot, Slack, and thousands more — OAuth handled, tools ready for your agent. No custom integrations to build. From Build → Packages, search Skills.sh, Smithery, and Microsoft APM — then attach packages into the agent folder your Rails app already syncs. Skills.sh Browse and attach skills from the Skills.sh registry into your agent. Smithery Discover Smithery packages and wire tools into Build → Packages. Microsoft APM Install Microsoft APM packages alongside your Rails agent files. Start from a real Rails use case — support, sales, finance, ops, and more. Clone a playbook to scaffold agent.rb, prompt.md, and metadata, then customize in Overview. Step 1 Browse Pick a playbook for your team Step 2 Get It Opens signup with the playbook loaded Step 3 Customize Agent files land in app/agents/ Slack, Teams, WhatsApp, web chat, email, Discord, Telegram, SMS, GitHub, Linear, cron, and HTTP — the same agent brain, delivered through OpenWire so you are not rebuilding OAuth and event plumbing for each surface. Web chat Browser chat + eve HTTP API Slack Branded bot in channels & DMs Discord Slash commands & components Microsoft Teams Adaptive Cards in Teams Telegram Telegram bot in one install Twilio SMS + Voice SMS and speech-transcribed calls GitHub PR reviews & issue triage Linear Issue delegation & Agent Sessions Meta Cloud / Chat SDK Connect a company mailbox Facebook Messenger Messenger via Chat SDK Google Chat Google Workspace Chat HTTP API JSON run endpoint Cron / schedules Scheduled agent runs Add the gem, open /agents , follow docs or clone a Playbook. Four steps — no AI course required. "Docs first, ship fast." — Rails Agent In your Gemfile: gem "rails-agent-stack", github: "Tiny-Bubble-Company/rails-agents" bundle install, then bin/rails generate rails agents:install. Mounts the engine and prepares credentials. bin/dev, then open /agents — like Sidekiq. Sign up, add a provider key, follow Dashboard Docs. Blank agent or Playbook clone scaffolds agent.rb, prompt.md, and metadata. Edit locally or with your coding agent. Durability, sandboxing, human-in-the-loop, and evals are built into the framework. Focus on building your agent. Workflows survive crashes and restarts. Every step is checkpointed. Agents park when waiting, resume on the next message. Reuse shared knowledge and skills across agents. Clone Playbooks into app/agents/ and customize in code or the dashboard. Monthly token budgets with admin alerts, plus policies for model access, prompt-injection detection, and sensitive-info redaction. Connect Slack and Teams via Open-Wire. Connect company mailboxes Gmail/Outlook on the Email channel for inbound and outbound. Search Skills.sh, Smithery, and Microsoft APM from Build → Packages and attach skills into the agent folder. New agents remember across conversations out of the box. Turn memory off per agent when you want each chat to start fresh. Tools that need confirmation trigger approval gates. Sessions park until resolved, then resume seamlessly. Define eval suites with scoring rubrics. Watch runs, tokens, and cost from Monitor — then tighten budgets when needed. RubyLLM and ActiveAgent are great libraries. But you still need AI knowledge, build the dashboard, and run the runtime yourself. Rails Agent ships the whole platform — no AI expertise required. Swipe to compare → | Feature | Rails Agentus | RubyLLM | ActiveAgent | DIY LangChain + glue | |---|---|---|---|---| | Setup | One gem, one CLI command | Wire clients manually | Wire mailer-style classes | Python + YAML + glue | | AI knowledge required | None — just Rails | You wire models & prompts | You wire models & prompts | Deep LLM & ops expertise | | Dashboard included | ✓ | — | — | — | | Playbooks + docs-first scaffold | ✓ | — | — | — | | Hosted runtime | Included, one-command deploy | Bring your own | Bring your own | You build & run it | | Traces, evals, monitoring | ✓ | — | — | Wire it yourself | | Budget + Guardrails | ✓ | — | — | DIY | | Memory & knowledge RAG | Built-in, on by default | Wire pgvector yourself | Wire it yourself | Wire it yourself | | Model neutral OpenAI, Claude, Gemini | ✓ | ✓ | ✓ | ✓ | | Human-in-the-loop approvals | ✓ | — | — | DIY | | Uses your Rails app as-is | ✓ | ✓ | ✓ | — | | Time to first agent in prod | An afternoon | A week | A week | A month+ | $99 Launch or $179 Scale per company — 30-day trial. BYOK model spend stays with your providers. Enterprise is custom via enquiry. For building and testing on the cloud sandbox. Production harness — BYOK models. Add Budget & Guardrails when you need them. Everything in Launch, plus Budget & Guardrails included. Need SSO, SLA, or custom terms? Submit an Enterprise enquiry /pricing enterprise . Add the gem, restart, open /agents . Follow Dashboard Docs or clone a Playbook. No AI course required.