Currently in private beta
The harness already composed for Rails.
Fragua is an AI orchestrator where a Rails developer turns an idea into a shipped application — research, plan, spec, and ship, with agents that respect the craft.
Four phases. One workspace. Zero context loss.
From a one-line product idea to a merged pull request. Each phase produces durable artifacts that become context for everything that follows.
Tell Fragua a one-line idea. It studies similar apps and writes a synthesis you can edit and approve.
From the research, Fragua drafts the MVP plan that defines what to build first. Brand guide and technical guide ship as separate phases.
Define features as structured specs. Each one decomposes into a task list ready for an agent to execute.
Agents implement tasks on a feature branch and open a pull request. Watch the work happen live.
Every artifact persists. Every later phase has access to every earlier one.
Generic agents don't know your codebase. Fragua does.
Fragua's agents ship pre-loaded with the Rails 8.1 vocabulary. Hotwire for the front end. Solid Queue for async work. Rich domain models with concerns. The One Person Framework philosophy from Rails as the default — building like one developer can ship the whole thing.
When the agent reads your code, it knows what a Stimulus controller is. When it writes a test, it writes Minitest with fixtures. When it opens a pull request, it follows the conventions your team already lives by.
- Built-in awareness of maquina_components for views
- Rails MCP Server integration for runtime context
- Per-feature git worktrees for parallel work
- Per-Account GitHub App with short-lived tokens
Watch the work happen.
The durable events from each agent turn land in the database one row at a time — assistant text, mid-interview questions, terminal artifacts. The browser sees a live timeline as the agent produces it. Token usage and cost are written per turn into their own table. Cancel signals the runner and takes effect within five seconds.
Turbo Streams broadcasts each turn as the agent produces it. The browser never asks; the server pushes.
The database keeps what matters — assistant text, questions, terminal artifacts, per-turn cost. Tool calls and intermediate state stay in the runner stream.
Close the browser. Come back tomorrow. The run is still going. The event log is still complete.
You bring your AI key. We bring the receipts.
Fragua never proxies your AI calls. Every token your agents consume is billed directly to the provider your key belongs to. We can't see your billing or charge you for usage.
What we do is parse the usage events from every run and roll them up — per AgentRun, per Workspace, per Account, per month. Finance knows where every dollar went.
At launch Fragua runs on Anthropic Claude via Claude Code. Other AI services and platforms plug in as they ship.
Rails 8.1. SQLite. Solid Queue. Hotwire.
Fragua is built on the Maquina stack — vanilla Rails with the One Person Framework philosophy. The same conventions you'd want in a project Fragua builds for you.
The framework, modern and complete.
Five embedded databases, one server.
Database-backed async jobs.
Turbo + Stimulus, server-driven UI.
Passwordless sign-in via email codes.
Per-Account, short-lived tokens.
How we deploy Fragua to our infrastructure.
Continuous backup with ~1s RPO.
Flat per-account. BYOK. No surprises.
You pay Fragua a flat monthly fee for the platform. You pay your AI provider for tokens. We never sit between the two.
For independent Rails developers and personal projects.
- 1 user · unlimited workspaces
- BYOK — your AI spend, billed to you
- Full cost dashboard
- GitHub App, per-feature worktrees
For small consultancies and product teams.
- Up to 10 users · unlimited workspaces
- BYOK — your AI spend, billed to you
- Three-role authorization (admin / member / viewer)
- Soft budget caps with email · Slack alerts
- Priority access to new features
Currently free during private beta · Pricing begins 30 days after public launch with email notice
Tell us what you're building.
Fragua is in private beta with a small group. We're not opening seats on a schedule — join the waitlist and we'll reach out when there's room for the work you're describing.