Output.ai is live 🎉🚀

Marcos Marx — Mon, 13 Apr 2026 20:38:03 +0000

We just open-sourced Output: a TypeScript framework for building AI workflows and agents.

It came from our real work at GrowthX. We have two tracks: our product team is always building features that are AI-heavy, and our forward-deployed team is dedicated to building growth-focused AI agents for dozens of companies like Lovable, Airbyte, Augment Code, and Webflow.

Every project hit the same walls: scattered prompts, no way to test non-deterministic code, zero cost visibility, retry logic, how to write evaluation code, and dozens of dependencies on startups that keep getting acquired, pivoting, or shutting down.

So we built the framework we wished existed - as feature complete as we needed at GrowthX. And now it's yours, open source under Apache 2.

What it does? Output gives you everything production AI needs in one place:

Prompts as code .prompt files in your repo. Version-controlled, reviewable in PRs. No dashboards.
Evaluation-driven development — LLM-as-judge evaluators with confidence scores. Test prompt changes against dozens of scenarios before you ship.
Full tracing & cost tracking — Every LLM call, every step, every dollar. On your disk or S3. No third-party observability vendor.
Durable execution — Temporal.io under the hood. Retries, replays, workflow history. You write async/await, we handle the rest.
Encrypted credentials — AES-256, scoped per environment and workflow. No external vault subscription.

Built for Claude Code. Every convention is designed so your coding agent understands the full context. Describe what you want, Claude builds it.

Getting started npx @outputai/cli init your-project

That's it. One command.

You can check the project https://github.com/growthxai/output, if you like it star us on Github.

Forem: GrowthX.ai

Output.ai is live 🎉🚀