In just the last couple days I've seen a lot of people complaining about the same category of problems:

I expect many of you can relate to this pain. In fact, I tested it:

Most of you have more than one agent you're using weekly

The fact is that while these agents continue to get better all the time, there is some stiff competition and we find ourselves picking one over the other in a variety of situations.

It's not like the choice between an Android and iPhone that you make once and stick with for years.

It's more like what outfit you decide to wear based on the activities of the day.

But the Android/iPhone analogy is still relevant here because each of these companies is doing their best to build a walled garden where you'll be happy so long as you never leave ("Aye Calypso" or "Hotel California" style).

Yes, you can set up automations and integrations in Codex, but if you decide you want to try out Fable 5.1 or Grok Bot you've got to set all that stuff up again. And keeping track of where you're doing that is a pain.

Oh, and speaking of integrations...

Your .env file is not safe and you know it...

If these problems sound relatable to you, I have a solution that I've been building for the last several months and I think it's really unique. It solves the walled garden problem, makes it impossible for agents to get at your secrets or fall prey to prompt injection and send them somewhere you don't approve, and saves you tokens with deterministic software that runs in an isolated sandbox.

That solution is Kody.

Kody is a sandboxed runtime in the cloud where your agents can author and execute code. You can integrate it with literally anything that has an API, has built-in and generic OAuth support, and can connect to MCP servers. Then Kody makes all of that available in the Kody runtime and your agents can author ad hoc code to interact with those integrations.

Once you're happy with it, the agent turns that ad hoc code into a durable Kody-hosted repository and package so any of your agents can reference it in the future. Packages also have their own SQLite database, scheduled jobs, webhooks, or even full blown web apps for you to interact with those integrations. This makes it run faster, cheaper, and more reliably than any SKILL.md file you could come up with.

And then if your package is really good, you can publish it to the public community of packages so anyone can find it and fork it into their own personal software factory and adapt it to their own needs. There are already hundreds of packages for interfacing with services you use.

And Kody's secret storage means that you can safely store encrypted secrets (that are in a completely isolated database of your own) and agents don't have the ability to see those secrets even if they wanted to. They can use them to interact with services and your packages, but that's all. Safer than your .env.

I've been using Kody like crazy over the last several months as I've been building it, and I've given early access to some of you as well to validate the product as well as improve its features, positioning, and framing. I created an MCP server for my home so Kody can control my lights/blinds/TVs/speakers/etc. Kody's become an integral part of my software development process (or "factory" if you want to call it that) with the integrations with Stripe, Cursor, Cloudflare, Sentry, Kit, Fathom, etc. And not having to set up automations, memories, and integrations in every agent every time has been really nice.

I spent the last several months on a bet that you'd think it's really nice too and I'm excited to share it with you to try it out for FREE. I hope it helps you the way it helps me.

So what are you waiting for? The doors are finally open! Go connect your agents to Kody!

https://kody.codes

P.S. Got questions? Just ask your agent "What is kody.codes?". Got problems once you're connected? Just ask your agent to submit feedback. My kody kicks off an agent to triage the feedback automatically.