The Agent Integration Problem
Developers are increasingly working with multiple AI coding agents, often switching between them depending on the task. But this workflow comes with friction: each agent has its own set of automations, integrations, and memory that have to be configured separately. Trying out a new agent means starting that setup from scratch.
There's also a security angle. The environment variables and API keys stored in a project's .env file are exposed to whatever agent happens to be reading that codebase. And each agent platform is effectively a walled garden — useful tools built for one don't carry over to another.
What Kody Is
Kody is a cloud-hosted, sandboxed runtime designed to sit between your agents and the services they need to interact with. Agents can author and execute code inside Kody, which has built-in and generic OAuth support and can connect to MCP servers. That means Kody acts as an integration layer for anything with an API, without needing to wire those connections into each agent individually.
Once an agent figures out a workflow — say, talking to a particular service — the ad hoc code it wrote in the Kody runtime can be converted into a durable, Kody-hosted repository and package. Any agent you use can then reference that package later. Packages come with their own SQLite database, scheduled jobs, webhooks, and the ability to host full web apps. Because they run as deterministic software in an isolated sandbox, they execute faster and cheaper than relying on something like a SKILL.md file in a prompt.
Packages can also be published to a public community directory. Other developers can find and fork packages into their own "software factory" and adapt them, so patterns for common integrations are shared rather than re-invented.
Keeping Secrets Out of Reach
A core part of the design is secret management. Kody stores encrypted secrets in a dedicated and isolated database per user. Agents can use those secrets to interact with services and packages, but they never have the ability to read the raw secret values. Kody's runtime also prevents agents from sending secrets anywhere you haven't approved, which is a hedge against both accidental leaks and prompt-injection attacks.
In practice, the tooling includes an MCP server for home automation (controlling lights, blinds, TVs, speakers) and integrates with services like Stripe, Cursor, Cloudflare, Sentry, Kit, and Fathom. The value proposition is that you set up these connections once in Kody, and every agent you use gets access to them without duplicating configuration.
Availability and Getting Started
Kody is free to try and openly available at https://kody.codes. If you have questions about the product, one option is to ask your agent "What is kody.codes?" Ingesting the documentation is itself a demonstration of how the platform is meant to be used.
