AI workflow automation for developers
Build it with an agent. Keep the parts that work.
Kody gives your agents a home for saved code, connected accounts, memory, and background work. Decide where each workflow needs judgment and where it just needs to run.
Build your first workflow ↗Tell me which releases matter to my project.
- Input
- GitHub releases + project context
- Saved progress
- Processed release IDs
- Code
Collect new releases
Read the selected repositories and skip release IDs already processed.
- Judgment
Explain the impact
Ask a model which changes affect the project, with links back to the source.
- Code
Format a preview
Build the digest and return it for review before enabling Slack posts.
Check the relevance and source links before posting.
Spend reasoning
where it matters.
You don’t need a model to fetch releases and format links. Figuring out which changes affect your project is a better use for it. Keep those steps separate so you can check the code and the AI’s answer.
A saved package can run from a job or webhook without a chat session. If your code calls a model, you’ll need to account for that usage and cost, too.
How jobs and workflows run ↗One useful output.
Then build from there.
Connect the required accounts, try a manual preview, and test failures and repeated runs before adding a schedule.
Get started with Kody ↗Copy this prompt to your agent
Help me turn one repeated task into a Kody package. First work out what it needs, what it should produce, which accounts it uses, and what needs my approval. Use code for predictable steps and explain where a model would help. Show me a preview, then help me test failures and repeated runs before putting it on a schedule.