Kody weekly: Schedule code with one jobs system

Here is what shipped in Kody from April 11 to April 17, 2026. This week your code can run on a schedule, keep its own storage, and live in repos, and saved apps get real backends.

Schedule code with one jobs system

What shipped

Kody now has jobs. The job_upsert, job_list, job_get, job_delete, and job_run_now capabilities create and manage code that runs on a schedule. A schedule is a cron expression, an interval such as 15m or 1d, or a once run at an ISO timestamp, and cron schedules can use an IANA timezone. Each job keeps its run counts, last error, last duration, and a short run history, and you can disable a job or flip its kill switch without deleting it. Jobs run through the same codemode runtime as execute and get a stable storageId for durable data.

Why it matters

Recurring work like a morning digest or a weekly cleanup no longer depends on you remembering to ask. Your agent writes the code once, schedules it, and you can see whether it ran and why it failed. It is the first step from one-off answers to automations you own.

Example

Call job_upsert from execute with a name, code, and schedule:

{
	"name": "weekday-digest",
	"code": "async () => { /* gather and summarize */ return 'ok' }",
	"schedule": { "type": "cron", "expression": "0 8 * * 1-5" },
	"timezone": "America/Denver",
	"params": { "topic": "open pull requests" }
}

Use job_run_now with the returned id to test it right away, and job_get to read its run history.

PRs: #157, #170, #171

More updates

• Execute: Adds durable storage. execute accepts a storageId and writable, code can use the built-in storage.* helpers and storage.sql(...), and storage_query and storage_export inspect a store read-only by default. (#171)
• Apps: Gives each saved app an optional server backend. App code splits into client and server parts, the server part runs as its own Durable Object with an isolated SQLite database, and the client reaches it at /app/:appId/*. (#162, #166)
• Packages: Adds repo-backed sources. Apps, skills, and jobs can live in repos, and repo_* capabilities open sessions, read and search files, apply edits, run checks, and publish. repo_backfill_sources moves existing inline apps, skills, and jobs into repos without overwriting published sources. (#181, #187)
• Execute: Supports returning raw MCP content blocks, such as images, from execute through __mcpContent instead of text only. (#149, #151)
• Execute: Adds agent turn capabilities: agent_chat_turn for one bounded tool-using turn, and agent_turn_start, agent_turn_next, and agent_turn_cancel to stream one. (#180)
• Memory: Supports optional source URIs on memories, and search uses your query as memory context when you leave memoryContext out. (docs, #168, #154)
• Platform: Generalizes the home connector into remote connectors, with separate sessions, routing, and secrets per connector, plus connector status in search. (#155)
• Orgs: Adds a reset button on the authorize screen when a redirect URI mismatch leaves an MCP client stuck, so you can remove the stale client and sign in again. (#161, #169)

Fixes

• Apps: Fixes ui_save_app partial updates wiping fields you did not change. (#165)
• Execute: Fixes read-only storage SQL rejecting queries with semicolons inside strings or comments. (#173)
• Automations: Fixes jobs reusing stale repo sessions. They now fail clearly instead. (#184, #191)

Try it

Ask your agent to schedule one small recurring task with job_upsert, then run it once with job_run_now and check the result with job_get.

Questions or feedback? Join us on the Kody Discord.