A recurring orchestration problem.
Preparing inputs, running parallel agents, following failures, and saving responses are useful across many applications. The project separates that common machinery from any particular domain or database schema.
BatchDesk is a public introduction to codex_agenting: a reusable framework for coordinating Codex CLI agents on independent, database-backed tasks.
Preparing inputs, running parallel agents, following failures, and saving responses are useful across many applications. The project separates that common machinery from any particular domain or database schema.
The framework owns tasks, runs, attempts, and saved responses. Application developers own their data adapters, prompts, output rules, and export policies. A new integration should not require changes to the core.
Agent work is performed through Codex CLI subprocesses. The framework adds local coordination, bounded execution, a service queue, explicit retries, and an operator dashboard. It is not a hosted model service.
A first implementation and an offline test suite exist. A limited ten-task Windows pilot is documented for one CLI/account setup. Linux execution, broader CLI compatibility, and advanced delivery guarantees still need separate work.
BatchDesk is an independent presentation of this framework. It is not an official OpenAI product and does not imply endorsement. Codex CLI is the external agent runtime used by the project.
The source download and GitHub repository will be available soon. Until then, the documentation preview and synthetic demo explain the workflow.