Viscacha: A crash-safe, zero infrastructure job system for Python functions and AI pipelines.

The developer ecosystem for AI applications often necessitates robust asynchronous processing. While battle-tested tools like Celery and specialized message queues like Redis/RabbitMQ provide powerful job management, they introduce significant operational overhead, requiring external infrastructure setup, monitoring, and scaling effort. Viscacha directly addresses this common friction point. It positions itself as a minimal, Python-native solution for background job management, eliminating the dependency burden associated with external brokers. Technically, Viscacha eschews complex message passing systems for a simpler, yet highly effective, 'tuple space' model. Jobs are persisted into a simple, append-only log, treating the queue as a durable, single source of truth. Workers claim tasks using leases: they read a job, work on it, and release it. If a worker fails or disconnects, the lease expires, and the job automatically returns to the queue. This mechanism provides robust crash recovery, a feature that is notoriously difficult to implement reliably in pure application code. Its architecture is particularly well-suited for modern AI workflows. Instead of managing thousands of messages, a developer defines a Python function decorated with `@worker.job()`. This function encapsulates the entire business logic—whether it’s an HTTP call, an email send, or an external API call like Anthropic's Claude—and the framework handles the queuing, execution, and state management. This drastically simplifies the development cycle for complex data or LLM pipelines, allowing data scientists to focus purely on the logic, not the operational plumbing. For teams evaluating job queues, the choice often boils down to complexity vs. power. While a dedicated message broker offers infinite scalability potential, Viscacha offers an unmatched development velocity and operational simplicity for most common use cases. By restricting job coordination to Python and local file logging, it dramatically lowers the bar for entry, making reliable, distributed job execution accessible to smaller teams and individual developers alike.