Show HN: Run Claude Code unattended inside a fence, under a dollar budget (github.com)

SIwithin has launched a command-line interface (CLI) tool that allows users to run Claude Code or OpenAI Codex unattended within a controlled environment, ensuring execution up to a specified budget of just one dollar. This innovation is significant for the AI/ML community as it enhances the safety of running AI models by confining them to specific directories while preventing unauthorized access to the user’s home folder. It operates on macOS and utilizes Python 3.10 or later, requiring no additional network calls, thereby alleviating privacy concerns. The tool incorporates several key features: it restricts spending through a per-run budget cap, includes a timer for each run, and maintains a record of costs and successes. SIwithin employs a loopback proxy for the Claude adapter, so that API interaction is metered and secure, shielding the user's API keys. The CLI cannot run on unsupported operating systems like Linux or Windows, emphasizing its dedicated focus. With features designed to monitor and limit resource usage closely, SIwithin represents a purposeful shift towards more accountable and risk-managed AI model deployment.