Looking Glass is a local AI coding CLI that enables persistent, automatable sessions where a scheduler can trigger future AI turns, allowing automation of workflows like deployment checks, test fixes, and recurring project reviews.
Looking Glass is a local AI coding CLI built around persistent, automatable sessions. The key feature is that a session can keep existing after you close the terminal, and a scheduler can later trigger new AI turns inside that same session. Those future turns reopen the original workspace, retain the session configuration, inspect files or logs that changed in the meantime, use the CLI tools, run commands, modify code, and write the results back into the session. In practice, this lets you automate workflows such as: Ask the model to deploy or start a long-running task, then have the same session check the result later. Schedule a follow-up turn to inspect test output and fix failures automatically. Run recurring project reviews that read the current codebase, execute tests, and report or resolve issues. Continue debugging after new logs or external processes have produced more information. Leave a session working through scheduled turns even when the interactive terminal is closed. Combine exact scheduled shell commands with context-aware AI follow-ups. Scheduled prompts are not separate stateless jobs or fresh chats. They are future turns attached to an existing session, with access to its: Workspace Previous conversation and tool history Primary model and reasoning settings Permissions and remembered approvals Agent configuration Scheduler results and persistent artifacts Looking Glass also supports concurrent worker agents. The agent model and reasoning level are configured separately from the main model, allowing a stronger model to coordinate while faster or cheaper models handle parallel discovery, implementation, or review tasks. Additional functionality includes: Interactive TUI and one-shot CLI prompts OpenAI-compatible local or hosted gateways File reading, search, patching, and bounded Bash execution Persistent approval modes for interactive and automated turns Scheduled AI prompts, reminders, and deterministic commands SQLite-backed session, scheduler, and artifact state A user-level systemd scheduler that runs independently of the terminal Typical setup: start a coding session, give it a task, attach one-time or recurring follow-up prompts, close the terminal, and let the CLI continue that same workflow later.
A reflection on how early AI chat interfaces resemble command-line interaction patterns, and a speculation about when a GUI-like paradigm shift will emerge for AI interactions, where the AI can directly observe and act on the user's context.
agentglass is an open-source local dashboard that provides real-time monitoring, cost tracking, and fleet management for multiple AI coding agents like Claude Code, Codex, and Gemini.
CLI-Anything is an open-source framework that automatically generates command-line interfaces for any software, making them accessible to AI agents. It includes a hub for community-built CLIs and supports multiple AI agent platforms.
Lanes is a native macOS desktop application that serves as a mission control for managing multiple AI coding agents, featuring an issue board, live embedded terminals, and Git integration to streamline developer workflows.
The article explores the idea of an open-source layer to orchestrate CLI usage by AI agents, addressing challenges like permissions, sandboxing, and audit trails when agents interact with multiple CLIs.