26.8.8 Version Release 🚀

Mogick CLI is an AI agent that runs in your local terminal. This first public preview brings together the complete workflow: understand a project, make a plan, apply changes, verify the result, then review or recover.

What you can do

• Take a task through to verified code changes: The agent can understand the project structure, search the codebase, edit multiple files, and run Shell and Git commands. After running tests, builds, or linters, it can use failures to continue diagnosing and fixing the problem.
• Review the plan before anything changes: Plan mode is strictly read-only and focuses on exploring the project and producing an execution plan. Once you approve the plan, switch to Agent mode and use the four-level permission policy to control file writes, command execution, and external tool calls.
• See what changed and review the result independently: Use /changes to inspect the current session's Diff file by file, then run /review for an independent review that does not modify the code. You can also export session history, context usage, and Git evidence for retrospectives or team handoffs.
• Experiment safely and keep long-running work moving: Before writing a file, Mogick automatically saves the previous state of the affected file. Use /rewind to roll back files and conversation history, resume work after an interruption, and continue long conversations as older context is compacted while the complete history remains available locally.
• Carry project conventions across sessions: Long-term memory automatically retrieves and retains coding preferences, project conventions, and key decisions, reducing the need to repeat context in later tasks. Use /memory to turn it off at any time when working with sensitive projects.
• Use the same agent in the terminal, scripts, and host applications: mogick provides an interactive TUI, mogick run supports scripts and CI, and mogick acp embeds the runtime in IDEs, desktop applications, or other automation systems.
• Connect your own models, rules, and toolchain: Configure and switch between multiple model providers, reuse team workflows through Skills, and connect enterprise systems or specialized tools through MCP. Experimental subagents can run independent exploration and review tasks in parallel.

Guardrails

• The default auto permission policy automatically allows safe operations and non-delete writes inside the workspace. Other operations may require confirmation based on risk, and delete operations always require confirmation unless the policy is yolo.
• Prompts, code excerpts needed for the task, and tool results are sent to the configured model service. Session history, tool calls, and permission decisions are stored locally by default.
• Checkpoints cover only files the agent actually modifies; they are not a backup of the entire workspace. Continue to use Git branches or another backup mechanism for important projects.