Cloi

Cloi is a terminal coding assistant for developers who want an agent to work on their code using models on their own machine. It can inspect a project, make edits and run commands without sending workspace data to a cloud service.

• Model selection and fallback: Cloi checks your RAM and GPU memory to choose a primary model and a larger fallback, then hands a stalled task to the fallback with the conversation intact.
• Workspace checks: It checks file claims, quoted text and line references against what it has read, with an additional model review after a fallback takes over.
• Reusable results and Python: Saved file reads, searches and command results remain available across conversations, and a persistent Python interpreter can use them for calculations and work across files.
• Conversation continuity: You can resume sessions, while summaries keep older work in context as conversations grow.
• Permission controls: File access stays within the workspace, and writes, edits and shell commands ask for approval by default; Cloi also filters subprocess environments and redacts secrets in output.

Cloi is open source under the MIT license. It requires Node.js, Ollama and a model that supports tool use; Ollama is its only model backend. It has no web access, MCP support or sub-agents. Hardware limits matter: its benchmarks found that models fitting within 8 GB couldn't complete the difficult cross-file debugging tasks tested.