Developer forks Continue extension to create a lightweight, telemetry-free local autocomplete tool
A developer has released a fork of the Continue VS Code extension called "FIM-Autocomplete," stripped of all agentic features, chat sidebars, and remote telemetry. The tool focuses purely on inline code completion using local or hosted Large Language Models (LLMs) without requiring a subscription. This project offers a privacy-centric, lightweight alternative for developers who are frustrated by the bloat, subscriptions, and corporate acquisitions of mainstream AI coding assistants. It allows developers to maintain full control over their code and data by running completion models entirely offline. The extension supports various backends like Ollama, llama.cpp, and LM Studio, automatically selecting the correct Fill-in-the-Middle (FIM) prompt format for over 16 model families. It also utilizes cross-file context, including tree-sitter scopes and LSP definitions, to improve completion relevance while keeping telemetry strictly local and in-memory.
## BACKGROUND
Continue is a popular open-source AI coding assistant that recently joined Cursor (Anysphere), which was subsequently acquired by SpaceX. While traditional AI assistants have evolved into complex "agentic" systems that can rewrite repositories or chat, many developers prefer simple "ghost text" autocomplete that predicts the next line of code. Fill-in-the-Middle (FIM) is a prompting technique where a model fills in code between existing prefix and suffix contexts.