Onyx Code: local AI code editor focused on data privacy
Experience local AI assistance that runs on your machine, offering code completions, planning, and automated edits without mandatory cloud access. Onyx Code, from Loko Technologies Ltd., installs on Windows as a utilities tool built on the VS Code OSS foundation. It pairs a bundled inference engine and an Air-Gap Mode with file-aware agent tools and a reviewable Smart Apply workflow. The app suits developers and organizations that require AI help while keeping source code and prompts strictly on-device.
What does Onyx do for developer workflows?
Onyx extends the VS Code OSS editing environment with a local AI inference engine (llama.cpp) to provide completions, planning, and automated edits. The agent suite includes Plan, Ask, Edit-auto, and Auto modes for task automation, while Smart Apply shows generated changes as a diff before they merge into files. Support for Model Context Protocol servers lets the app interact with local filesystems, git repositories, and databases for context-aware suggestions.
How does it affect system resources during AI work?
The bundled inference executes on the user's hardware, so CPU and memory use depend on local model runs rather than remote servers. Because the engine runs offline, inference compute occurs locally and the app can operate without external dependencies. Background operations remain limited to local processing; optional cloud APIs are separate and not required for core offline assistance, which means resource load is confined to the desktop where the models run.
Is it safe for sensitive codebases?
Security features target high-assurance environments: Air-Gap Mode disables outbound network requests, and the app keeps code on the local machine so no prompts or completions are sent externally by default. An integrated prompt sanitizer removes sensitive data before optional cloud API use. These elements create a design that prevents transmission of source code outside the desktop unless the user enables external integrations.
Do non-technical teams need specialist skills to operate it?
Because the interface builds on VS Code OSS, familiar extension support through Open VSX reduces learning overhead for developers already used to that editor. Advanced agent modes and MCP server use require configuration and an understanding of local model hosting, whereas basic editing and offline completions work within standard editor workflows. The app therefore serves technical users directly while offering accessible editing for routine tasks.
Clear fit for privacy-focused development, with one operational trade-off
Onyx suits security-conscious engineers and teams that require AI assistance without exposing code offsite; its local inference and Air-Gap Mode support that aim. The main trade-off is dependence on local compute for model inference and any initial Pro activation that requires an internet connection, which means organisations must provide adequate hardware and a controlled activation step before full offline use.
