Locaible is a desktop app that runs pre-tuned AI coding agents on your own machine and exposes them to Cursor through a local, OpenAI-compatible endpoint. Instead of sending your codebase to a hosted model, you point Cursor at localhost and, as the project puts it, your code never leaves your machine. For developers and creators building with AI who work under privacy or compliance constraints, it is a way to keep the agent workflow while dropping the data exposure.

What Happened

Locaible launched as a Show HN offering local agents that appear directly in Cursor's model dropdown. It ships pre-tuned specialists, including a security-aware code reviewer, a legal-focused agent called Atlas, and a tech-support agent backed by runbooks, alongside a marketplace and the option to build custom per-project agents loaded with your own style guides and docs. Setup is pitched at three minutes with either Cursor or Continue.dev.

Why It Matters

The privacy pitch is the differentiator. Cloud coding agents are powerful, but they route proprietary source through third-party servers, which is a non-starter for regulated or sensitive work. By serving models locally on an endpoint Cursor already understands, Locaible keeps the editor experience intact while removing the network hop. It also means zero per-token API billing, which changes the cost math for developers running agents all day.

Key Details

Under the hood, Locaible runs local models through an Ollama-style stack, supporting options from qwen3:8b up to deepseek-coder:33b depending on your hardware. Output quality therefore scales with your machine: a laptop will run the smaller models, while a workstation with a strong GPU unlocks the larger coders. The free tier covers three agents and one device, with a paid tier for more. Because it speaks the OpenAI API format, the same endpoint works with any tool that accepts a custom base URL, not just Cursor.

What to Do Next

Install Locaible, pick a model that fits your GPU, and point Cursor at the local endpoint to test it on a non-critical project first. Benchmark a real task against your current cloud agent so you know the quality trade-off before committing sensitive work. If reliability and guardrails matter more than privacy, weigh it against our coverage of Forge for local agent reliability and Cursor Composer 2.5.