Show HN: Compilr.dev, multi LLM AI workspace

Compilr.dev, built by Carmelo (HN: scozzola) over six-plus months, launched in beta as an ecosystem of tools for building and using AI agents. It spans TypeScript libraries (`compilr-dev/agents` and `compilr-dev/agents-coding`), a CLI, and a Desktop app that share a common core.

The Desktop app ships with pre-built specialized agents, each with a tailored system prompt, defined skills, and access to dedicated tools. When a project is created, a team of agents is assembled automatically; users can add, remove, or create custom agents at that point. Agents maintain individual context but also share project-level context, and can interact via direct inquiry or task handoff between agents.

The workspace uses a bring-your-own-key (BYOK) model, letting users choose and switch LLM providers freely. It also incorporates a spec-driven workflow — a `/design` command prompts an agent to capture project scope, generate work items in a database, and produce a PRD document.

Token consumption is a stated design priority. Techniques include loading tool and skill definitions only when needed, pruning or summarizing tool outputs after several rounds, and spawning subagents to summarize large files rather than passing them in full to the main context.

The project is seeking beta testers for real-world feedback, per the Hacker News post.