Candle – a helpful tool for running local dev services
- Source
- andyfischer.dev
- Author
- apf6
- Date
Candle is a small tool aimed at a problem that grows with parallel agent work. When several coding agents run dev servers across git worktrees, they tend to collide on shared processes and ports, and nobody can easily tell which log belongs to which checkout. Candle runs local services in the background and scopes them to the project directory they were started from, so each worktree gets its own set. Logs go into SQLite, which means any terminal or agent session can query them rather than tailing a file that another process owns. The author also ships an AGENTS.md prompt that tells coding agents how to start, stop and inspect services with it. That last detail matters more than it looks: agents tend to start a server in the foreground and then stall, and a documented background workflow gives them a clean alternative. It is an early, single-author project, so treat it as an option to try rather than settled infrastructure.
Context
Candle is a process manager for local development: a tool that starts, stops and logs the background services a project needs while you work. Its author, Andy Fischer, describes it as comparable to systemd, pm2 or supervisor but specifically designed for local development. A service launched through it can outlive the terminal or session that started it, and the tool keeps process state and logs in a single SQLite database under the user's ~/.local/state directory.
Candle's commands are scoped to the current project directory, so listing or starting processes affects only that project's services. For coding agents, the post says Candle refuses the interactive candle watch log stream when an calls it, prints an error and points to candle logs instead, and it suggests adding a note to AGENTS.md so an agent knows to use the tool. The post states two limits: Candle runs on Mac and Linux, and it does not help with unique port assignment. These are the author's own description of the tool, and the post reports no benchmarks or user results.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
Checking sign-in…
Loading comments…


