Solix FAQ

No. solix demo seeds a fake galaxy with planets, comet streaks, and a pending permission so you can explore the UI without Claude Code installed. You only need Claude Code to watch your real agents.

Solix works with any recent Claude Code release. It syncs through Claude Code's Agent View / background-session bridge, so there is no extra config — start a session and a planet appears.

You don't — the first run does it. Starting Solix patches your Claude Code hooks itself, after backing up ~/.claude/settings.json, and prints what it changed. So the whole install is:Then open a Claude Code session and watch the planet appear:solix install still exists for scripted setups, and solix uninstall removes only Solix's own hook entries.

Run solix doctor — it checks that the hooks are wired, the port (127.0.0.1:4242) is free, and Node is ≥ 20, and reports whether the installed hooks match the running build. The most common fix is solix install --force after a Claude Code update overwrote the hooks. Note that Solix only sees sessions started after the hooks were wired, so restart any session that was already running.

Yes — that is the whole point. Each running session becomes its own planet orbiting the sun (you). Run three claude sessions in different terminals and you will see three planets, each with its own model color and context size.

One running Claude Code session. Its color shows the model (opus purple, sonnet blue, haiku cyan) and its size reflects how much context it has consumed.

A subagent spawned by Claude Code's Task tool. Moons orbit their parent planet, so you can see when a session has delegated work and how busy each subagent is.

A task Solix launched is awaiting your permission for a sensitive action — it has paused and needs a yes/no. The count sits on Decisions in the left rail — open it to read the exact command or diff, then press Y or N right from the browser — no alt-tabbing back to the terminal. The keys deliberately only work while that panel is open, so you cannot approve something you have not read.

Blocking — for the tasks Solix runs. It registers a Claude Codehook that fires before the tool call executes, and for an agent Solix launched that hook holds the write or shell command in place until you answer. Nothing is approved by default and nothing is merely logged after the fact. If Solix becomes unreachable mid-decision the call is denied, not waved through.

A claude session you started in your own terminal is a different case: Solix watches it and costs it, but Claude Code's own permission prompt stays in charge there. Solix does not interpose itself on a process it did not start.

Give the task a branch when you launch it and Solix runs it in a dedicated git worktree under ~/.solix/worktrees, so several agents can work in parallel without touching each other's files or the branch you have checked out. You review the result like any other branch. Leave the branch field empty and the task runs in the project directory as usual.

The cap is per task. As live spend approaches it the budget ring turns amber and Solix warns you; at the ceiling the run is paused instead of continuing. You can raise the cap and resume, or stop the agent there.

Yes. Solix wraps the session two-way rather than tailing its output, so you can send input into a live Claude Code session from the galaxy view — answer a prompt, redirect it, or stop it.

Each comet is a tool call firing in real time — a read, a write, a search. Streaks let you feel the tempo of an agent without reading its log.

Set a cap on an agent and its planet wears a ring showing live USD spend estimated from token usage, shifting sky → amber → red as it approaches the ceiling. Planets without a cap have no ring.

What happens at the ceiling depends on what is running. A Maestro plan warns once on the way up, then pauses before the spend crosses the cap, and you resume it when you're ready. A session Solix launched stops accepting new work. A session you started in your own terminal gets the ring and an alert — Solix gates the commands that session tries to run, but it cannot halt a process it didn't spawn.

The planet stops orbiting and fades, but it is not gone — the Timeline lets you scrub back through everything that agent did. You can replay tool calls, approvals, and cost over the session's whole life.

Recurring work — like a nightly security scan from Sentinel — with its own visible rhythm in the galaxy. A routine runs on an interval, when an inbound webhook fires, or only when you trigger it, and it keeps a record of its past runs.

Each one tells you what it will cost before you enable it: an interval routine starts a real agent session and spends money on every run, until you disable it. That is easy to miss in a one-click install, so Solix says it in plain words rather than burying it.

No. Solix runs entirely on 127.0.0.1. Your code, prompts, and file contents never leave your machine — the dashboard only reads local state from Claude Code's session files.

Zero. There is no phone-home, no usage tracking, no account, and no cloud sync. That is also why a “live agents online” counter would be impossible — Solix has no way to know what other machines are doing, and that is the point.

Solix never touches your Anthropic API key. Authentication to the model happens inside Claude Code itself; Solix only observes the session state that Claude Code already writes locally.

No. It binds to 127.0.0.1:4242 only — localhost, not 0.0.0.0. Nothing on your LAN or the internet can reach it unless you explicitly tunnel it yourself.

Yes — MIT licensed, source on GitHub. You can audit every line before you run it, and the hooks Solix wires on first run are plain text you can read and revert.