@sleds/cli
v0.1.23
Published
SLEDS CLI — the cost and limits of AI agent work, decided at the wall. Point Claude Code at a SLEDS proxy; estimate before you run; hold, switch model, or run overnight when you hit a limit.
Maintainers
Readme
@sleds/cli
The cost and limits of AI agent work, decided at the wall.
SLEDS sits in front of Claude Code. It meters what you spend, sizes a job before it runs, and when you hit a limit — a rate limit, a weekly cap, exhausted credits — it gives you a choice right there in the terminal: continue on the org key, hold until the window resets, switch to a cheaper model, or run it overnight at half price.
Install
curl -fsSL https://sleds.ai/install | sh
sleds init --signupThe installer puts sleds on your PATH (via npm while the single-binary build is on its way — it needs Node 22.5+). npm i -g @sleds/cli or npx @sleds/cli init --signup work the same.
No org? --signup asks: just you, or your team. Just you is your email plus your own Anthropic API key. SLEDS checks the key, stores it encrypted, and it's the only key your seat ever spends. Your team is a work email and a team name — you become its owner; a platform key is optional (a Team plan without one still gets seat tracking, hold and hand-off at a wall). If your team already runs SLEDS, use the seat token your admin gave you instead:
npx @sleds/cli init --token sled_…init points Claude Code at the proxy, installs the /sled skill and the status-line meter, and registers two hooks: one that records a usage-limit moment, and one that sizes spec files as you write them.
The package is @sleds/cli; the command it installs is sleds. Requires Node 22.5+ (or Bun). Claude Code must be installed. seat-sync is macOS-only.
Every day
sleds status switch state, proxy health, wall mode
sleds meter built today · queued · delivered overnight · estimate accuracy
sleds on hit a limit on your seat? move Claude Code onto the org key
sleds off back to your seat
sleds defer hit a limit, no org key? this Mac resumes the session itself at the reset (the musher)
sleds musher the agent behind it: --status · --install · --uninstall
sleds run <jobId> run one deferred job now, by hand
sleds verify <id> was this job built with the model it claims? checks the org's signature locally
sleds seat-sync read your Claude subscription's own limits into SLEDS (macOS)
sleds doctor check the whole install: seat, Claude settings, proxy, hooks, Node, plugin, timer
sleds mcp MCP server: usage / estimate / overnight / jobs, for Cowork and other clientsInside Claude Code:
/sled estimate <what you're about to do> a band, a cap, and the price on each model
/sled 1 · 2 · 3 · 4 at a wall: continue · hold · overnight · switch model
/sled overnight <spec> queue a batch job with no wall
/sled jobs · /sled resume <id> what ran overnight, pulled into the sessionNothing changes until a wall.
Docs
Repo: github.com/spdyadam/sleds · Site: sleds.ai
