moooon
Motir
Vibe the project. Bring an idea — Motir's three AI layers plan it, track it, and ship it, end to end. You're looking at Motir, built in Motir.
- Vibe Project
- Open Source
- AI Agent
- AI Loop
- 6
- 需求
- 4
- 贊同
- 534
- 已規劃
- 4,982
- 已發布
- The graph-routing architecture guard matches comment text, so prose that mentions vms.<app> fails it on main
- The fleet runner image has no pipx, so motir-ai's three Vitest shards die at `pipx install lizard` the moment CI moves onto the fleet
- The "Run in my agent" refusals acceptance spec times out in the merge queue, holding every PR behind it
- The graph service scales itself: it adds machines on memory against budget, read p95 and download backlog, removes them after sustained low use while keeping at least 2, and its refresh notices no longer time outPlanning capacity follows demand — a conversation turn never waits behind a planning run, the planning machines and the graph service grow and shrink on their own, and a load-tested cost ceiling is the only fixed limit
- (motir-codegraph) Stage the machine-management token on the graph app — a Fly token scoped to motir-codegraph, set as the secret CODEGRAPH_FLY_API_TOKEN, so the service can create, signal and destroy its own machinesPlanning capacity follows demand — a conversation turn never waits behind a planning run, the planning machines and the graph service grow and shrink on their own, and a load-tested cost ceiling is the only fixed limit
- (motir-ai) End-to-end across a simulated scale-out, deploy and drain — reads issued throughout are all answered or answered warming, a brief spike adds nothing, the floor of 2 holds and the count survives the deployPlanning capacity follows demand — a conversation turn never waits behind a planning run, the planning machines and the graph service grow and shrink on their own, and a load-tested cost ceiling is the only fixed limit
- (motir-ai) Story integration gate — coverage floor over the signal, settings, engine, actuation and deploy files, and the seams between them against real Postgres and a fake Machines APIPlanning capacity follows demand — a conversation turn never waits behind a planning run, the planning machines and the graph service grow and shrink on their own, and a load-tested cost ceiling is the only fixed limit
- (motir-ai) deploy-codegraph keeps the machine count — it updates every running machine, created ones included, to the new release and never resets the count to 2Planning capacity follows demand — a conversation turn never waits behind a planning run, the planning machines and the graph service grow and shrink on their own, and a load-tested cost ceiling is the only fixed limit
- Scale-out and scale-in against the Fly Machines API — create a machine with the same image, size and process group, and remove one by drain: stop it with SIGTERM, let in-flight reads finish inside the 60 s kill_timeout, destroy it only once it has left vms.motir-codegraph.internal, so a move never fails a readPlanning capacity follows demand — a conversation turn never waits behind a planning run, the planning machines and the graph service grow and shrink on their own, and a load-tested cost ceiling is the only fixed limit
- (motir-ai) The scaling decision engine — a signal past its threshold for its whole window adds, a brief spike adds nothing, one scale-out at a time, sustained low removes one, never below 2, and every decision is logged with signal, value, threshold, window and resulting countPlanning capacity follows demand — a conversation turn never waits behind a planning run, the planning machines and the graph service grow and shrink on their own, and a load-tested cost ceiling is the only fixed limit
- (motir-ai) Scaling settings read from config — per-signal thresholds, windows, sustained-low window, floor of 2 and an optional ceiling, as the decision's placeholders, with no hard-coded capPlanning capacity follows demand — a conversation turn never waits behind a planning run, the planning machines and the graph service grow and shrink on their own, and a load-tested cost ceiling is the only fixed limit
- (motir-ai) Each graph machine exposes its scale signals — memory use against budget, read p95 and the download backlog (the depth of the hydrate queue behind the 2-download limit)Planning capacity follows demand — a conversation turn never waits behind a planning run, the planning machines and the graph service grow and shrink on their own, and a load-tested cost ceiling is the only fixed limit
- (motir-ai) Fix the notice timeout at its own layer, with a regression test that fails before the fix — the notice stays fire-and-forget, replicas-only and never slows the pointer writePlanning capacity follows demand — a conversation turn never waits behind a planning run, the planning machines and the graph service grow and shrink on their own, and a load-tested cost ceiling is the only fixed limit
- (motir-ai) Trace the [graph-notify] refresh-notice timeouts of 2026-10-16 to a named cause — which leg ran past 2 s, with every ruled-out hypothesis recordedPlanning capacity follows demand — a conversation turn never waits behind a planning run, the planning machines and the graph service grow and shrink on their own, and a load-tested cost ceiling is the only fixed limit
- The fleet job inventory is re-measured only by hand, so any workflow edit in a private repo marks image coverage inventory_stale and blocks the Motir-CI-on-fleet switch until someone notices
- The sign-in probe that gates Send to my agent always reads a signed-in agent as signed out — its exec runs without the agent-config environment, so it stats ~/.claude and ~/.codex instead of where the sign-in livesYour own agent instances — host Claude Code, Codex or any sandbox agent in Motir's cloud on your own credential, see them on one page, work with one beside it, and let idle ones hibernate
- The open agent's panel on My agents is cut off below the boot read-out — the tab strip shows but the terminal area under it is clipped
- An undecided plan is unreachable (no Approve or Decline) when a later plan in the same session is decided
- Fleet runners have no disk Docker can use: overlay is refused on the Machine's rootfs and the vfs fallback fills it, so every services: job dies with ENOSPC in its install step
- After the chat story MERGES — release the `motir` CLI that carries the chat server: merge the Version Packages pull request, see `cli-v*` publish the sandbox images, pull the Claude Code and Codex images by digest and see the chat server in them, and chat on a production agentYour own agent instances — host Claude Code, Codex or any sandbox agent in Motir's cloud on your own credential, see them on one page, work with one beside it, and let idle ones hibernate