Status: 2026-11-13. Two separate projects. This repository is pre-alpha: it
holds docs, repository scaffolding and the dsh-rayrun-bots A1 skeleton, no
publishable code yet. The cloud-computer plugin at
RayrunOS/dsh-rayrun-computer
is untouched and stays its own project; this repository builds the bot layer and
reuses that plugin as an external dependency — two plugins in one DSH profile.
This page has two tracks. What you can run today installs the separate cloud-computer package, which is the required dependency. The target path covers the bot product this repository is building — none of its steps exist yet.
| You need | Why | Status |
|---|---|---|
| DeepSeek Harness | the agent runtime this plugs into — desktop app or dsh web |
the computer package was verified on 0.2.1-rc.5; the design baseline is 0.2.2-rc.1 and re-verification is in progress |
| dsh-rayrun-computer | the cloud-computer layer — a separate project, required by Rayrun Bot | not on npm yet — install from a checkout (below) |
| A Rayrun account | supplies your cloud computer; Rayrun bills it by the second | required — you pay Rayrun directly for compute |
| A DeepSeek API key | the model calls made by the agent on the machine | required — the key and the model bill are yours |
| Node / build tools | not needed: plain ESM, no build step | none |
Nothing is hosted by this project: no account, no relay, no control plane.
Rayrun Bot does not vendor the cloud-computer layer: it is a separate project and a required dependency. Today two plugins must be installed in the same DSH profile; one install is planned once bundle-dependency support is verified. Start with the dependency — it is not on npm yet, so install it from a checkout:
git clone https://github.com/RayrunOS/dsh-rayrun-computer.git
cd dsh-rayrun-computer
dsh plugin --profile web add "$(pwd)"
# restart DSH: the client half only loads when the page startsOn Windows, run the same commands in PowerShell with "$PWD" as the path.
Then, in the DSH page:
- Open Cloud Computer → Connection settings and sign in with your Rayrun account, or paste an API token. A token is created for you and stored in the Harness vault.
- New computer → choose an image, give it at least 2 GB of memory, turn on the remote desktop → Create and open. It takes about 15 seconds.
- Ask the agent to list your machines — it should return the one you just
created. The negative check is worth doing once: before you sign in, the same
request must answer
token_required, never pretend the work happened.
What that package does today:
| Works | Notes |
|---|---|
Seven rayrun_* tools |
sandbox lifecycle, exec, desktop computer use, files, agent delegation, worker layer, workspace sync |
| noVNC desktop in the page | machines created with a remote desktop publish their desktop port |
| Per-session cloud/local routing | in "cloud" mode the host refuses local write/execute tools |
| Verified workspace sync | git bundle + dirty patch, SHA-256 checked on both sides; pulls land in .dsh/incoming/ and are never applied |
| Task delegation | an objective + acceptance criteria, multi-round, start/result can outlive one turn |
| Optional idle stop | idle_stop_minutes; stops the machine, never destroys it |
| Login and vault | password/WeChat/API token; the token stays host-side |
Everything above comes from the separate dsh-rayrun-computer project, not from
this repository. It is verified on DeepSeek Harness 0.2.1-rc.5.
When the bot half lands, the first-run path should be:
- Install
dsh-rayrun-bots(npm is not published yet; a checkout will work). The bundle is intended to declaredsh-rayrun-computeras a dependency so one install brings both plugins — that path is the W0 spike S6 and is TBD. Until then, install the separate plugin first (see above). - Sign in to Rayrun once; the first-run wizard creates your cloud computer.
- Create a bot: a name, instructions, and the skills it can load. It gets its own session on your computer.
- Talk to it, or let a routine wake it up while you are away.
- Approve its first gated action from the panel.
None of steps 1–5 exist yet in this repository. See roadmap.md for what each milestone needs.
| Works today (separate project) | the seven tools, noVNC desktop, per-session cloud/local routing, hash-verified workspace sync, task delegation, worker projection, optional idle stop; verified on DSH 0.2.1-rc.5 |
| This repository | README, docs, issue/PR templates, agent rules, and the dsh-rayrun-bots A1 skeleton with plain-Node checks and a checks-only CI workflow. No publishable code. |
| Planned, unverified | two-plugin install: the Rayrun Bot bundle declares dsh-rayrun-computer as a dependency so one install loads both host and client halves (W0 spike S6) |
| Not built yet | bot roster and bot↔session mapping, a persistent DSH host on the machine, remote approvals, routines and triggers, cost bar, IM channel, npm packages and the release/publish CI path |
| Blocked on the Rayrun platform | safe inbound networking (the token ingress is unusable today and only public works), idle auto-stop, a spend cap, the session-resume contract |
token_required. No Rayrun token is configured. Open Cloud Computer → Connection settings; the page never fakes a successful call.- The machine has no outbound network, and its allowlist is fixed at
creation. Set
egress_allowwhen you create the machine. Runtime changes are not verified: the documented runtime endpoint was not usable in the 2026-11-01 test. destroyis irreversible. The machine and its workspace are deleted together. Use stop/pause to keep the disk; a stopped machine still occupies storage, which is free up to the image's included allowance and billed above it.- Background work costs money and is not self-healing. Rayrun bills by the
second, has no platform-side idle auto-stop, and
max_lifetime_secondsstops a machine without restarting it. If your Rayrun balance is exhausted, the platform can force-destroy your organization's machines including their disks. Watch your balance, and setidle_stop_minutesif you want the plugin to stop idle machines (stop only, never destroy). - A second task gets
409 sandbox_busy. One cloud computer runs one agent task at a time; it does not queue. Wait, or use another machine. dsh-rayrun-computeris not installed. Two plugins are needed today: the bot layer and its separate dependency. Install the cloud-computer plugin from its checkout into the same DSH profile and restart DSH; without it, the cloud tools the bot layer relies on are missing.npm i dsh-rayrun-botsfails. Expected: neither package is published on npm yet.
- Architecture — the layered picture and its tradeoffs.
- Security — what is protected, and what is not.
- FAQ — short answers, including "why is this repo mostly docs?".
- Roadmap — v0.1 / v0.2 / v1.0 exit criteria.
- README — the product page.