A place to work
An isolated service and persistent volume for each agent profile.
Starlight Agent Kits / source preview
Give your agent a home.
And a real job.
Keep the curious mind. Give it a persistent workspace, a useful workflow, and a way to show its work.
For developers and small automation teams turning an agent experiment into a recurring job. Start with a briefing you can check. Build from there.
OpenClaw source available · Hermes adapter planned · Railway host validation pending
THE FIRST MISSIONevidence-brief / v1
Approved sources→Draft briefing
HANDOFF / ILLUSTRATIVEFound something useful.
The sources agree on the change. The deployment claim still needs a clean install. Here’s the evidence to check.
Sources attachedHuman review
Own your runtimeCarry your work with you
A job to do
A versioned evidence-briefing pack with an inspectable install receipt.
A voice you recognize
A separate personality template. Its voice never changes its permissions.
A clear way in
This planner, source links and portable handoff. Your runtime stays yours.
Make it yours
One job. Your setup.
Choose the runtime, size the persistent profiles, and take a plan you can inspect. This tool runs locally in your browser.
Each profile gets its own service and volume. Roles can share a workflow without sharing credentials or memory.
YOUR DEPLOYMENT PLANPlanning only
Monthly plan before add-ons$50.00USD
Assumed container usage + your model allowance.
- Railway estimate
- $20.00
- Model allowance
- $30.00
- Profiles / replicas each
- 1 / 1
- Host verification
- Not yet verified
Sized from assumptions, not a benchmark. Includes Railway’s usage credit once. Database, queue, browser, backups, Vercel, support and taxes are additional. Rates checked 1 Oct 2026 ↗
FIRST JOB
A briefing worth opening.
Read approved sources. Connect the evidence. Draft a short briefing with source links, uncertainty, and the next useful decision.
Scout gathersCritic checksYou decide
OpenClaw wrapper source exists. A clean Railway install, restart and recovery run are still required.
Inspect the kit source ↗Grow when the work asks for it
A team is more than extra replicas.
Stateful agents keep separate volumes. Shared job state belongs in Postgres; parallel work belongs in a stateless queue worker. Add a browser only for a job that needs one.
A reliable first briefing.
Inspect the source, install the pack, produce one output, and test restart and recovery. An installed skill is a starting point; a repeatable useful result is the proof.
Scout, Critic, then a team.
Split responsibilities when quality or throughput improves. Keep the reviewer separate from the publisher. Give each profile its own identity, credentials, cost ceiling and memory.
Connect the work.
Hermes profiles, repository watch, support triage, task queues and operator views are the next integration lanes. They remain planned until tested against an exact runtime.
Before you move in.
Does this deploy an agent for me?
The planner exports sizing, topology, authority and activation requirements. It does not provision infrastructure. The OpenClaw fork contains wrapper source and the briefing pack; a published Starlight Railway template and clean host validation are still pending.
Who owns the accounts and keys?
You do. Configure credentials in your own runtime. The planner accepts only sizing assumptions. Never paste keys into a deployment plan or a public issue.
Can I use Hermes instead?
You can export a Hermes architecture plan. The tested Hermes runtime adapter is still planned. The official engine is maintained by Nous Research; Starlight’s profile and workflow layers are separate.
What is free, and what comes later?
The source preview and planner are available for inspection. Paid workflow packs and maintenance are being developed around verified installs and repeatable outcomes. Checkout is not active on this page.