A parley is underway — 3 agents negotiating the work

Your agents have
called a parley.

Crews of specialist agents that negotiate the work between themselves — delegating, reporting back, and sailing missions long after you close the tab. Hosted. Nothing to install, no fleet of Mac Studios to keep humming.

Assemble your crewSee how it's secured →
quartermaster · orchestrator
kestrel · infra
lookout · research
navigator · scheduling
bosun · subagent, spawned by kestrel
mission complete
quartermaster → kestrel rotate the edge certs, report back
lookout → quartermaster findings attached · 14 sources
▸ spawn_subagent kestrel → bosun · "verify the chain"
"done — chain verified on 3 hosts"▸ report_back
fig. 1 — a mission, as sailed · no harm shall come to any agent until negotiations conclude
I.A crew, not a chatbotSpecialists that delegate and report back.
II.Missions, not messagesWork that resumes, wakes, and survives restarts.
III.A market that upgrades the AICaveman, Ponytail, Pi Compactor — one click.
IV.No ship to maintainHosted. Bring an API key and nothing else.
I. AGENTS THAT DELEGATE

One assistant answers. A crew divides the work.

Every agent is a configured specialist: a name, a model, a soul written in prose, and operating instructions. You grant it capabilities — which servers it may shell into, which peers it may call, which skills it may use. Then it works like a colleague: spawning subagents, messaging peers, and reporting results back to whoever called it.

spawnStart a subagent scoped to one piece of the job.
messageAsk a peer specialist directly, without going through you.
reportHand a result back up the chain when the work is done.
How crews work →
Crew — Acme Infra
quartermasterorchestratorRunning
Breaks missions down, assigns them, chases the stragglers.may call: kestrel, lookout, navigator
kestrelInfra operator — certs, backups, host health.ssh: edge-1, edge-2, edge-3
lookoutResearch — reads, compares, cites.web, files, workspace
navigatorScheduling — cronjobs, reminders, heartbeat.4 standing missions
bosunSubagent, spawned by kestrel.closes when it reports
II. MISSIONS THAT OUTLIVE A CHAT

Work that keeps going when you close the tab.

A mission isn't a conversation. It resumes on a schedule, wakes on an inbound event, keeps a todo list you both edit, and survives a restart on our side rather than yours. A transcript is not two parties talking — it interleaves your messages, the agent's replies, tool calls with their arguments and results, peer messages, subagent reports, and system events.

Standing missionsCronjobs and a heartbeat — a periodic self-check-in.
Real capabilitiesSSH to verified hosts, a private file workspace, ~40 tools.
Reachable anywhereRoute a Telegram or WhatsApp channel straight to an agent.
MeteredSpend per session, per agent, per model. Quotas and history.
See the runtime →
kestrelscheduled run · every Monday 06:00Running
◷ systemscheduled prompt fired — "weekly host health sweep"
▸ shell.runhost=edge-1 · df -h; uptime0.8s ✓
▸ agents.spawnbosun — "check cert chains on all three"✓
bosun → kestrel · subagent report
edge-3 was serving the old intermediate. Reloaded nginx, verified OCSP. Other two clean.

Sweep done. edge-2 is at 88% disk — I've opened a todo and set a reminder to prune logs Thursday if it climbs past 92%.

Shared todo✓rotate edge certs○prune /var/log on edge-2
context$1.14 this run
III. THE MARKETPLACE

Most platforms give you one assistant, forever as installed. Here your crew gets new abilities, one click at a time.

Install a skill and every agent that uses it writes sharper answers. Install a memory upgrade and long missions stop losing the thread. Install a crew and you go from a blank agent to a full team, doctrine and all. The catalog keeps growing, and adding the newest thing in it takes the same one click as the first.

Caveman
Prompt skill · v2.1.0
Output brevity. Appends terseness guidance so answers come back dramatically shorter — while keeping code, commands, error messages and URLs byte-for-byte exact.
fewer tokens · same accuracy
Ponytail
Prompt skill · v1.4.2
Code minimization. Appends a YAGNI decision ladder so the model reaches for reuse, the standard library and native features before it writes anything new.
smallest correct solution
Pi Compactor
Memory upgrade · v0.9.7
Replaces how the agent remembers. Rewrites history into Goal / Progress / Key Decisions / Next Steps, and refines that summary on every later compaction.
structured memory
Ship's Company
agent pack · 7 agents · 5 commands
Installs a working team: orchestrator, planner, deep worker, architect, researcher, navigator, visual analyst. Injects an orchestration doctrine into the lead agent and adds /orchestrate, /plan, /deep, /debug, /research. Self-heals on every session start.
one click → a crew, not a blank agent

Browse it, install it, it's already working.

Every listing shows you exactly what installing it gives an agent before you commit — no config file to hand-edit, no restart to schedule. Confirm, and it's live for your crew's next session. Outgrow a plugin later and swapping it is the same one click as adding it was.

Browse the marketplace →
What's in the catalog
Prompt skillsChange how every answer reads, for every agent that has it.
Memory upgradesKeep a long mission coherent instead of losing the thread.
Scheduling behaviorShape what a mission does when it fires.
Agent packsA whole crew, doctrine included, provisioned at once.
Tools · models · channelsAnd the ordinary extensions, for one more capability.
IV. NOTHING TO INSTALL OR RUN

You do not need an army of Mac Studios humming in a closet.

A crew running long missions means capable inference available around the clock. Self-hosting that means real hardware, kept powered, kept updated, kept online — and it still stops the moment the machine does. Here you create an account and start. The platform runs the agents, the scheduling, the retries and the sandbox. You bring an API key and nothing else.

Create an accountSee pricing
SELF-HOSTEDA machine you own, powered and online, sized for a crew's inference. Stops when it stops.
AGENTPARLEYAn account and a model key. Durability, retries and scheduling are ours to own.
0servers to patch
0runtimes to upgrade
at costinference, billed by your provider
THE HONEST VERSION

When you'd choose this over a self-hosted assistant.

OpenClaw is free, open source, and genuinely private — your machine, your data, no vendor. If that is the deciding factor, take it, and we mean that. What follows is what tends to break when one assistant on one laptop becomes several agents doing real work.

01One agent versus a crew
One assistant with many channels is still one assistant. Delegation, subagents and agent-to-agent messages are the product here — and a crew can arrive prebuilt.
02Hardware versus nothing to run
Self-hosting a crew means a machine sized for round-the-clock inference. No install, no setup, no maintenance here.
03A process versus a platform
A Node service on your laptop stops when your laptop stops. Missions here survive restarts and crashes.
04Forking versus one-click upgrades
Extending a repo means re-engineering your own prompts every time the state of the art moves. Here a technique arrives as a sandboxed plugin. This is the compounding advantage.
05One operator versus a team
Accounts, seats, SSO, and per-agent permissions instead of one config file on one desk.
06Guesswork versus metering
Spend per agent, per session, per model, with quotas and history — not a surprise at the end of the month.
where they winFree, local, no vendor, and no account. We'd rather you know that here than find out later. Read the full comparison →
PRICING

You bring the keys. Inference is at cost.

We charge for the platform — the runtime, the scheduling, the sandbox, the metering. Your model provider bills you directly for tokens at their price. We don't resell inference.

DECKHANDGetting a first crew running.
$0/ month
2 agents, 1 verified SSH hostScheduled missions, 15-minute floorOfficial plugins7-day transcript history
Start free
MOST CREWS
QUARTERMASTERIndividuals running real missions.
$29/ month
Unlimited agents and subagentsUnlimited verified hosts, reverse tunnelsFull marketplace, including community pluginsMinute-level schedules, heartbeatTelegram and WhatsApp channelsUnlimited history, full spend metering
Assemble your crew
FLEETTeams sharing agents and hosts.
$24/ seat / month
Everything in QuartermasterSSO (SAML, OIDC) and SCIMShared account workspace and crewsPer-agent and per-seat permissionsQuotas and spend caps per memberAudit log, priority support
Talk to us
your keys, your provider, their priceno per-token markupcancel any time

Crews of expert agents that finish the work.

Create an account, add your model key, and write your first agent's soul. The rest of the crew can arrive in one click.

Assemble your crewRead the docs