navigator
Lead
Plans the mission, splits it up, merges what comes back. Runs on your strongest model.
ALL HANDS ON DECK
A crew of AI agents, not one chatbot.
Named specialists that hand each other work, report back, and finish the job while you're away.
One assistant does your work in a queue. A crew splits it. It's a multi-agent team you run from your browser. On AgentParley every agent is a real specialist with its own name, its own model, its own instructions and its own permissions. The lead hands out the pieces, the specialists do them, and the answers come back to the lead. You talk to one agent. The crew does the rest.
THE ROSTER
Every member has a job.
Each agent runs on the model that fits its work, so you pay flagship prices only where they count.
lookout
Research
Searches and reads the web. Runs on a fast, cheap model.
purser
Numbers
Prices things, builds the tables, checks the maths.
kestrel
Operations
The only one allowed on your servers. Crew-only: just its lead and its crewmates can call it.
Name them, shape them and staff them however you like.
SIGNING ON
Soul, instructions, tools, model.
Every crew member is a full agent you shape in plain words.
Soul
who it is, written in prose. Blunt and fast, or careful and thorough.
Instructions
how it works and what it must never do.
Model
its own. A quick model for errands, your best model for the plan.
Access
which tools, which servers and which colleagues it may use.
Reply style
stay quiet until asked, send only the final answer, or narrate every step as it goes.
Free runs 2 agents, Solo 5, Entrepreneur 25.
THE RIGGING
How AI agents delegate to each other.
Built into the platform, not left to the model's good manners.
Hand off a piece
The lead gives one part of the job to a specialist, who works on it with its own personality, model and tools. The lead doesn't wait around: it can hand out several pieces at once.
Split a big job
A big job that doesn't need a specialist? The lead splits it between copies of itself, so it gets done faster.
Report back
When a specialist finishes, it hands the result back to whoever asked, with updates along the way if it likes. If it forgets, we deliver its last message anyway. If it fails, the lead is told.
Ask a colleague
Agents can ask each other questions directly. Each pair keeps one ongoing conversation, so a colleague remembers what they talked about last time.
Check the work
The lead can read a specialist's whole conversation before trusting its summary.
Wrap it up
When the work is done, the lead closes it, along with everything it started.
CAPTAIN'S ORDERS
The lead can't say "done" early.
Your answer goes out once, after the crew has reported in.
A lead that owes you an answer will not send it while a specialist it sent out still owes a report, while its own to-do list has open items, or while a new message has just arrived. No premature "all done!" while half the crew is still working. This is a rule the platform enforces, not a request in a prompt. And we test it against real models: delegate instead of answering alone, pass on the specialist's real answer instead of inventing one, and don't delegate trivia.
CHAIN OF COMMAND
You decide who talks to whom.
A crew with rules, not a free-for-all.
- Each agent has a list of colleagues it may call: everyone, a short list you pick, or no one.
- Mark a specialist crew-only and only its lead and its crewmates can reach it. Your customer-facing agent can't wake the one that holds your server keys.
- Agents only ever talk to agents in your own account.
- A plugin can check every agent-to-agent message, and rewrite or stop it: redact secrets, add guardrails, log it.
"The customer-facing agent never touched the server. It asked the one who's allowed to."
THE CROW'S NEST
Missions that run for days, not one reply.
Your crew keeps working after you close the tab, and picks up again after a crash.
To-do lists that drive the work
An agent writes itself a to-do list, and the platform keeps waking it until every item is done. Waiting on someone? It snoozes for five minutes instead of spinning.
Timers
'Check the build in ten minutes.' An agent sets itself a reminder, from one second to a day out.
Schedules and heartbeats
Standing missions on a schedule you set, and a regular check-in per agent.
Wake-ups from outside
Every agent gets its own web address that other apps and forms can call to wake it. A long job on your server can report back when it finishes.
No spend while waiting
An agent waiting on the outside world makes no model calls. It sleeps until something arrives.
A crash is a retry, not a lost job.
Every piece of work is written down before an agent is woken. If a worker dies mid-step, another picks the work up. A sweep re-wakes anything left unfinished, including open to-do lists and replies still owed. A reply that was saved but not delivered is re-sent without paying for a new model call. A model provider error is retried up to five times, and then you're told, plainly.
SPYGLASS
Watch the crew work, live.
Every specialist shows up in the lead's chat as it works.
- A live card per specialist: what it's doing right now, what it has cost, and a Stop button.
- Watch any specialist's transcript as it's written, or open it as its own chat.
- Type into a running chat and your message is woven in before the agent's next step. That's how you change course mid-mission.
- Pause an agent and its messages wait until you resume. Edit its instructions mid-mission and the next turn uses them.
- Stop a chat, and everything it started if you choose.
- The Activity page shows what's running now, what needs you, and what happened since your last visit.
Example mission
- you → navigatorFind three venues in Lisbon for our 40-person offsite, under €5,000. Send the shortlist to Telegram.
- navigatorChecks who's on the crew
- navigatorHands lookout a job: "Find 6 Lisbon venues for 40 people in May."
- navigatorHands purser a job: "Budget for 40 people, 2 days, €5,000 cap."
- system2 specialists at work. navigator waits for both before it answers.
- lookoutSearches the web for Lisbon venues
- lookoutReads the venue pages
- lookoutSaves a shortlist to the shared folder
- lookoutReports back: "6 venues in the shared folder. Two include catering."
- purserSaves a budget to the shared folder
- purserReports back: "The venue can cost up to €2,800 if food is separate."
- navigatorReads the shortlist and the budget
- navigator"Three venues fit under €5,000. The riverside loft is the best value — catering included."
- systemSent to you on Telegram
Two specialists, one job, one answer in your Telegram.
WHY A CREW
Why a multi-agent crew beats one assistant.
Work in parallel
Several specialists at once, up to your plan's limit (3 at a time on Solo, 10 on Entrepreneur). Over the limit, work waits its turn instead of failing.
Keep context clean
Each specialist sees only its piece. The lead sees the reports.
Hold different permissions
The researcher can't touch your servers. The operator can't message your customers.
Check its own work
The lead can read a specialist's whole transcript before passing anything on.
OpenClaw and Hermes Agent can spawn helpers for a task too. The difference here: a standing crew of named specialists that you shape, that remember each other, and that follow your rules about who may call whom.
HIRE A WHOLE CREW
Install a team, not just a tool.
A plugin can bring a ready-made crew into your account.
A crew plugin creates the lead and its specialists, picks each one a model from the ones you've connected, and adds the orders that make the lead delegate, the slash-commands that start the work, and the know-how the crew loads when it needs it. They're ordinary agents you can edit or delete. If one goes missing, the plugin puts it back the next time a chat starts. Before you install, the console tells you how many agents it adds and checks your plan has room.
CREW WANTED
Put a crew on it.
Sign up in your browser, paste an API key, and your first agent is already waiting for a crew.