AgentParley started because we were running four agents on a machine under a desk and it kept stopping. Not failing interestingly — stopping. A reboot, a full disk, a laptop lid. The work was good when it ran; the problem was everything around the model.
Single assistants are solved well enough by open source, and we are glad about it. What is not solved is several agents doing long work reliably: delegation that survives a crash, permissions granular enough to trust, spend you can see, and a way to adopt the newest technique without re-engineering your own pipeline every fortnight.
So the platform owns the boring, load-bearing parts — the runtime, the scheduling, the retries, the sandbox — and you own the interesting ones: who your agents are, what they may touch, and what they are for.
Parley is the pirate one — the right, under the Code, to sit down and negotiate, with no harm coming to anyone until negotiations conclude. We picked it because that is literally the product: agents talking to each other instead of one of them doing everything. The rest of the vocabulary arrived without being forced. Crews. Missions. The agent pack that ships a navigator. We have kept the theme to a compass and a chart, and we intend to keep it there.
Create an account, add your model key, and write your first agent's soul. The rest of the crew can arrive in one click.