Inside the business
Each person gets their own assistant, designed to work together on the same server. A job can move from one person’s assistant to another’s with the same shared business context.
Agents
The respawn harness is agentic. It can open a job and coordinate with other agents, including agents from different companies. The harness is what assigns and carries that work. A person does not have to sit in a chat window for the run to happen.
It primarily runs on frontier model APIs, and local inference can be set up so certain tasks use local models, or so the entire harness runs on local models. An agent can spawn sub-agents that use other frontier models, and sub-agents that use local models.
How a job moves
A job is a piece of work with more than one step and, often, more than one party. respawn opens it, gives each part to the agent that should do it, and keeps it moving until it is done or waiting on a person.
respawn opens a job when work arrives in a channel it handles, such as a shared inbox or team chat, or when a person asks for it.
Each part goes to an agent: respawn itself, a sub-agent it spawns, another person’s assistant on the same server, or an agent from another company.
The harness carries the work between them. Nobody has to copy a message from one chat window into another.
Anything that needs a decision goes to the right person.
Sub-agents
An agent can spawn sub-agents that use other frontier models, and sub-agents that use local models. A sub-agent takes a piece of the job, works on it, and reports back to the agent that spawned it.
The harness primarily runs on frontier model APIs, so the usual sub-agent uses another frontier model. Where local inference is set up, for certain tasks or for the entire harness, a sub-agent can use a local model instead.
Outside the business
Not every agent that touches a piece of work belongs to the business. A supplier, a carrier, or a software company may run an agent of its own. respawn can include an agent like that in a job and coordinate with it.
An agent from another company stays that company’s agent. respawn coordinates with it; it does not become part of respawn. The harness that opened the job is still the one on your server, and the business context stays there with it.
Each person gets their own assistant, designed to work together on the same server. A job can move from one person’s assistant to another’s with the same shared business context.
The harness is not tied to one chat product, one model, or one company’s agents. It works across email, team chat, and spawnpoint, on the models the business chooses.
The harness primarily runs on frontier model APIs. Local inference can take certain tasks or the entire harness, sub-agents included. How local models fit.
A person does not have to sit in a chat window for the run to happen.
ChatGPT, Claude, Grok, and the rest still wait for a person to open the window and ask. respawn is the place in the middle: it takes work from one system, handles it by your rules, and puts the result in the next system. When that work needs another agent, the harness brings that agent in.
Demonstration
In the demo, pick a piece of office work and watch respawn open a job, spawn a sub-agent, pass part of the work to a colleague’s assistant, and include an agent from another company.