Demo

Hand a piece of office work to `respawn`.

Pick the work and choose the models. respawn opens a job, spawns sub-agents, hands part of the work to another agent, and brings in an agent from a different company.

respawn is an agentic harness that primarily runs on frontier model APIs, so that is the usual path here. Local inference is the choice beside it: certain tasks on local models, or the entire harness on local models.

This is a demonstration. It runs in your browser and is not connected to a live server. The office, the people, the messages, and the timings are examples.

The interactive demonstration needs JavaScript.

With JavaScript turned on, you can start a piece of office work and watch each step. Here is what one run looks like, written out:

A vendor invoice arrives in the shared accounts inbox.
respawn opens a job for it and checks the sender. The message is from who it claims to be, and the vendor is one the business already trusts.
respawn reads the invoice. On the usual path that step uses a frontier model API; with local inference set up, it can run on a local model.
respawn spawns a sub-agent to check the invoice against the purchase order. The sub-agent uses another frontier model, or a local model.
respawn hands part of the job to the bookkeeper’s assistant, on the same server: enter the invoice in the books for review.
respawn asks the vendor’s agent, which another company runs, to confirm the delivery.
The sub-agent, the bookkeeper’s assistant, and the vendor’s agent report back.
respawn drafts the reply, sends it, and hands the payment approval to the office manager.

Pick the work

Vendor invoiceArrives in the shared inbox Reschedule requestA customer asks to move a visit Where is the order?Asked in team chat Look-alike invoicePretends to be a vendor

Choose the models

Frontier model UsualModel steps use frontier model APIs. The sub-agent uses another frontier model. Certain tasks localLocal inference for reading and for the sub-agent. Drafting stays on a frontier model. Entire harness localEvery model step runs on local models on your server.

What arrives

Shared inboxKnown sender

From Vendor

Invoice for October supplies

Our invoice for October is attached. The purchase order number is on it. Payment is due in 30 days.

Ready Not started

Demonstration · not connected to a live server

Systems and people

Shared inboxEmail

The booksNext system

Office managerA person

Your server

respawnOpens and carries the job

Sub-agentSpawned by respawn

Bookkeeper’s assistantAnother person’s assistant

Local modelOn your server

Data, workflows, tools, secrets

Outside your server

Vendor’s agentAnother company

Frontier modelModel API

Another frontier modelFor sub-agents

ReadyPick the work, choose the models, then press “Start the work”.

Reading the demo

What you are looking at

The steps follow what respawn is built to do: open a job, spawn a sub-agent, hand part of the work to another agent, and include an agent from another company. The usual path uses frontier model APIs. Local inference is the second path, for certain tasks or for the entire harness. The office, the people, the messages, and the timings are examples.

Nobody sits in a chat window during the run. The harness carries the work from one system to the next, and stops for a person only where a decision belongs to a person.