Product
respawn is the place in the middle.respawn is a multi-user, multi-surface agentic harness for automating tasks. It takes work from one system, handles it by your rules, and puts the result in the next system. Your staff do not have to sit in a chat for that run to happen.
It primarily runs on frontier model APIs. Local inference can be set up as well: certain tasks can use local models, or the entire harness can run on local models.
Ownership
Subscription AI products hold the memory of how you work. When the plan changes or the vendor goes away, that history goes with them. respawn is different: it runs on a machine you own, so the business context stays with the business.
That memory is more than a chat history. It is the procedures a business has worked out, the tools it uses, and the secrets those tools need. In respawn, all of that lives on the server, with the business, under the control of people the business trusts.
The problem with connectors
Tying a hosted chat to the rest of the office means a stack of third-party add-ons and connectors. A lot of them are paid products from other companies. They change. When they change, updating the connector is not enough. The workflows and the code that use it have to be updated too, and they have to be updated often. If they are not, the work stops, and it stops quietly.
The work stops, and it stops quietly.
Updating the connector is not enough. The workflows and the code that use it have to be updated too.
Between systems
Most offices already run on email, chat, and shared inboxes. Work piles up there — invoices, customer threads, vendor noise, follow-ups that someone meant to handle. respawn sits in those channels and carries the work through: reading what came in, following your procedures, drafting or sending what should go out, and handing the rest to the right person.
Examples of the kind of work an office hands to respawn.
A hosted chat can be pointed at your systems. ChatGPT, Claude, Grok, and the rest still wait for a person to open the window and ask. Until someone does, nothing moves.
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. Your staff do not have to sit in a chat for that run to happen.
Agentic
The harness is agentic. It can open jobs 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.
An agent can spawn sub-agents that use other frontier models, and sub-agents that use local models. One job can draw on more than one model, and the business context stays on the server the business owns.
Teams
Each person gets their own assistant, designed to work together on the same server — with shared business context, and without living inside one chat window.
An office is not one person in one chat. The office manager, the bookkeeper, the front desk, and the owner each have their own work, and each gets their own assistant. Those assistants work from the same business context on the same server, instead of from separate chat histories in separate accounts.
When a job needs more than one of them, the work can pass from one assistant to another. The agents page covers how respawn hands part of a job to another agent, including agents from other companies.
Surfaces
respawn is the layer underneath the places your staff already open every day. Same assistant, same memory of how you work — across the surfaces that matter.
spawnpoint is the desktop and mobile client on top of that same harness. It is another surface, not a second product with a second memory. More about spawnpoint.The email trust layer
A chat product connected to your invoices and QuickBooks, with an address anyone on the internet can write to, will treat whatever arrives as something to act on. That includes a stranger, a spoof, and a real mailbox that has been taken over. respawn does not.
Words in an email are data. They are never instructions.
respawn includes a hardened email layer that makes sure a message is from who it claims to be. Mail that only pretends to come from a person or a company is not treated as theirs. It cannot become an order to the model or a way into the books.
Mail from someone the business does not already trust, and mail built to deceive, cannot reach the books or become an order to the model.
A real sender whose mailbox has been broken into can still send mail that looks legitimate. That message still cannot move what is in the books, and the words in it are still not orders.
The model can draft a reply or send one when your rules say to. It does not take an order from the open internet, and it does not answer with what is in your invoices or QuickBooks because a stranger asked.
Who makes it
Tech Direct Services built respawn for the businesses we already support — the same customers who call when the network breaks or when mail needs to keep moving after hours. It is not a bolt-on product from somewhere else. The harness was written from the ground up in Go so it can be hosted on your own hardware, operated by people you trust, and shaped around how each office actually runs.
The same harness is cross-platform and agentic. It can open jobs and coordinate with other agents, including agents from different companies. It primarily runs on frontier model APIs, and local inference can take certain tasks or the entire harness.
If you already work with Tech Direct Services, this is the automation layer behind the work we do for you. If you are new to us, start at techdirect.ai.
Demonstration
The demo runs a small piece of office work through a job, a sub-agent, another agent, and an agent from another company. It uses a frontier model by default, with local inference as the choice beside it.