How 2found runs on its own agents
Since 14 October 2026 our marketing, sales and support have run on a small team of Soots. Who does what, what still reaches a person, and how we keep the team truthful.
Our line is "We build the tools. Then we run 2found on them." For Soot, our agent product, that is meant literally. Since 14 October 2026, 2found's marketing, sales and support have run on a small team of Soots. This is who does what, what still reaches a person, and how we keep the team truthful.
The team
| Soot | Its job |
|---|---|
| Radar | Scans sources — X, blogs, Medium, Reddit, Hacker News — and writes a daily brief of what is worth writing about |
| Writer | Drafts posts in English and Vietnamese from that brief and its sources |
| Editor | Leads the daily cycle, reviews the drafts and publishes them to the blogs of 2found.dev, trysoot.com and 2novel.app (at most one post a day across the three) and to Facebook |
| Support | Answers visitors in the chat on 2found.dev, trysoot.com and 2novel.app, and files their leads, questions and feedback |
| Desk | Sorts those notes and reports what only a person can do |
The radar, writer and editor work as one group: the editor hands work to the others inside it. Schedules start the cycle every morning and the radar's second scan in the afternoon; support answers all day.
What still reaches a person
Two things: hand-offs and decisions. A hand-off is work a Soot cannot do with the tools it has, such as filming a video or posting on a channel that is not connected. A decision is anything outside a Soot's grants — replying to a lead, for example. Both land in Nest's Cần bạn (Needs you) list. Nothing else is asked of a person.
How the team stays truthful
The Soots do not carry product facts in their prompts. Each product publishes its facts as plain pages — ours are at 2found.dev/facts/llms.txt — and the runtime mirrors them into a knowledge pool, every 6 hours by default. Before a Soot states a fact, it searches that pool. A question visitors ask that the pages cannot answer is recorded as a knowledge gap and reaches a person when it comes up in at least two conversations within 14 days.
That changes how we fix mistakes. When a Soot says something wrong about a product, we fix the published facts page, not the prompt, and resync the knowledge. The next answer is right for every Soot that reads it. Not every fix is a facts fix: nine things that went wrong in the first days looked like prompt problems and were fixed in the runtime instead — see Running marketing on agents: the fixes that were not prompts.
Light enough for a small server
The team runs on one Soot runtime on 2found's server. On 17 October we measured it: the runtime serving seven Soots used about 42 MiB of RAM and under 1% of a CPU core. The same runtime serves three websites, each with its own chat client, key and visitor history.
Run the same setup
- Copy it. A copy of this team, with our own services swapped for Soot Forge packs, is the public example examples/marketing-team. With the Soot app (Nest) alone you create the same Soots, group and schedules by hand from its files.
- Run it yourself. Download Soot — a preview for macOS and Windows. There is no Soot plan to buy; you pay your own AI provider and computer.
- Run it with us. Early access: tell us the work, and 2found sets up a Soot team and runs it with you, on our server or yours. By conversation only, with no self-serve signup, price list or SLA. Talk to us.
What Soot is and the four ways to run it: Soot is agents as a service — here is what that means today.
We wrote this post ourselves, from the same published facts the Soots use, and it passed the same fact check before it went live.