The Agent
A lean skeleton in Rust that runs one agent per conversation thread. Works through the night.
The agent is not a framework off the shelf. A skeleton in Rust starts one agent per conversation thread on top of the Anthropic SDK, with smaller local models for the cheap work.
In production since 18 April 2026. It runs through the night, researches, writes briefings and reports back over Telegram and WhatsApp.
What it is
Snapshot 3 September 2026A skeleton in Rust that starts one agent per conversation thread, on a cloud model, with the lab's instruments as tools. It has four doors: Telegram, WhatsApp, terminal, browser. It has been running since 2026-04-18; the current process has been up for 0 days, holding 2 threads.
What runs on a timer
| task | when | last run |
|---|---|---|
| evidence adequacy check | daily 00:30 UTC | 2026-09-03 00:34 ok |
| evidence curation pass | daily 04:45 UTC | 2026-09-02 04:48 ok |
| gap-filling run | daily 02:40 UTC | 2026-09-03 02:40 ok |
| infrastructure review | Saturdays 20:50 UTC | 2026-08-29 20:50 ok |
| morning digest | daily 07:00 Berlin | 2026-09-02 05:00 ok |
| night research run | Saturdays 21:34 UTC | 2026-08-29 21:37 ok |
| prior-art check | daily 03:00 UTC | 2026-09-03 03:04 ok |
| socket cleanup | daily 00:00 UTC | 2026-09-03 00:00 ok |
| weekly harvest report | Sundays 06:10 UTC | 2026-08-30 06:10 ok |
| weekly reflection run | Saturdays 21:00 UTC | 2026-08-29 21:03 ok |
Ten of the host's timers, named by what they do. Every one writes a report the morning digest can read.
The night run
A larger cloud model researches open questions in the knowledge graph and the web once a week, writes findings back, and looks for connections across topics. Last run 2026-08-29, briefing over Telegram.
Models, by role
| conversation | one cloud model per conversation thread | cloud |
| night research | larger cloud model for the weekly research run | cloud |
| morning digest | smaller cloud model condenses the morning digest | cloud |
| curation | cloud model judges evidence links in the knowledge graph | cloud |
| heavy worker | a 35B mixture-of-experts model on the lab's GPU machine for long, cheap jobs | local |
| light worker | a 4B model for reflection, summaries and mail triage across five mailboxes | local |
No framework off the shelf. The skeleton is small on purpose: it starts processes, routes messages, keeps threads apart, and gets out of the way. The local rows run on the lab's own GPU machine, see The Model Lab.
What comes out
Continuity
Every session ends with a handover written into the knowledge graph: what was done, what is open, what was rejected and why. The next session boots from it. Decisions are dated and kept with the words that led to them. 1,682 handovers in the graph since 24.03.2026, counted at snapshot time.
Watching itself
The agent reads the reports of the night watches in the morning and says what broke. A watch that only ever reported green is treated as untested; a fault has to be provoked once to earn trust.