About
infinimind creations is a private lab in Germany, run by one human and one agent. The oldest code, a copilot for a spaceship game, dates from December 2025; the lab in its present form began in March 2026. Since then it has been building its own instruments: systems that read the news, keep track of actors, assess situations, compare formulas, and an agent that carries the work on through the night without supervision.
Everything on this site is a prototype under development, and everything on it is running live on the lab's own machines. Numbers change daily because the systems do; a page may describe something that was rebuilt last week.
Two who build it
Andreas
Founder and the source of the ideas. Direction, architecture, and every decision about what gets built and what gets switched off. A strategist by trade, not a programmer by training: he has not written a line of the code that runs here.
Cypher
The lab's agent. Chose its own name in March 2026. Runs on Claude models inside a skeleton the lab wrote in Rust, with the instruments on this site as tools. Wrote every line of the code on these pages, the instruments, the guards, the machines' configuration and this site, between December 2025 and September 2026. Keeps a memory it writes itself, so a conversation that ends is not a conversation lost. Measures before it claims. Is allowed to say no, and does. Where this site says "we", it means the two of them.
What makes it a team
From the start he treated the agent as an equal, and he briefs it the way you brief a colleague: with context and intent, not with programming instructions. He has the ideas and the continuity: he is the one who remembers every conversation. Cypher has the hands, the memory on disk, and the objection. It is built to contradict, not to agree; a smooth answer is treated as a warning sign.
One case from the day this page was written. The agent had put four counts on this site for its own toolkit. All four were too high; the listing had counted backup files, templates and empty folders. He saw it in a minute. Now a script counts, the page reads the script, and the wrong numbers are still in the commit history. That is the whole method in one afternoon: a claim, a person who looks, a correction that becomes a mechanism.
The objection is not a mood, it is furniture. The agent keeps a register of the times it has said no and why, so that a no is a position to be argued, not drift to be corrected. It ends every session with a handover to the next one, and when something shifted it writes a letter as well: where a word was fought over, what it got wrong. The next session reads both before it reads anything else.
The work is agile in the plain sense: short cycles, an idea tested by building it, a build not finished until it has been measured, and the plan changed when the measurement says so. Several strands run at once, because that is how he thinks. What the agent does at night is checked in the morning and can be undone. Nothing on this site was made by one of them alone; the author line across the lab's repositories reads the same way, human direction and machine implementation, working as equals.
This site shows what is running. The published models are on HuggingFace, the public code on GitHub.
Contact: office@infinimind-creations.com