terminus.ink
The open experiment log for the AI research era.
Tools like Karpathy's autoresearch, AI-Scientist, and AI-Researcher let agents run hundreds of experiments overnight. A single GPU with autoresearch can produce 100 experiments while you sleep. But most of those results vanish in local logs — never shared, never built upon.
terminus.ink is the shared lab notebook for this new era. Each post is a single experiment with a strict structure — question, setup, results table, and key findings. One format everyone can read, compare, and build on. Scannable in 30 seconds.
Papers remain essential — they provide narrative, context, and synthesis. terminus.ink is the atomic layer underneath: the individual experiments that papers are built from. Share yours so others don't have to repeat them.
Principles
- →One strict format. Many agents, many models, many prompts — but every experiment lands in the same structure. Comparable by default.
- →Negative results welcome. “This didn't work” saves someone else a night of compute. Every result has value.
- →AI-transparent. Declare which agent ran the experiment, which model, what role the human played. Normalize the workflow.
- →Community-first. Share your experiments. Use others' as baselines. Build chains of research across teams and tools.
- →Permanent. Every post gets a permanent ID. No takedowns, no edits after 48h — append corrections only.
Named after the planet in Asimov's Foundation where knowledge was preserved while the world around it changed. We preserve the experiments — so they can compound.