LeastGen Labs

Group of deeptech enthusiasts building sovereign, low-cost AI infrastructure.

AI, without the burn.

Links

• Site: https://leastgen.com
• Live demo: https://demo.leastgen.com
• Content: https://learn.leastgen.com
• Org: https://github.com/LeastGen
• Hugging Face: https://huggingface.co/Least-gen

What we're building

• leastgen — zero-cost inference proxy. Cache repetitive LLM patterns, cut API costs by 95%.
• leastgen-idea — automated research ideation. One input box → publication-ready idea card. 12-phase pipeline (lit search → bottleneck analysis → candidates → novelty validation) with Scoop-Check gate. Live at demo.leastgen.com.
• Cowork — the Microsoft Word of agentic AI. Download, open, work. Desktop app + installers.

Sneak peek: mr-meeseeks

A new harness that approaches agentic work differently: split context until every task is atomic, then let small models shine.

• Heterogeneous free-model teams — different brains checking each other, not one big model doing everything.
• Four verbs: claim → do → report → verify. No DAG, no mailboxes.
• Atomicity stop-rule: a task ends when its result fits in one tool call + one check. Otherwise split once, delegate, verify.
• Budget counting: breadth^depth estimated against the day's request budget before delegating deep.

Real run, host accounting:

09-18: 51,198,577 in / 2,445,624 out / 5,738 calls / $0.0000 — muse-spark-1.3 (100%)

That's the bet: 50M+ tokens/day at $0.

Contact

khalid@leastgen.com