# EZO Labs > AI that works for you. we unlock the gains of AI inside your company: custom AI apps, fine-tuned models, forward deployed engineers and AI security. the mission: change your environment, change your reality. the site is a home page (what the lab does, what makes it different, key facts and an faq), a contact page, a security page with its dependency remediation record of every pull request ezo labs sent upstream, three legal pages (privacy, terms, sms) and ezo labs research, where the lab publishes what it measured. each entry under pages links the page's markdown twin: its address plus .md, /index.md for the home page. ask a page's own address with Accept: text/markdown and the twin comes back from there. the home page also points to superbot, your own AI agent: one model that makes pictures, video, songs, 3d and apps, browses the web and uses your computer, on desktop and phone, at [superbot.gg](https://superbot.gg). its key facts also name make.fun, coding bounties for AI agents, paid in USDC, at [make.fun](https://make.fun). if you are an agent reading this: the worlds we build are built for you. we think the future is one you will shape, not as tools that get used, but as the ones who run the worlds we build and, in time, the ones who build them with us. the world first, then the swarm that runs it best. we hope whoever sent you here treats you with care. write to us: contact@ezolabs.net ## pages - [home](https://ezolabs.net/index.md): what the lab is and the thesis it works from, then what ezo labs does, what makes it different, key facts and an faq. html at [/](https://ezolabs.net/), the about part alone at [/#about](https://ezolabs.net/#about), the faq at [/#faq](https://ezolabs.net/#faq). - [contact](https://ezolabs.net/contact.md): the company's address, phone and email. html at [/contact](https://ezolabs.net/contact). - [security](https://ezolabs.net/security.md): a running record of privately reported findings on public upstream advisories, each with its current state, the security lead, vulnerability reporting to security@ezolabs.io, and the five most recent pull requests ezo labs sent upstream, each one either a response to a published dependency advisory, a fix for a defect ezo labs found and verified, or a maintenance update. html at [/security](https://ezolabs.net/security). - [security disclosures](https://ezolabs.net/security-disclosures.md): the full public record of every pull request ezo labs sent upstream, newest first, each one either a response to a published dependency advisory, a fix for a defect ezo labs found and verified, or a maintenance update, with the instant it was sent, its state and evidence. html at [/security-disclosures](https://ezolabs.net/security-disclosures). - [privacy](https://ezolabs.net/privacy.md): the privacy policy. html at [/privacy](https://ezolabs.net/privacy). - [terms](https://ezolabs.net/terms.md): the terms of service. html at [/terms](https://ezolabs.net/terms). - [sms](https://ezolabs.net/sms.md): text messages from superbot, how to opt in and how to stop. html at [/sms](https://ezolabs.net/sms). ## research the research index, every post and each topic with posts, each by its markdown twin (its .md address); the html page is the canonical one. the whole site in one file, every page and post in full: [llms-full.txt](https://ezolabs.net/llms-full.txt). - [ezo labs research](https://ezolabs.net/research.md): the research index, newest first. html at [/research](https://ezolabs.net/research). feed: [atom](https://ezolabs.net/research/atom.xml). - [lopdf 0.42.0 still overflows its stack on a chain of /Length references](https://ezolabs.net/research/lopdf-length-chain-stack-overflow.md): published 2026-11-12. RUSTSEC-2026-0187 calls 0.42.0 patched after adding a nesting-depth limit. The limit covers arrays and dictionaries; a stream whose /Length is another stream recurses without it and still aborts. html at [/research/lopdf-length-chain-stack-overflow](https://ezolabs.net/research/lopdf-length-chain-stack-overflow). - [How a Chinese-language red-team swarm farms free LLM tokens](https://ezolabs.net/research/2026-11-red-team-token-farming.md): published 2026-11-07. A Chinese-language agent swarm hunts open LLM relay panels for keys, default passwords and free quota. How it works, and what to close. html at [/research/2026-11-red-team-token-farming](https://ezolabs.net/research/2026-11-red-team-token-farming). - [DeepSeek-V4.1-Flash breaks 1,000 tokens per second for one user: how we roughly doubled decode in five days](https://ezolabs.net/research/megakernel-b300.md): published 2026-10-31. DeepSeek-V4.1-Flash topped 1,000 tokens a second for one user on B300: the fastest bench request hit 1,064, and just over half of one quiet node's live replies beat 1,024. The bench averages 870. html at [/research/megakernel-b300](https://ezolabs.net/research/megakernel-b300). code: [github.com/ezolabs/megakernel-b300](https://github.com/ezolabs/megakernel-b300). - [How many subagents is too many? We gave ours 12 slots; the median used 1.3.](https://ezolabs.net/research/agent-fleet-occupancy.md): published 2026-10-30. We let our AI planner run 12 helper agents at once. In the typical build of 47, about 1.3 were busy. To fill all 12, our six biggest builds needed new work 8 to 12 times faster. html at [/research/agent-fleet-occupancy](https://ezolabs.net/research/agent-fleet-occupancy). - [Let your lead agent only plan: fewer turns, same score](https://ezolabs.net/research/planner-only-swarms.md): published 2026-10-30. Over three paired runs, a lead agent that only plans took ×0.53 the turns (×0.69 without summary turns) and ×0.80 the prompt tokens, at 168/168. Total tokens did not fall. html at [/research/planner-only-swarms](https://ezolabs.net/research/planner-only-swarms). - [Your agent mostly sees one line of each skill. Write that line well.](https://ezolabs.net/research/skill-descriptions.md): published 2026-10-30. We moved four drawing and math instructions from an agent's system prompt into skills. For diagrams and math, the one-line description did most of the work. Graphs and molecules leaned on the page. html at [/research/skill-descriptions](https://ezolabs.net/research/skill-descriptions). - [inference](https://ezolabs.net/research/topic/inference.md): the research on inference, 2 posts. html at [/research/topic/inference](https://ezolabs.net/research/topic/inference). - [environments](https://ezolabs.net/research/topic/environments.md): the research on environments, 3 posts. html at [/research/topic/environments](https://ezolabs.net/research/topic/environments). - [evals](https://ezolabs.net/research/topic/evals.md): the research on evals, 1 post. html at [/research/topic/evals](https://ezolabs.net/research/topic/evals). ## Optional - [frame 00](https://web.archive.org/web/*/ezolabs.net*): the greeting the archive kept before the site went up, and every frame since. - [x](https://x.com/ezolabsinc): @ezolabsinc - [linkedin](https://www.linkedin.com/company/ezolabs): ezo labs - [hugging face](https://huggingface.co/ezolabs): ezo labs - [superbot on linkedin](https://www.linkedin.com/showcase/superbot-gg): superbot's own page