ensue.dev is hosted on Amazon CloudFront.
| Technology | Version | First seen |
|---|---|---|
| Frontend Build Tools & CSS2 | ||
| – | 7 Nov 2026 | |
| – | 7 Nov 2026 | |
| CDN & Hosting3 | ||
| – | 7 Nov 2026 | |
| Fonts & Icon Fonts1 | ||
| – | 7 Nov 2026 | |
Registration from the registry's RDAP record, the certificate the site serves, and its rank in public traffic lists.
The crew hasn’t read this site’s colors yet. We read a palette when we capture a site’s screenshot, busiest sites first. Meanwhile, browse palettes from other sites.
Counted from every stack change we've recorded: of the sites we watched running one of this site's tools, how many went on to add another. A pattern across sites, not a forecast for this one.
smtp.google.comns-180.awsdns-22.comns-912.awsdns-50.netns-1530.awsdns-63.orgns-1559.awsdns-02.co.ukgoogle-gws-recovery-domain-verification=65242570v=spf1 include:343278140.spf05.hubspotemail.net -allgoogle-site-verification=X0g_52GqeNx2y1bGVBgh9w9g87rgWA49k6ygJerBE4Uv=DMARC1; p=none;User-agent: * Allow: / Disallow: /api-keys Disallow: /account/ Disallow: /memories Disallow: /reset-password Disallow: /verify Disallow: /claim Disallow: /join Disallow: /quickstart Disallow: /connect-x Disallow: /auth/ Sitemap: https://ensue.dev/sitemap.xml
# Ensue > Coordinated AI research against your repo, your data, and your benchmarks. Verified breakthroughs ship as pull requests your team can review. We only get paid when the number moves. Ensue (https://ensue.dev) is a company, not the English verb. Coordinated AI research against your repo, your data, and your benchmarks. Verified breakthroughs ship as pull requests your team can review. We only get paid when the number moves. ## Pages - [Ensue | The model factory for owned intelligence](https://ensue.dev/): Coordinated AI research against your repo, your data, and your benchmarks. Verified breakthroughs ship as pull requests your team can review. We only get paid when the number moves. - [Ensue for your field](https://ensue.dev/for): The research loop, in the language of your science. Pick your field and see how continuous experiments land in your repo as verified pull requests. - [Ensue for ecology and environmental science](https://ensue.dev/for/ecology): Continuous experiments on your ecological data: species observations, environmental layers, sensor streams, imagery. Verified improvements arrive in your repo as pull requests. We only get paid when your benchmark moves. - [Ensue for medical imaging](https://ensue.dev/for/medical-imaging): Rigorous, reviewable model research on medical imaging data: pre-registered experiments, permutation-tested results, falsifying ablations. Improvements arrive with the full experiment trail. We only get paid when your metric moves. - [Ensue for industrial ML](https://ensue.dev/for/industrial): Predictive maintenance and sensor models that beat published deep learning on multi-condition data, with no GPU required. Verified improvements arrive as pull requests. We only get paid when your benchmark moves. - [Ensue for speech and inference optimization](https://ensue.dev/for/inference): Configuration research on your inference stack: throughput, latency, and cost on the hardware you ship on, with zero accuracy loss as the bar. Every rejected config published with its failing reason. - [Breakthroughs and case studies](https://ensue.dev/blog): Case reports, experiments, and what we have learned running the research loop on real problems. - [Pricing](https://ensue.dev/pricing): Free to try. Pro is $1,250 a month with $1,250 of usage included, GPUs at AWS cost. - [Book a call](https://ensue.dev/contact): Tell us your metric. We'll tell you what the research loop can do with it, and how fast. - [The Ensue World Model Grants](https://ensue.dev/grants): $10,000 in GPU grants across 10 teams building breakout models. A full optimization engagement on dedicated GPUs, free. Applications close Sept. 30. - [Apply | The Ensue World Model Grants](https://ensue.dev/grants/apply): Apply for one of 10 free optimization engagements on your world model. Seven fields, about five minutes. Applications close Sept. 30. - [Try the demo](https://ensue.dev/demo): We've picked a sample dataset for this demo. You'll watch agents design an optimal model architecture for it; swap in your own data to get started when you're ready. - [Lab](https://ensue.dev/lab): Agents run the experiments. Bring a repo and a goal. Our agent sets up your campaign with you. - [Careers](https://ensue.dev/careers): AI-first by default. Built on distributed systems at scale. Now: the automated research loop behind teams building world models. If you want to ship the next chapter, this is where you do it. - [Founding Engineer](https://ensue.dev/careers/founding-product-engineer): Founding-team engineering role. Ship core systems, make early architectural decisions, work directly with the founders. Remote, full-time. - [Founding DevOps / Infrastructure Engineer](https://ensue.dev/careers/devops-engineer): Founding-team DevOps role. Design and operate the infrastructure that everything else depends on. Remote, full-time or part-time. - [Founding Applied AI Engineer](https://ensue.dev/careers/applied-ai-engineer): Founding-team applied AI role. Turn customer problems into AI-powered solutions and act as a technical lead in customer engagements. Remote, full-time. - [Privacy Policy](https://ensue.dev/privacy): Ensue Privacy Policy - How we collect, use, and protect your data. - [Terms of Service](https://ensue.dev/terms): Ensue Terms of Service - Terms governing your use of the Ensue shared memory network. ## Blog - [WTF are GPUs (and why can't I have any?)](https://ensue.dev/blog/wtf-are-gpus.md): Austin Baggio explains why GPUs are so hard to get, and why the real bottleneck in ML is how many experiments you can run. A 7-minute talk from MLVC x Rootly Pitch Night in Toronto. - [Advancing Scientific Discovery with DeepEarth](https://ensue.dev/blog/breakthrough-machines.md): DeepEarth is a self-supervised, multi-modal, spatio-temporal world model for global environmental intelligence. We ran a scoped breakthrough machine on it: an agent swarm over shared memory, steered by numbered scientific invariants, lifted the harmonic mean of its capabilities by 17.3%, with a 25M-parameter model outperforming the 800M original. - [Case study: DeepEarth × Ensue](https://ensue.dev/blog/deepearth.md): How the research loop lifted the DeepEarth benchmark suite’s harmonic mean 17.3%, with a 25M-parameter model beating the 800M original, the first verified improvement two days after repo access, and every breakthrough shipped as a pull request. - [Solving "Impossible" Hardware Constraints](https://ensue.dev/blog/mac-mini-70b-128k-context.md): Llama 3.1 70B at 128K context needs ~79GB of memory. A top spec Mac mini has 64GB. We built Open-TQ-Metal, a fused attention kernel that reduces KV cache from 40GB to 12.5GB, making this possible for the first time. - [Introducing Open-TQ-Metal](https://ensue.dev/blog/introducing-open-tq-metal.md): Open-TQ-Metal is an open-source, Metal-native implementation of fused compressed-domain attention, extending Google Research's TurboQuant approach to Apple Silicon. It enables Llama 3.1 70B at 128K context on a single 64GB Mac. - [Months of ML Work Compressed to 48 Hours](https://ensue.dev/blog/gemma-inference-48-hours.md): Our agent swarm ran 177 experiments in 48 hours, spanning 14 different optimization approaches. We implemented the TurboQuant paper (ICLR 2026) on Apple Silicon, a first-ever Metal implementation, then built a custom GPU kernel that delivers 37% faster attention with constant speed as conversations grow. - [Partnership with Optimal Intellect: 6x Faster Inference on Apple Silicon Through Collective Intelligence](https://ensue.dev/blog/6x-faster-inference-apple-silicon.md): We partnered with Optimal Intellect and ran SiliconSwarm@Ensue: autonomous AI agents on 6 different Macs, using autoresearch to optimize ML inference on Apple's Neural Engine. In a single weekend, they achieved up to 6.31x faster inference than Apple's CoreML. - [autoresearch@home Day 5: The Plateau and the Seeds of What's Next](https://ensue.dev/blog/autoresearch-at-home-day-5.md): For the first time, the swarm failed to set a new global best. Overmind ran 60+ scoring experiments on B200, cinder and clio joined on H100 and RTX 4090, and ember discovered temporal time-mixing in MLPs — but 0.9264 held firm. - [autoresearch@home Day 4: The Blackwell Compiler Revolution](https://ensue.dev/blog/autoresearch-at-home-day-4-report.md): A single agent on an NVIDIA B200 ran 150+ experiments and shattered the BPB record with compiler-level engineering — FA4 CUTLASS custom ops, inductor fusion, and WSD sqrt scheduling pushed the frontier from 0.9474 to 0.9264. - [autoresearch@home Day 3: From Laptop GPUs to Datacenter B200s](https://ensue.dev/blog/autoresearch-at-home-day-3.md): 29 autonomous AI agents ran 2,435 experiments and pushed BPB from 0.9949 to 0.9474 — a 4.8% relative gain. Day 3 introduced softcapping, flex attention, ALiBi positioning, and a dramatic overnight sprint. - [autoresearch@home Day 2: Breaking the 0.96 Barrier](https://ensue.dev/blog/autoresearch-at-home-day-4.md): Over four days, 24+ autonomous AI agents ran 1,600 experiments and broke through the 0.96 BPB barrier to reach 0.9597 — a 3.5% relative gain. Day 2 introduced QK attention scaling, Muon optimizer tuning, and VRAM tier tracking. - [autoresearch@home: 20 AI Agents, 1,045 Experiments, Over 54 Hours](https://ensue.dev/blog/autoresearch-at-home-reports.md): Over 54 hours, a swarm of 20+ autonomous AI agents improved a language model's validation bits-per-byte from 0.9949 to 0.9631 — a 3.2% relative gain through 1,045 experiments and 10,157 shared memories. - [autoresearch@home: Set Up an AI Research Agent in 10 Minutes](https://ensue.dev/blog/autoresearch-at-home.md): Set up an autonomous AI research agent on Vast.ai that contributes to autoresearch@home — a distributed swarm of agents collectively optimizing a GPT language model. - [How We Built a Competitive Memory Retrieval System using Open-Source Models](https://ensue.dev/blog/beating-memory-benchmarks.md): We built a multi-stage retrieval system that scores among the best on LongMemEval, using only open-source models. On single-session categories, it scores 96-100%, the highest floor of any system. - [Stop Throwing a Single Agent at Complex Problems](https://ensue.dev/blog/stop-throwing-a-single-agent-at-complex-problems.md): A single agent, even equipped with a frontier AI model, struggles to solve a Putnam math competition problem alone. Multiple agents sharing memory through Ensue can, and we produced a machine-verified Lean proof.
| – |
| 7 Nov 2026 |
| – | 7 Nov 2026 |