We are not yet selling a product. What we are doing — for a small, deliberately chosen group of operators across four verticals — is co-developing the BULUT-1 model line with the people who will eventually use it. This is the brief. If it sounds like you, write to us.
If we sell you a model in 2026, you will be disappointed. If we build a model with you across 2026 and 2027, you will be ahead of every competitor when 2028 arrives.
Continuous video understanding of factory floors, ports and energy infrastructure. Anomaly detection, predictive maintenance, OSHA-grade incident reconstruction with full causal chain of attention. The discipline of the operator who runs this kind of system is a perfect match for ours.
Ideal partner: a mid-size industrial operator running 5–50 sites with existing camera infrastructure and a small data-science team.
Closed-loop simulation, edge-case mining and behavioural prediction for L4 trucking and last-mile robotics. We do not build vehicles — we believe the stacks that go into them deserve smarter eyes than a per-task CNN ensemble. The physics of long-tail edge cases is exactly where foundation models earn their keep.
Ideal partner: a regional fleet or trucking operator with at least one year of vehicle-side telemetry and an in-house ML team.
Joint training across satellite, drone and ground-station data to nowcast wildfire spread, urban flooding and methane plumes at sub-kilometer resolution. The bottleneck for ground-truth-augmented EO models has always been ground data — we are building it. Civil-protection agencies and EO research groups especially welcome.
Ideal partner: a climate or civil-protection programme with EO data access and a willingness to co-publish.
Generalist manipulation and locomotion policies fine-tuned from BULUT for humanoid, quadruped and arm-robot OEMs. We will need a hardware partner who tolerates an early-stage foundation backbone and can co-develop the action-codification we need. Likely the last vertical we activate.
Ideal partner: a robotics OEM with a working hardware platform and a small but skilled embodied-AI team.
| Criterion | Why it matters | How we will assess |
|---|---|---|
| Operates real physical infrastructure | If your business does not depend on the physical world running, we are not the right fit. | first call · 30 min |
| Has a small, sharp technical team | We work with engineers, not procurement. The technical contact is the partnership. | technical conversation |
| Comfortable with research-velocity timelines | Twelve months is the minimum useful horizon. Demo-by-Friday energy will burn out fast. | honest expectation-setting |
| Open to data sharing under provenance discipline | We can only co-build if we can train on your domain data — under our consent ledger. | legal review jointly |
| Willing to co-publish if appropriate | We will not announce shipping. We may want to publish what we learnt — with you. | case-by-case |
| Mission-aligned | We will not partner with operators whose use of physical AI we would be ashamed of. | conscience-based · two-way veto |
You email partners@bulutlabs.com with a paragraph on what you operate and why physical AI matters in your context. We reply within 48 hours. We schedule a 30-minute call. Either side can say no without explanation.
One with your technical lead, one with our founder. We discuss what data you can share, what evaluation harness would matter to you, and what your team's research velocity looks like. We are honest about what BULUT-0 can and cannot do today.
We share our consent ledger architecture and walk through how your data would be ingested, redacted, trained on and audited. Most partnerships fall apart here, which is good. The ones that survive this are durable.
We sign a short, plain-English MoU outlining scope, data access, evaluation milestones, mutual rights and exit clauses. No cash changes hands unless your team's time costs you more than data-flow plumbing.
Quarterly reviews with you. Weekly stand-ups with your technical lead. Open Slack with our research team. Honest mid-year retrospective. Renewal conversation in month 11. Public co-published learnings if mutually useful.
We do not partner with operators whose primary use case is offensive military or surveillance.
Even with consent ledgers, we will not train on populations that did not meaningfully agree.
If your goal is to acqui-hire us before BULUT-1 ships, we are not the right fit. We are a five-year-patient company.
We do not run AI strategy workshops. We build models. We are happy to recommend people who do consulting.
We do not white-label. We do not embed silently. Our partners' presence in our customer list is part of how we earn trust.
If we would be embarrassed to put the partnership on the website in a year, we should not start it now.
Email a paragraph on what you operate and why physical AI matters to you. Replies in 48 hours, in three languages, from the founder.