ColibriV Design AI · Engineering agent

Set the goal tonight. Get verified designs by morning.

ColibriV Design AI is an engineering agent for physical hardware. You set the goal; it explores the design space overnight, runs real physics and hands you ranked designs with a full evidence trail.

We built it to design our own hydrogen engine. Now we're opening it to hardware teams in other sectors.

COLD PLATE · TEMPERATURE
COMBUSTOR · FLAME
ILLUSTRATIVE · LOWER-LEFT IS BETTER

Goal tonight. Designs by morning.

  1. 01 · Evening

    Set the goal

    Describe what you want in plain words. Design AI turns it into objectives and constraints, and you confirm each one.

    “lower cold-plate thermal resistance 15% at the same pressure drop”“cut burner NOx with 30% hydrogen blend”“lighter bracket, same stiffness”
  2. 02 · Overnight

    Explore overnight

    It builds parametric designs, screens thousands of them with fast physics models, and picks the most promising and the most uncertain to test further.

  3. 03 · Overnight

    Verify with real physics

    The best candidates run on real solvers: flow, heat transfer, chemistry and structures, with mesh and convergence checks.

  4. 04 · Morning

    Decide in the morning

    You get ranked designs, trade-off charts, CAD you can open, and an evidence pack that shows where every number came from.

Built for your sector.

Design AI uses one engine for every sector. What changes is the sector pack: the parametric designs, physics setups, quantities that matter and validation data for your kind of hardware.

Design partners · 2027

Thermal management

Cold plates, heat sinks, coolant manifolds and heat-exchanger cores for data centers, AI servers and power electronics.

Typical goalLower thermal resistance at the same pressure drop, within your manufacturing limits.

Design partners · 2027

Industrial combustion and clean heat

Hydrogen-ready and hydrogen-blend burners, boiler furnace sections and gas-turbine combustors.

Typical goalLower NOx and safe flashback margin as the hydrogen share goes up.

Design partners · 2027

Structures

Brackets, mounts and frame nodes for drones, eVTOL, robotics and industrial equipment.

Typical goalLess mass at the same stiffness, strength and fatigue life.

Planned · late 2027

Battery and EV thermal

Cooling plates and module thermal design for EV, eVTOL and energy-storage packs.

Typical goalMore uniform cell temperature with less coolant flow.

Planned · late 2027

Fluid machinery

Pump impellers, fans and valves, including coolant pumps for liquid-cooled racks.

Typical goalHigher efficiency across the operating range.

Controlled deployments planned

Aerospace propulsion

Combustors, fuel injectors, nozzles and cooling channels. This is where Design AI started.

Before you share dataContact us before sharing any controlled data.

Don't see your sector? Tell us about your problem when you apply. Partners will be able to build their own sector packs on our toolkit.

Why it’s different

The language model plans the work. It never invents the numbers.

Verified, not guessed

The language model plans the work; it never invents the numbers. Every result you see has passed a real solver run, and you can trace it to that run.

Evidence you can show a customer or regulator

Every requirement, setup, result and decision is logged in a tamper-evident record. Export it as an evidence pack.

Hardened on one of the hardest problems in engineering

Design AI was built to design hydrogen combustors, where flow, heat, chemistry and safety margins all interact. The same rigor carries over to every sector we support.

Your data stays yours

Your designs and results are isolated from every other customer. We don’t train shared models on your data unless you opt in.

Capa- bilities

Design
Parametric CAD for each sector; STEP import with guided parametrization 2027
Physics
Flow (CFD), heat transfer, chemistry and structures (FEA), from quick estimates to full solver runs
Search
Space-filling sampling, Bayesian optimization, multi-objective evolutionary search
Physics models
Fast learned models with uncertainty, retrained as results come in
Manufacturability
Design rules for your process, such as minimum fin thickness or wall thickness
Evidence
Tamper-evident record; exportable evidence packs
Access
Web studio, API, command line and an MCP server for AI agents 2027

Who it’s for

  • Thermal
  • Combustion
  • Structures
  • Batteries
  • Flow

Hardware teams of 5 to 200 engineers that need more design studies than their simulation capacity allows.

Teams we build for

  • Thermal hardware makers
  • Burner and boiler OEMs
  • Battery and EV integrators
  • Drone and robotics companies
  • Pump and fan makers
  • Aerospace and energy startups

Design partner program 2027

Shape Design AI with us.

We're selecting a small group of teams, across several sectors, to shape Design AI with us.

What you get

  • Dedicated design studies on a real engineering problem of yours
  • A direct line to our engineering team
  • Founding-partner pricing

What we ask

  • A real problem with clear goals and constraints
  • Regular feedback calls
  • Permission to publish a case study (anonymized if you prefer)

Honest limits

Design AI is an engineering tool. Its results are advisory engineering evidence, not certified analysis. Every result it reports is confirmed by a physics solver, and final decisions stay with your engineers. Sector availability above reflects current plans and may change.

FAQ

Which sectors can I use it for today?

Design partners in 2027 start with thermal management, industrial combustion and structures. Battery and fluid machinery follow later in 2027. If your problem fits none of these, apply anyway and tell us about it.

Is my data used to train your models?

Not by default. Your data trains only models for your own organization. Sharing with our general models is opt-in.

Which solvers does it use?

Validated open-source solvers for flow, heat transfer, chemistry and structures. Running your own licensed commercial solvers is planned for private deployments.

Does it replace my CAD or simulation tools?

No. It runs studies your team doesn’t have time to run, exports CAD you can open in your tools, and shows every solver setup so your analysts can check it.

Can it run in our own cloud?

Private deployment is on our roadmap. Tell us your requirements when you apply.

Can it handle export-controlled work?

Controlled deployments are planned. Contact us before sharing any controlled data.

What does it cost?

Design partners get founding pricing. After the program, we offer fixed-fee design studies and team subscriptions. Contact us for details.

Bring us your hardest design problem.

Thermal, combustion, structures, batteries or flow: if it can be simulated, we want to see it.