Build
Setups from native operators
Assembles simulation, geometry, USD and render setups from the DCC's own native operators, built and validated in one step. Your layout stays untouched.
Private beta · Q1 2027 · for SideFX Houdini
Houdini work, executed inside the scene.
Hardeen is an agent that builds, resolves and validates setups in your live Houdini session. Run it on your GPU, in Hardeen Cloud, or privately inside your studio.
Runs where your scenes liveValidated, not assumedNative operators, no opaque codeStudio models stay yoursNo per-token billsBuilt by artists, for artists
The bottleneck
Technical direction doesn't scale with headcount.
Every show runs on a few people who know how the pipeline actually works.
Standard setups get rebuilt on every show. Getting productive on a studio pipeline takes months. Pipeline knowledge lives in a handful of senior heads, and leaves with them between projects.
And cloud AI is off the table
Your scenes can't leave the building.
Client NDAs and security audits keep shots, assets and pipeline code away from public AI services. It's a hard rule, not a preference. Hardeen is built to run where your scenes already are.
How it works
Install. Connect. Instruct.
Hardeen is a desktop app that runs next to your DCC and operates on your open session through a bundled plugin. Artists keep working the way they already do.
01
Install
One installer sets up the app, the model runtime and the DCC plugin.
02
Connect
Open your DCC (Houdini in the beta) and start Hardeen. It attaches to the live session automatically.
03
Instruct
Describe the outcome. Hardeen reads the scene and executes it.
“Build a scatter on this grid.”
“Explain this pyro setup.”
“Validate this setup before caching.”
What it does
Pipeline work, executed in the scene.
200+ operations across the scene: geometry and attributes, simulations, USD stages, materials and lighting, caches, task graphs, rendering, animation and custom tools, each validated against the running session.
Inspect
Evidence, frame by frame
Reads geometry, attributes, simulation state, USD stages and caches frame by frame, so an issue is traced to the step where it starts, with evidence.
Explain
Pipeline knowledge, on call
Walks artists, contractors and partner studios through the scene in front of them, in the studio's own conventions.
Execute
The repetitive layer of the job
Caches, simulations, materials, lights and USD, executed deterministically while the artist keeps creative control. Grounded against trusted sources and information before executing.
On every task
Every step validated against the live scene.
Hardeen reads before it acts, changes one thing at a time and confirms the result in the running session. Even though it's a beta, it validates instead of assuming.
01
Read
Takes stock before acting: scene structure, cook state, geometry, attributes and the current selection.
02
Build
Makes one change at a time with native operators and exact parameter names, checked against verified references.
03
Validate
Re-cooks the affected part of the scene and checks the result frame by frame before reporting anything as done.
04
Report
What changed and why, with direct explanations, so the artist stays in charge of the scene.
Why a specialist
Inside the scene, not next to it.
General chat models answer from memory: errors get copy-pasted, scenes get described by hand, and parameter names get guessed.
General cloud AI
- Answers from memory
- General-purpose, guesses tool specifics
- Your scene leaves the building
- Per-token bills that grow with use
Hardeen
- Operates on your live scene
- A specialist model, trained on how VFX scenes are actually built
- Runs on your machines, or servers run for you alone
- Flat cost on hardware you choose
For artists
An agent inside your scene, on your own GPU.
The general Hardeen model runs locally and works on your live scene. No GPU to spare? Use it in Hardeen Cloud instead.
LocalBeta
Runs on your machine
Your scene never leaves your workstation.
- 16 GB VRAM
- Compact mode: lower precision, about 78k tokens of context
- 24 GB VRAM
- Recommended: higher precision for longer, more complex tasks
Hardeen CloudOpt-in
For machines without the GPU
The same agent on our servers, with access to larger Hardeen models as they ship. Usage limits apply.
- Requires
- Windows or Linux and SideFX Houdini. No local GPU needed.
Founding testers get
- Free access through the beta
- A say in what Hardeen learns next
- Credit as a founding tester
We ask for
- Real use on your own scenes
- Reports on what didn't work
- Two short feedback calls
For studios
A private model, trained on your pipeline.
General AI learned VFX from the public internet. It has never seen your tools, your naming conventions or the way your shows are built. Hardeen's training pipeline turns your approved setups into a private model that has.
On-premises
Inside your building
Runs on your workstations or your render farm, shared across the team. Nothing leaves your building.
Hardeen Private Cloud
Dedicated servers, yours alone
Your private model on dedicated GPUs we run for your studio only. Never shared with anyone else's workloads.
200+operations Hardeen runs in a live scene
Privatemodels trained only on your studio's setups, scoped to your agreement
0bytes of studio data on shared servers
2027first studio pilots, with design partners
01 · Scope
Map the pipeline
Your pipeline, your hardware, and where senior time goes today.
02 · Train
Setups become training data
Your approved setups, converted on your premises and under NDA.
03 · Deploy
On-prem or private cloud
On workstations, on the render farm, or on dedicated servers run for you.
04 · Retrain
Keeps pace with your shows
Retrained as conventions and projects change.
Design partners get
- Early access to a private model trained on your setups
- Direct influence on the roadmap
- A free pilot when studio pilots begin in 2027
We ask for
- Two or three of your artists in the beta
- A 30-minute call each month
Security
Built for confidential work.
Unreleased shots, director's cuts, client assets and pipeline code stay where they already are.
For studios
Studio data stays private
Studio models run on-premises or on dedicated Hardeen Private Cloud servers. Studio data never runs on shared Hardeen Cloud.
Your data trains only your model
Never used to train the general Hardeen model or anyone else's without a separate agreement. All under NDA.
You decide what's kept
After your model is built, training data is deleted, or kept on your own servers for future retraining. Your call.
No content in telemetry
Usage statistics are optional and content-free: no prompts, no scenes, no file names.
For artists
Local stays local
Running on your own GPU, your scene and your prompts never leave your machine.
Hardeen Cloud is opt-in
Your cloud sessions are never used to train the general model unless you choose to contribute them.
No token limits on your own hardware. No per-token bills for studios.
Roadmap
Houdini first. Then the rest of the pipeline.
What ships in the beta, and what comes after it. Everything in the right column is roadmap.
In the beta
- Live scene connectionOperates on the open session through a bundled plugin.
- Native TD workflowsValidated after each change. Your layout stays untouched.
- Grounded in verified referencesAnswers and lookups come from an indexed reference, far faster than searching manuals by hand.
- Hardeen's own modelRuns locally on your GPU.
- Moderate to long working sessionsSession length scales with your GPU. Optimised to handle moderate tasks even on upper mid-range consumer GPUs.
Next Roadmap
- Multi-step agentPlans a task, defines and verifies the workflow, validates and recovers on its own, with real workflow approaches.
- Hardeen Cloud and Private CloudLarger models for artists; dedicated servers for studios.
- Private studio modelsTrained on each studio's own setups.
- Viewport awarenessChecks results visually and reads reference images.
- Memory and studio knowledgeConventions, preferences and pipeline guidelines, as editable plain text.
Beyond Houdini Roadmap
HoudiniNukeCinema 4DUnreal EngineDaVinci Resolve
Houdini is the most technical tool in the pipeline, which makes it the right place to start. The order of the next tools is set with our design-partner studios.
- Status
- Private beta, Q1 2027
- Platform
- Windows, Linux
macOS planned - Works with
- SideFX Houdini
More DCCs on the roadmap - GPU (local)
- NVIDIA, 16 GB+ VRAM
24 GB recommended
Team
Dmytro Rohovyk
Co-founder · CEO
VFX and technical artist. Builds Hardeen's training pipeline and infrastructure.
Dmytro Kyrylov
Co-founder · CTO
Data engineer. Builds Hardeen's training pipeline and dataset, and manages its ETL/ELT pipelines.
Private beta · Q1 2027
Join the waitlist.
Artists get early access to the beta. Studios get a conversation about design partnership.