KaijuTank Labs
Products
Eleven products and research projects we build and run ourselves. Each one is listed with what it does, where it honestly stands today, and how you can get it.
CloudLab
Cloud delivery and handoff platform
CloudLab takes a cloud system from a developer's machine to production with every step recorded: what was validated, who approved it, and what the next team needs to run it.
- Local-first validation before anything touches a shared environment
- Promotion from local to development, sandbox and production through explicit approval gates
- Releases with their promotion and rollback history
- An architecture canvas and component inventory; a change goes through a reviewed draft
- Security controls, policy checks and findings, with monitoring, logs and alerts beside them
- A traceability ledger, an audit timeline and access roles
- Infrastructure as code with Terraform and Packer: AWS, with adapters for Azure and Google Cloud
- Starts from a Jira ticket and a gated Jenkins pipeline; works with GitHub, GitLab, plain Git and Perforce
- An evidence-aware handoff package: infrastructure code, runbooks and operating notes
- Browser experience built; the Windows and Linux app is in final testing
- Who it is for
- Teams that inherit, hand over or modernise production cloud systems.
- How to get it
- Available now through the CloudLab Delivery Assessment. Ask for a walkthrough of the current build.
Aether
Our fine-tuned language model
Aether is a language model we fine-tune, evaluate and release ourselves. It is a model, not an app: it plans, calls tools and returns structured output inside whatever product uses it, and it runs entirely on hardware you control.
- Runs on a single 8 GB consumer GPU, with no per-token cloud cost
- Persistent memory with combined keyword and semantic search
- Planning loops, structured output and tool orchestration
- Vision, voice and image generation in the same system
- Who it is for
- Organisations that need private AI workflows where data cannot leave the building.
- How to get it
- Delivered today as part of an AI implementation sprint: we deploy and tune it on your hardware. Standalone early access is by request.
- Base model
- Fine-tune on our own data
- Evaluation and quality gates
- Release a version
- Served locally by Silo
Silo
Inference engine for small GPUs
Silo is a Rust engine for serving language models on limited hardware. It is where we measure what a small GPU can really do, and it serves Aether in our own lab.
- Hardware-aware execution planning: latency, throughput and memory budgets
- Verified streaming compression with a stable container format
- KV-cache planning and CUDA work for low-VRAM machines
- Benchmarked under sustained load, not only in short bursts
- Who it is for
- Engineering teams evaluating local inference on constrained hardware.
- How to get it
- Not a commercial product. We share findings and run technical briefings, and the work feeds directly into client deployments.
Cinder
AI-native browser and creative studio
Cinder is a desktop workspace that brings research, building, automation and media creation into one application, running on local models wherever possible.
- A private browser with research and automation built in
- Image, video, audio and 3D creation in one workspace
- Local speech recognition and local language models
- Skills that package repeatable production workflows
- Who it is for
- Creators and small teams who want one private workspace instead of a dozen cloud subscriptions.
- How to get it
- In development. Join the early access list and we will contact you when a build is ready to try.
Frameward
Production platform for film, VFX, animation and games
Frameward connects production tracking to the things tracking tools usually ignore: where an asset came from, which build it shipped in, and whether it passed review.
- Configurable tracking with immutable publishing and asset lineage
- Colour-managed review with playlists, annotations and approvals
- Perforce, Unreal, Unity, Blender and Godot integrations
- The core control plane is implemented and tested; later phases are in progress
- Who it is for
- Studios that have outgrown spreadsheets and want tracking tied to source control and builds.
- How to get it
- In development. We are looking for a small number of design-partner studios to shape the next phase.
Borealis
Unreal Engine AI agent, in the editor and in VS Code
Borealis turns a plain-language request into a reviewed plan and then into Blueprints, C++ and assets inside Unreal Engine. You approve the plan before anything changes, and every step can be undone.
- Ask mode for questions about your project; Agent mode to plan and build
- Plan, dry run, approve, execute, undo: nothing changes without your say
- Routes work across local and cloud models, with a free local option
- Asset pipeline from concept to import, plus a headless mode for CI
- Who it is for
- Unreal Engine developers and studios who want automation they can inspect and control.
- How to get it
- Private preview. The editor plug-ins are being rebuilt for current engine versions; a free local tier is planned alongside paid studio plans.
KaijuCode
Local-first agentic developer runtime
KaijuCode is a coding assistant that runs against your own models. It reads and edits files, runs commands in a sandbox and keeps long tasks on track, without sending your code to a third party.
- Multi-step tool loop with sandboxed file and shell access
- Connects to any MCP server it is pointed at
- Routes each task to the model that has performed best on that kind of work
- Works with any OpenAI-compatible local server
- Who it is for
- Developers and teams whose code cannot leave their machines.
- How to get it
- Private preview by request.
Runtime QA Agent
Software QA that operates the real application
A Rust runtime that drives real desktop applications and games the way a tester would, records everything it saw and did, and files findings with the evidence attached.
- Observes through Windows UI Automation and screenshots; acts with keyboard and mouse
- Every run can be replayed step for step
- Detects crashes, errors, stuck screens and failed assertions
- Optional Jira reporting with duplicate suppression
- Who it is for
- Teams shipping desktop software or games who need repeatable runtime testing.
- How to get it
- Offered as a QA automation engagement: we set it up against your application and hand over the runs.
- Observe
- Act
- Record outcome
- Detect finding
- Report with evidence
Interview Coach
Desktop coach for technical interviews
A small desktop overlay for practice and mock interviews. It follows the problem, the code and the conversation, and gives hints, solutions and review according to the role and mode you set.
- Candidate, coach, interviewer and reviewer roles with separate views
- Practice and mock modes with a central policy on what help is allowed
- Local speech recognition and local models
- Capture and transcription are disclosed to participants by design
- Who it is for
- Engineers preparing for interviews, and coaches who run mock sessions.
- How to get it
- In development. Join the early access list.
EdgeLab
Evidence-first strategy research
EdgeLab turns a trading hypothesis into a backtest, an adversarial review, walk-forward validation and an audit report. It is a research instrument: it makes no profit claims and cannot place a live order.
- Backtests with explicit transaction costs and slippage
- Walk-forward validation and an 81-point parameter sensitivity sweep
- A significance hurdle that rises with every variant tried on the same data
- Paper trading only, and only when explicitly enabled
- Who it is for
- An internal research tool, shown here as an example of how we build evidence into a system.
- How to get it
- Not offered as a product or a service, and nothing here is investment advice.
- Hypothesis
- Backtest
- Adversarial review
- Walk-forward
- Audit report
Kaiju Engine
Engine and editor research
An early engine project: a Rust core with an editor built on web technology. Scripts in several languages compile to sandboxed modules, and a stable C interface lets other languages drive the engine directly.
- Rust core with a stable C interface
- Editor built with web technology, with starter templates
- Scripts compile to sandboxed modules that run on every platform
- Rollback networking for deterministic games
- Who it is for
- Our own research into editor and engine tooling.
- How to get it
- Not available. We publish notes as the work matures.
- Rust core
- C interface
- Script modules
- Web editor
- Templates
Access and pricing
How to work with us today
The products above reach clients through engagements while they mature. These are the three ways to start.
Assessment
Start hereCloudLab Delivery Assessment
Fixed fee Rates on request, quoted in writing
You want to know where your cloud delivery is at risk before you change it.
- Review of how the system is built, tested, released and handed over
- Written findings with risks ranked
- Cloud readiness and CI/CD risk report
- A modernisation roadmap you can act on with or without us
- Scope and fee agreed in writing before work starts
Build
AI / Cloud Implementation Sprint
Per sprint Rates on request, quoted in writing
You have a defined thing to build and want it delivered, tested and documented.
- One defined deliverable: a pipeline, service, model deployment or product milestone
- Acceptance checks agreed up front
- Delivered with tests and documentation
- Built with our own tooling: CloudLab, Aether, Runtime QA Agent
- Runbooks so your team can own it afterwards
Ongoing partner
Fractional Principal Engineering
Monthly Rates on request, quoted in writing
You need senior hands on architecture and delivery without hiring a full-time lead.
- Architecture decisions and design review
- Hard debugging alongside your engineers
- Delivery support from prototype to production
- One point of contact, no account managers
- Month to month
We send our rate sheet on request: ask for it by email. Every engagement is quoted in writing before any work begins. Product access (previews, early access, design partnerships) is by request, and product pricing will be published when each one is generally available. See what you can ask for on each product.
