One platform. One team that delivers it.
FlatClaw is built and delivered by Kirk Tech Solutions. Kirk stands the platform up in your own cloud tenancy, builds the skills, connectors and agents your work needs, and runs it with you. No resellers, no hand-offs, one team from the first call to production.
Everything between the repository and a platform in production.
The open-source platform is free to pull and audit. Kirk turns it into your platform: deployed, integrated, taught your work, and kept running.
Deployment in your tenancy
The platform stood up inside your Azure, AWS or Google Cloud account, your Northflank project, or your own hardware: identity, networking, GPU inference, backups, and the privacy proof run on your egress.
Estimating and quoting engines
Takeoff from drawings and specifications, reconciliation against your proposals, pricing on your rules, benchmarks from your job history, and proposals in your house format.
Data lakehouses and reporting
Governed lakehouses over your ERPs and CRM with lineage to the source transaction, consolidated financials, forecast and pipeline by business unit, and plain-English inquiry on top.
Voice agents
Two-way phone agents on your own lines, with speech, reasoning and recordings inside your tenancy, sized for the concurrency your call volume actually needs.
Custom skills and MCP integrations
Model Context Protocol servers that expose your systems to agents safely, per-user credentials, approval gates on consequential actions, and skills authored for the work your teams do.
Managed operations and support
Certified releases and security patches on a cadence you approve, monitoring, backups, capacity, an audit pack, and a named team with business-hours response.
Fixed scope, value in weeks, milestones paid on acceptance.
Every engagement follows the same shape, from a two-person team to a multi-brand group. Pricing is fixed per phase and quoted after discovery. Nothing is metered per token or per seat.
- 01
Discovery
Two to three weeks, fixed scopeWe map your systems, identity, data and the first workflows, and come back with a priced plan. Where an unknown cannot be priced responsibly, discovery is where it gets retired.
- 02
Phase 0: a working deployment
About six weeksThe platform stands up in your tenancy on your cloud. First users sign in with your identity provider, the first connectors and skills go live, and you see value before the larger build starts.
- 03
Phased build, paid on acceptance
Monthly milestonesEach milestone is scoped with your team and accepted against agreed outcomes before it is billed. Value first, finesse last.
- 04
Run
Annual platform subscriptionPer inference tenant: certified releases, security patches, monitoring, backups and support. Your cloud bills you directly for the infrastructure. We never sit in that bill.
- Your cloud, your account, your bill for the infrastructure.
- Your data and the model never leave the tenancy; the proof runs on your egress.
- Every line of the platform is open source and yours to read.
Your cloud, our delivery.
One image, one control plane, one privacy proof, on whichever cloud you already run. Northflank is the reference lane the platform is built and verified on; Azure, AWS and Google Cloud tenancies are stood up by Kirk in your account.
Microsoft AzureThe lane for Microsoft-first organizations.
A resource group inside the customer's own subscription: AKS or a single GPU virtual machine (NC H100 v5 class), Entra ID for sign-in, private endpoints for everything, and outputs into Fabric and Power BI when the customer wants them.
- Entra ID sign-in
- NC H100 v5 GPU nodes
- Fabric / Power BI hand-off
Your account, your VPC, no public inference endpoint.
An account and VPC the customer owns: EKS or a single GPU instance (p5 or g6e class), IAM roles for access, KMS for secrets, and egress limited to the services users explicitly connect.
- IAM + KMS
- p5 / g6e GPU instances
- Private VPC egress only
A project the customer owns, fenced with VPC Service Controls.
GKE with A3 (H100) nodes or a single GPU VM inside the customer's project, Workload Identity for the services, and VPC Service Controls around the tenancy so nothing crosses the boundary unnoticed.
- Workload Identity
- A3 (H100) nodes
- VPC Service Controls
NorthflankThe fastest path from zero to a running tenant.
The managed-GPU platform FlatClaw was built and verified on. One project per tenant, H100 plans by the hour, and the lane scripts in the repository bring inference up and down with one command. Every release is proven here first.
- One project per tenant
- Managed H100 plans
- Lane scripts in the repo
For data that cannot be in any cloud at all.
Bare metal or a private Kubernetes cluster in your building: an NVIDIA H100 or RTX PRO 6000-class card, the same containers, the same image. No cloud account, no cloud bill, no cloud dependency.
- H100 or RTX PRO 6000 class
- Private Kubernetes or a single host
- Zero cloud dependency
Microsoft Azure, Amazon Web Services, Google Cloud and Northflank are trademarks of their respective owners. Logos identify supported deployment targets; no endorsement is implied.
The part that does not change per cloud.
The same public inference image from GHCR runs on every cloud. Per-tenant differences live on the weights volume and in the tenancy's secrets, never in the image.
A resource group, an account, a project, or a rack — one per customer, with the control plane and the GPU inside it. No shared state across tenants.
The cloud bills the customer directly. We never sit between a customer and their substrate, and we never touch the bill or the data.
Run tcpdump on the tenancy's egress for a full session. Zero packets to any third-party inference endpoint, on any of these clouds.
Kirk Tech Solutions
The company behind FlatClaw
Kirk Tech Solutions builds custom AI applications and Model Context Protocol integrations for organizations whose data cannot leave their own infrastructure. FlatClaw is the platform that came out of that work, released as open source and delivered by the team that wrote it.
For each deployment Kirk handles the end-to-end: discovery, the tenancy on your cloud, skill design and implementation, MCP server authoring, role and approval policy, integration with the systems you already run, and the operations that keep it in production.
Thirty minutes: your workflow, your cloud, and a plan.
Tell us the process nobody likes doing and where the data has to stay. We will tell you what Phase 0 looks like.