faxl

Sovereign AI

navigate the frontier

If you buy billions of tokens a day, it is time to manage your tokenomics directly: rent tin in your own VPC, run your own models behind one proxy, and own the cache.

Book a meeting

Why faxl

Are your AI costs and estate getting out of control?

Critical agentic workflows require enterprise-grade AI architecture. If your POCs proved agentic AI works, it's time to get serious about running it long term.

Cloud APIs make it impossible to manage costs at the caching and prefill layer, which could be 90% of your bill. They also expose you to foreign policy, pricing shocks, and an inability to manage AI audit and governance centrally. faxl offers you Sovereign AI: a product you run as an internal SaaS solution inside your VPC. It routes traffic, optimises tokenomics, adds governance, billing and adoption reporting, and does it through deep understanding of the hardware.

Book a call with us to find out how you can take control of your AI estate and optimise your tokenomics.

Enterprise VPC GPU compute · cloud or on-prem rent as needed Agent nightly batch Agent morning corpus query Agent long-running workflow prompts · infrequent, long gaps faxl proxy prefill cache owned · on cheap disk never invalidated warm resume from disk GPU ● on GPU ○ off GPU ○ deallocated or reconfigured
Everything stays inside your VPC. GPU compute — cloud or on-prem — is rented only when needed and resumes warm from disk you own.

Model control

Deliberate, secure and trusted

Downloading model weights from untrusted sources is risky. Will they write code that is deliberately vulnerable? This is not a new problem. Enterprises already use software like Artifactory or Nexus to manage distributed trusted copies of open source code, and faxl offers the same peace of mind for open weight AI models.

Don't delegate model management to public hubs where bad actors can operate. faxl's enterprise gateway includes a central vault for vetted models and their optimised versions, served from inside your network — reducing risk, bringing auditability, and reducing the cost of security review.

Model registry
Approved models and revisions, served from your estate — not pulled from a public hub at run time.
One address
An OpenAI-compatible endpoint for the whole enterprise. Clients change a URL, nothing else.
Cache prefill control
You set what is warmed and kept per model, rather than discovering the vendor's policy from your bill.

Enterprise controls

What a registry gives you, applied to inference

faxl is to prefill state and models what an artifact repository is to builds. These are the controls a platform team expects before it will put shared infrastructure into production.

Audit trail
Every request logged durably — prompt hash, prefix-hash array, lookup probes, cache decision, outcome — in deterministic order, and replayable.
Per-tenant telemetry
A canonical namespace over tenant, model, revision, backend, dtype and config is stamped on every record, so every request in the estate is attributable. It also stops state crossing a model or engine boundary, where reuse would be wrong.
Verified before use
Exact prefix-hash plus token-for-token confirmation before any resume. Sellers load on trust; faxl verifies.
Retention policy
You choose what persists and for how long. Correctness never depends on the choice.
Cost attribution
Tokens served, tokens skipped and bytes stored, per department, per model, per application. Invisible re-prefill becomes a line item someone owns.
Internal cross-charging
A monthly statement per team for AI spend on shared infrastructure, so each department pays for what it used.
Adoption analytics
Who is using AI, how much, and whether it is growing. Measured from the same record, not surveyed.
Survives churn
State persists off-GPU and restores across restart, deallocation and cross-node routing — the conditions where a vendor cache evaporates.
Long-gap workloads
A warmed context survives between runs. A frozen corpus hit once a morning, or a batch job that wakes and sleeps, keeps its cache across the idle gap.
Portability
The cache is an off-GPU blob keyed by namespace, not tied to a live process. Move a workload between clouds or on-prem and the warm state moves with it.
Air-gapped operation
No phone-home, no licence server, no managed dependency. Lift the whole thing onto sovereign bare metal and every control comes with it.

Roadmap

faxl is your gateway to Sovereign AI

A shared AI team inside a large organisation is trying to offer one secure, sovereign AI capability to thousands of staff and their agents. Everything that request touches — which model, which documents, what is remembered, what it cost — is decided at the proxy. That makes the proxy the service, not a component of it.

The direction of travel is a faxl interface your staff can use directly: the ergonomics people expect from a public chat assistant, serving your own models, with retrieval against your own corpora rather than an external service. Same gate, same audit, same attribution — and nothing leaves the estate.

Measured

What is proven today

  • Repeated prompt on Kimi Linear 48B: first token 5,056 ms → 246 ms, byte-identical output.
  • Tested on Granite 4.0-h, Kimi Linear 48B and Qwen3-Next-80B — chosen because published analysis infers Kimi K3 shares their hybrid recurrent design, the design that breaks ordinary prefix caching. What works on these works on K3 the day its weights land.
  • Eight-turn agentic coding session, warm store: 98% of prompt tokens skipped, whole session 4.4× faster (77.0 s → 17.6 s), 8/8 replies byte-identical — Qwen3.6-35B-A3B-4bit on a 128 GB MacBook Pro. Qwen3.8-Flash-Next-OptiQ-2bit: 98.1% skipped, 3.1×.
  • Exact re-run of a 40-task workload: 92% skipped.
  • 570/570 adversarial reuse decisions correct; no wrong or stale state served.
  • The NVIDIA/vLLM port passes the same byte-identity bars on GPU.

Results on your estate depend on your models, your traffic, and how much your prompts repeat. Bring your traffic profile and we will measure it.

Limits

faxl needs access to your GPUs

It sits in front of the inference engine, so it works with self-hosted vLLM on NVIDIA and MLX on Apple silicon. It doesn't work with external services from AI providers.

Talk to us

faxl is in alpha and improving daily. Bring your traffic profile and model mix, and we will show you what a warm cache takes off your bill, what it would take to run on your own tin, and how we can help you navigate this new AI frontier.

Book a meeting

Andrew Morgan, Gamakon Ltd — andrew@gamakon.ai