BONACCI FOUNDRY · MODEL SERVICES

The model runs on your GPUs.
We build the model.

Domain models delivered as weights you own outright. Built inside a network with no route to the internet, on your own hardware, or on our GPUs if nothing requires a boundary. From a corpus you supply, a few hundred examples we expand, or one we curate for you.

One week corpus audit first, free for now · full builds $15k to $120k · no corpus required to start · air-gap capable

Why this exists

A general model that has never seen your vocabulary

Every open checkpoint you might reach for was trained on public web text. It has never seen your product taxonomy, your regulatory vocabulary, your customers' languages, or your internal document formats. Running it on your own GPUs makes it private. It does not make it good at your job.

The hosted alternative is worse on both counts. The frontier labs will not train inside your network, they will not hand over weights at any price, and a general API on a repeated narrow task is the most expensive way to solve it.

Foundry closes the gap: a model that knows your domain, serving on hardware you own, with the weights in your hands and no licence attached to them.

vllm.internal.acme-bank.corpsame box
general checkpoint
> classify: "NOSTRO recon break, val dt T+2"
label: payment_issue
confidence: 0.41
tokens in: 31
private. does not know your vocabulary.
your domain model
> classify: "NOSTRO recon break, val dt T+2"
label: nostro_break.value_date
confidence: 0.96
tokens in: 14
private. trained on your ground truth.
When to build

Three reasons to commission a model

A custom model pays off when one of these is true.

Your vocabulary breaks general models

Domain jargon, regulatory language, internal taxonomies, and document formats that public web text never covered. Accuracy on your task, not on a benchmark.

Inference cost is a line item

A general frontier API on a repeated narrow task is the most expensive way to solve it. A small model trained on that one task runs at a fraction of the cost, on hardware you already own.

The data cannot leave the building

Training data is the most sensitive data you have. If a hosted fine-tuning API is closed to you for the same reasons hosted inference is, this is the only remaining path.

What goes in

You do not need a corpus to commission a model

Most teams think they don't have enough data. Usually they mean volume, and we can solve for volume.

You bring the corpus

Your documents, tickets, filings, logs, code. We measure duplication, PII, class balance and template collapse before a GPU hour is spent, and tell you in writing if what you hold is too thin to justify a build. This is the cheapest path and the one with the least risk in the eval.

You bring a few hundred examples

This is how most projects start. A small set of real cases—500 is enough—characterises your domain. We expand it into full training volume synthetically, grounded in the distribution of your seed set. Your examples set the target; we build the corpus. Fine-tuning, continued pretraining, even pretraining from scratch: all three rungs are available this way.

You bring nothing but the domain

We curate the corpus from public sources, filter it for licence terms, and build against an evaluation suite agreed with you before training starts. This is how our own models were trained, including the finance model published on Hugging Face.

Which path you are on changes the token budget and the timeline, and it is the first thing the audit week settles. It does not change what gets delivered: the same weights, the same eval harness, the same configs.

What we build

Most people who ask for a model from scratch should not buy one

Three rungs. Picking the right one usually saves you money.

01

Fine-tuning

You want new behaviour, format, or task performance.

Instruction tuning or LoRA on an open-weights base. Delivered as safetensors and GGUF, deployable on your existing vLLM or Ollama setup.

2 weeks
from $15,000
min 500 examples
02

Continued pretraining

You want new domain knowledge, not just new behaviour.

Corpus construction from your own data, continued pretraining on an open base, then post-training. Delivers most of the value of a from-scratch build at a fraction of the cost. This is the right answer for the large majority of teams who arrive asking for rung 3.

3 to 5 weeks
from $30,000
min 100M tokens or 500 seed examples
03

Pretraining from scratch

Only correct when one of these is true.
  • Your languages tokenize badly on existing vocabularies, making general models both expensive and inaccurate
  • Your data is not natural text: sequences, telemetry, logs, structured signals with no usable pretrained base
  • Every training token must be certifiable, so inheriting a model with an unauditable corpus is not permitted

Tokenizer design, corpus pipeline, architecture selection, pretraining, post-training, eval harness, technical report.

6 to 9 weeks
from $60,000
compute billed at cost plus 15%
min 10B tokens or 500 seed examples
Tell us which rung you need Book the corpus audit

The clock starts at data access, not at signature. The training run is rarely the long pole: a 9 billion parameter model over 122 billion tokens is days of wall clock on the right pod, so we compress by widening the pod and passing the compute through at cost, not by cutting evaluation. Corpus readiness is what actually moves the date.

Don't have a corpus? The "500 seed examples" path means we expand your examples synthetically into full training volume. You provide a few hundred real cases that define the domain; we build the rest. Every rung is available this way.

Where the build runs

Built for sealed networks first, available to everyone else

Built for isolated networks, but not limited to them. Where it runs is a compliance decision, not a product tier.

Inside the air gap

Offline install from images you mirror into your own registry, training on isolated hardware, delivery on physical media. No licence server, no telemetry, no callback anywhere in the path. Your firewall rules are the control, not our promises. Read the security page →

On your hardware

The run executes on the cluster in your data center or your cloud account, against storage you control, with our engineers driving it. The right answer when residency rules bind the data but the network is not actually sealed. Nothing for us to bill on compute, engineering time only.

On our GPUs

Not everyone who needs a domain model is under supervisory rules. If nothing about your problem requires an air gap, we do not charge you for one: you bring the domain, we bring the compute and the training stack. Compute is passed through at cost plus 15%.

Whichever of these applies, delivery is identical: safetensors and GGUF, the evaluation harness built on your ground truth, the corpus pipeline, and the training configuration. Owned outright, with no licence to renew.

Process

Scoped before it is sold

Every engagement starts with an audit that can end with us telling you not to build.

STEP 1

Scoping engagement

One week, and free of charge at the moment. We audit your data, assess tokenizer fit, estimate a token budget, and recommend a rung. You get a written report whether or not you continue. Roughly half of these conclude that retrieval or a fine-tune is enough, and we say so.

STEP 2

Corpus construction

Whether the corpus is yours, expanded from your seed examples, or curated by us, it is built as a versioned pipeline with contracts enforced at compile time: PII redaction verified, class balance checked, language mix asserted. Lineage is emitted, so you can answer which source rows produced which model version.

STEP 3

Training and evaluation

Runs on our GPUs, in your environment, in your cloud account, or on isolated hardware inside your air gap, whichever your compliance boundary requires. Evaluation is against a held-out set built from your ground truth, not public benchmarks.

STEP 4

Delivery

Weights you own outright, in safetensors and GGUF. Eval harness and training configuration handed over. Serve them on your own vLLM or Ollama, or anywhere else you like. The engagement ends here and nothing further is required of you.

Billing is by milestone: 30% on corpus delivery, 30% at the mid-training checkpoint with evaluation, 40% on final delivery. Compute is passed through at cost plus 15% and billed separately.

Depth

We have been to the bottom of this stack

Most vendors offering custom models are running someone else's fine-tuning API behind a project plan. We pretrained a 9 billion parameter mixture-of-experts model from scratch: 122 billion tokens, Google TPU v6e Trillium, MaxText and JAX. Tokenizer, corpus pipeline, architecture, training loop, evaluation.

We published the full technical report, including the benchmarks where it underperforms and the decisions we would make differently. That is the level of detail you should demand from anyone you let near your training data.

Read the technical report →  ·  See the architecture and training stages →

9B
parameters pretrained from scratch
122B
training tokens
TPU v6e
MaxText and JAX
Open
published report and eval harness
Ask first

Not ready for a call? Send us the details.

A model engineer reads every one of these and replies in writing, usually within two working days. No sequence, no nurture track.

  • 1
    The task you want the model to do, in one or two sentences.
  • 2
    Roughly what data you hold for it: volume, languages, and whether it is natural text. “Almost none” is a valid answer and does not end the conversation.
  • 3
    The constraint that rules out a hosted API, if there is one. If there is not, say so. Plenty of builds run on our GPUs.

Prefer email? Write to contact@thinkingdbx.com. If your question is really "which rung do I need", the audit week answers it faster.

Start here

Start with the corpus audit

Bring the task and whatever data you have. We'll tell you which rung you need—including when the answer is don't build.

Or email contact@thinkingdbx.com.

  • 1

    Weights

    safetensors and GGUF, owned outright, no licence to renew.

  • 2

    Eval harness

    Built on your ground truth, so you can re-run it on every future version.

  • 3

    Corpus pipeline

    Versioned, contract enforced, with lineage from source rows to model version.

  • 4

    Training configuration

    The exact configs and report, so another team could reproduce the run.