We train domain models you own, build the self-hosted data platform that feeds them, and run AI teammates on your business data from our cloud. Inside your air gap when the data cannot move, on our infrastructure when it can.
Weights delivered, not licensed
Air-gap capable (Foundry, Studio)
Open weights on Hugging Face
Use one layer or all three. Each works on its own.
Domain models trained inside your perimeter, delivered as weights you own outright. A few hundred examples is enough to start.
An agentic data platform that ships pipeline code to your Spark and Kafka and serves your models on your own Ollama or vLLM. Self-hosted, air-gap capable.
AI teammates that query your databases, check every number, and ask before changing anything. Hosted on our cloud, so there is nothing to install: connect a database and give a kin its first job.
Pretrained from scratch on our own GPUs. Every model ships with evals and a written account of where it fails.
Grounded financial analysis. Reads live market data from context and quotes it exactly, or tells you it does not have the number. Pretrained on SEC EDGAR filings and financial news. 9B parameters with 2.6B active, Apache-2.0, commercial use permitted.
weights, evals and limits →Search a codebase by describing what the code does, in plain English. Pretrained from scratch on 3.97B tokens of permissively licensed code, with every licence decision and every redaction written down. 110M parameters, runs on a CPU, Apache‑2.0.
weights, evals and limits →Analyzes COBOL programs to identify files, copybooks, data fields, called programs, and dead code. Trained on real and generated COBOL from The Stack. 1.5B parameters, Apache-2.0, commercial use permitted.
weights, evals and limits →An analyst drops a quarterly filing into context and asks for the segment margin: Kautilyaa returns the figure and the line it came from, or says it does not have that number. Across 150 test cases it produced no invented figures at all.
On Kautilyaa we traced a silent optimizer bug that had been degrading every run: weight decay applied to normalisation gains, invisible in the loss curve. Catching that class of failure before your compute is spent is what a Foundry engagement is for.
# checked before a single GPU hour is billed P1 weight decay excludes norm gains BLOCK exclusion list is empty: every parameter decays, including the RMSNorm gains. invisible in loss. P2 checkpoint interval fits tenure BLOCK 10000 steps on spot capacity: every preemption restarts at zero P15 licence permits delivery terms pass P16 eval sets decontaminated pass P17 tokenizer fertility measured warn # 2 blocking. run not started.
Built for sealed networks first, available to everyone else.
Foundry and Studio. Offline install from mirrored images, no licence server, no callbacks. Your firewall rules are the control, not our promises.
Foundry and Studio. On your GPUs, in your data center or your VPC, with our engineers driving the work. For when residency rules bind the data but the network is not sealed.
Foundry builds on our GPUs, and Kin. For teams with no compliance boundary: you bring the domain or the database, we bring the compute.
If you can't send your data to a frontier lab, building a model is the only path left.
Classified and controlled-unclassified networks where air-gap is a hard requirement and offline delivery is the only install path.
Core banking, transaction, and PII data under supervisory rules that forbid third-party model processing.
Patient records and trial data, where auditors care about where the data physically sits rather than how well it is encrypted in transit.
Energy, utilities, and industrial operators running OT networks that are isolated from the internet by design.
If you have a pursuit with an on-premise or sovereign model requirement, we build it under your delivery. Your client ends up owning the weights outright. There is no telemetry and no callback in anything we hand over, so it should clear your security review without argument, and the audit report is written so it can go into your bid as it stands.
We built these tools to build our own models. Three are open source, so you can read the code before you trust us with a build.
A free one-week corpus audit tells you whether you need a model at all. Half end with us recommending you don't build.
Your domain, the data you actually have, and where a build is allowed to run.
A token budget, a recommended rung, and eval targets. Or the recommendation not to build.