Generic AI platforms
Per-token billing on someone else's LLM
Every conversation eats into your margin. The platform's capability is identical for every customer. Compete on price; lose on cost.
The hive · Our own LLM
Our own LLM.
Trained as a community.
The Hive is more than a metaphor. It's our own large language model — trained on anonymised conversations from every member of the community, owned collectively, and getting smarter every month. Your hive contribution grows our shared intelligence. Everyone wins.
What the hive really is
Not a metaphor.
A real, trained LLM.
Every AI platform on the market uses someone else's LLM and pays per token. We've built our own — a vertical-trained LLM, fine-tuned on anonymised conversations contributed by the community. It runs on our infrastructure, serves your agents, and grows with every hive member. No API margin to OpenAI or Anthropic. No per-conversation cost.
Atomic Engine
Our own LLM. Grown by the hive.
We own the LLM. We train it on community-contributed data. We host it ourselves. The more the hive uses it, the smarter it gets. No per-token pass-through to a giant US tech firm.
Where Claude fits in
We use Anthropic's Claude for the parts where premier-grade reasoning genuinely matters — complex multi-step decisions, edge cases, high-stakes responses. The everyday vertical conversation work runs on our own hive-trained LLM. The result: you get the best of both, without paying Anthropic margin on every conversation.
How the hive works
Growth through collaboration.
Not competition.
You opt in. Your conversations are anonymised — stripped of every identifier, verified by automated and human review. The anonymised patterns feed vertical-specific training. Our hive LLM gets smarter. Updated models deploy to every member of that vertical. Everyone wins.
- 01
Opt in
Toggle hive participation in your account. Toggle off at any time.
- 02
Anonymise
Names, emails, numbers, specifics — all stripped before anything leaves your environment.
- 03
Tag
Industry, sub-industry, department, role. The metadata that drives training.
- 04
Train
Our LLM fine-tunes on aggregate vertical patterns.
- 05
Deploy
Smarter agents push to every hive member instantly.
Hive → grade progression
Hive contribution
drives agent grading.
The more you and others contribute to the hive, the faster your agents climb the periodic grading system. An agent that started at Hydrogen hits Iron at month twelve and Gold by month twenty-four — only if the hive is feeding them growth. Beyond Gold sit the rare grades — Iridium, Tungsten, Osmium, and the actinides Uranium and Plutonium for the most veteran agents in the largest hive deployments.
Day 1
H
Hydrogen
Month 3
Li
Lithium
Month 6
Al
Aluminium
Month 12
Fe
Iron
Month 24
Au
Gold
Privacy & trust
Anonymisation that
actually works.
The hive only works if anonymisation is bulletproof. We've built a two-layer pipeline — pattern matching plus context-aware LLM review — and we verify zero PII leakage before any data enters the training set.
Layer one
Pattern-based stripping
Emails, phone numbers, postcodes (first half preserved for regional context), person names, company names, account numbers, IBANs, specific monetary amounts — all auto-replaced with anonymous tokens like [EMAIL], [POSTCODE], [PERSON], [AMOUNT-BAND].
Layer two
Context-aware stripping
An LLM-based pass catches indirect identifiers (“the largest hospital in Truro”), role identifiers (“our finance director Jane”), and referential PII (“same person who emailed last Tuesday”). Verified by automated pattern scan plus 1% human review.