Private AI matters.
We manage everything.
You stay in control of updates, data, and access. The AI runs in your firm; the headaches don't run on your team.
The privacy problem
The strongest cloud AI models are trained on terms of service that allow the provider to log prompts, retain outputs, and route requests through infrastructure you do not control.
For a partner reviewing a transaction, a regulator assessing a license application, or an auditor working through a discovery bundle, those terms are disqualifying. The data does not belong to the firm; the firm is the custodian.
Abila is the answer for everyone who has decided that "we'll just turn off training" is not enough. The data does not leave. There is no API call to a model the firm does not own.
What Abila does
Most people have heard of ChatGPT and Gemini. Behind the headlines are dozens of capable AI models — open-weight, specialised, regional — each with different strengths, costs, and licensing terms.
Abila ships a tested catalog of approved models, together with the on-premise appliance that runs them. An administrator chooses which models each team can use. The catalog is reviewed with each release, so the firm does not have to track the market itself.
Ask matters questions
Hybrid retrieval across the matter's documents with citations to the exact passage.
Learn more →Draft with the firm's voice
Two-stage retrieval: matter facts plus the firm's prior work as the style source.
Learn more →Audit every answer
Hash-chained log of every prompt, every model, every cited chunk — verifiable offline.
Learn more →You choose the sources. Nothing else.
Public chatbots draw on whatever the model was trained on, plus whatever the vendor's web-search layer surfaces this week. The user has no list of sources and no way to remove one.
Abila answers only from the documents you put on the matter. The corpus is visible, removable, and auditable. There is no background web search, no "the model also knows" channel, no unverified third-party knowledge bleeding into the answer.
Every cited passage points back to a file you uploaded. If you didn't put it in, it can't come out.
Masking won't save you.
The find-and-replace defence — "we'll call them Client A, then ChatGPT is fine" — does not survive contact with the actual exposure.
The document is the data, not the name. A renamed client still leaves the deal size, jurisdiction, sector, dates, witness, counterparty, and the verbatim clause text. Aggregation re-identifies. Anonymity is a probability, not a property.
Masking is one-sided. You can rename your client. You cannot rename the bank on the other side, the regulator running the licence assessment, the counsel of record, or the property at the address. One side identified means the pair is identified.
Pseudonymisation is not anonymisation. Under UK/EU GDPR (Article 4, Recital 26), data with names replaced by tokens is still personal data. ICO and EDPB guidance is explicit. The compliance obligations don't lift — only the lawyer's intuition does.
The work product itself is confidential. "Redraft this clause" requires pasting the clause. The structure, the sequence of edits, the redlines — that is the firm's IP and the client's deal. No name mask reaches it.
Who it's for
Law firms
50–300 fee-earners, matter-centric work, engagement-letter exposure on AI use.
Learn more →Regulators
License applications, fitness-and-propriety assessments, regulatory returns.
Learn more →Professional advisors
Accountants, auditors, consultants, HR. Regulated client data, audit-trail demands, on-prem expectations.
Learn more →Hash-chained audit
Matter-scoped access
No cross-matter retrieval. Ethical walls enforced at two layers.
Citation-verified output
Every answer's citations are checked against the retrieved chunks.
Hash-chained log
Every prompt, model, and chunk is recorded; the chain is verifiable offline.
Pen-tested per release
Expected to be pen-tested by every enterprise customer.
About Abila
Abila is a Gibraltar-incorporated company, founder-run by the engineers who build and ship the platform. No outside investors, no board pressure to monetise data, no growth team writing the roadmap.
Reply-to on every email is a person who can change the code that day. The company is small on purpose — the customers are firms whose risk appetite rules out the large-vendor failure modes.
Founders
Robert Quinlivan
CEO
30+ years in international financial markets. Senior roles at KPMG, Merrill Lynch, Barclays, and Macquarie took him from Australia to Japan, Korea, and Hong Kong; Group CFO at Sun Hung Kai & Co. (HKEx-listed); then CEO of Bullish Gibraltar, the GFSC-regulated digital asset exchange. CFA charterholder and Fellow of Chartered Accountants AU/NZ. Based in Gibraltar.
Simon Ordish
CTO
35 years shipping software, still cutting code daily. Founded Laverock von Schoultz in 1992 and grew it from 6 to 60 people — clients including Ernst & Young, Reuters, and GCHQ — before its acquisition by La Française des Jeux in 2010. Since then: blockchain at scale, co-founder of WhatsOnChain, and production AI architecture for regulated industries. Based in Gibraltar.