Products
Our own product line.
Nine products, all of them standing on Mnemos, our own governance platform. These are at varying stages and none is generally available. We list them by name, with their real status, because a client is entitled to know which part of a proposal is proven today and which is our own roadmap.
Ask your organisation anything. Get an answer you can check.
Every customer conversation, understood, and every next step, evidenced.
Turn the paperwork you run on into structure you can query.
Agents with a job, a manager, a budget and a record.
Professional creative you can control, with provenance you can prove.
An engineering assistant for teams whose code cannot leave the building.
Specification to a tested, governed, auditable system.
From a plain-English outcome to a running, governed application.
The same idea, sized for one person or one household.
Everything we deliver commercially today runs on the two production platforms. We keep the two clearly separate — including inside a proposal.
The flagship
KGP: one operating loop with five parts.
Enterprise · public sector
Knowledge platform KGP
Private beta
Ask your organisation anything. Get an answer you can check.
A governed brain for organisational knowledge. Connect the places knowledge already lives, ask in ordinary language, and get an answer with its sources attached — or a clear statement that the material does not support one. Then organise the work that follows, automate the routine part inside approval limits, and learn from what went wrong. One operating loop with five parts.
One operating loop, five parts
- Ask · Noesis
- Employees ask in plain language and get an answer cited to the real source, limited to what they are allowed to see, and honest enough to refuse when the material does not support one.
- Map · Topos
- Captures how work actually gets done — the real steps, decisions and exceptions, and turns fragile tribal knowledge into an asset the organisation owns and can audit.
- Do · Praxis
- Performs routine work, but never blindly: it previews exactly what it is about to do, a named person approves, and it executes and keeps a receipt — inside a budget.
- Improve · Sophia
- Real corrections make the system better without silently rewriting it: every change is reviewed before it lands, kept narrow in scope, and fully reversible.
- Build · Copilot Build
- A guided surface for building new views, flows and assistants on top of the governed layer, inside the same policy and evidence model as everything else.
Why it matters
- Answers, not documents — Minutes of reading replaced by a sourced answer.
- Verifiable by default — Every claim resolves to the passage behind it.
- Safe to say no — It declines rather than guessing, and tells you why.
Sales · service · revenue operations
Customer contact intelligence HelixArc
In development
Every customer conversation, understood, and every next step, evidenced.
An AI-native platform for customer contact: system of record and system of intelligence in one. Accounts, contacts, leads, opportunities and activity are held under audit, with lead conversion that cannot double-count. On top of that sits the intelligence layer — conversation analysis that reads what actually happened on a call, qualification and opportunity scoring against recognised frameworks, and coaching that tells a manager what to work on with whom. Built agent-native: every action available to a person is available to an agent under the same rules and the same record.
Why it matters
- One customer truth — Record and intelligence in one place, not a CRM plus three bolt-ons.
- Reads the conversation — Signal extracted from what was actually said, not from what someone typed afterwards.
- Evidenced, not asserted — Every score and every recommendation traces to the material behind it.
Enterprise · mid-market
Document intelligence Mosaic
Private beta
Turn the paperwork you run on into structure you can query.
Contracts, reports, forms, correspondence and scans go in. Structure comes out — parties, dates, obligations, amounts, clauses — each linked back to the exact place it was found. It reads files where they already live, and sends only the passage it needs to a model.
Why it matters
- Nothing retyped — Structure extracted once, used everywhere.
- Always traceable — Every value points back to its source.
- Exceptions surface — The routine majority handled; the unusual raised.
Enterprise · shared services
Digital workforce Praxeotron
Private beta
Agents with a job, a manager, a budget and a record.
A registry of governed agents configured as roles instead of chatbots — an analyst, a reconciler, a triage worker. Each with a profile, a named owner, exactly the tools it needs, a spending limit and an approval boundary. Registered, versioned and supervised from one place, so the estate is an inventory rather than a rumour.
Why it matters
- Supervisable — Give work, watch it, correct it, stand it down.
- Budgeted — A limit that stops, not a dashboard that reports.
- Accountable — Every action in the same record as your people's.
Communications · education · brand
Creative platform Kromas
Private beta
Professional creative you can control, with provenance you can prove.
Images, diagrams, video and audio produced through templates, a guided studio and structured composition, with governed quality-and-rights review so every output is safe to publish, and a record of what was asked for, what produced it, and on what terms.
Why it matters
- Defensible — You can prove what you made and how.
- At scale — Consistency without a studio.
- Rights-clean — Provenance that survives a challenge.
Developers · regulated environments
Local-first coding agent Kaivor
Private beta
An engineering assistant for teams whose code cannot leave the building.
A fast, Rust-first coding agent that runs locally by default. No account, no code uploaded, and your choice of a model on your own hardware or a hosted one you point it at. It is also the proof of how we build: it was produced by our own agentic tooling as a separate, governed repository that ships on its own.
Why it matters
- Nothing leaves — Your code stays on your machine.
- Your model, your choice — Local or hosted, a configuration change.
- No procurement — An engineer can run it today.
Internal · then engineering organisations
Agentic software factory ASF
On the roadmap
Specification to a tested, governed, auditable system.
Specialist agents for specification, design, code, quality and security, with a human approval gate between each stage and an evidence bundle proving every step. Each stage runs on the model priced for it, and the total build cost is capped and provable.
Why it matters
- Evidence by construction — The trail is a by-product of building.
- Compounding — Each product cheaper than the last.
- Consistent — Everything emitted is governed the same way.
Enterprise
Agent-native app generation Numen
On the roadmap
From a plain-English outcome to a running, governed application.
Business software built on the assumption that some users are not people: every action available to a person is available to an agent under the same rules. Described in ordinary language, generated, tested, and delivered with an evidence receipt for what was built and how.
Why it matters
- Work actually leaves — An agent completes the task instead of suggesting it.
- One process — No parallel 'AI section' to reconcile.
- Auditable per action — Who did what, person or not.
Consumer
Personal knowledge space Cortex
On the roadmap
The same idea, sized for one person or one household.
A private memory for a life: capture anything, have it organise itself, and recall it by asking. It lives in an encrypted vault on the owner's own device and works offline, with recall running locally and only what is strictly needed ever reaching a model.
Why it matters
- Private — It runs where you put it.
- Portable — It leaves with you.
- Durable — It outlives any device or provider.
The platform underneath
Mnemos.
The platform every product we build stands on — built once, so governance is never re-argued.
Mnemos is the layer that makes a product governed by construction rather than by policy document. It holds what the organisation knows, decides whether an answer may be given at all, registers agents as principals alongside people, gates execution, keeps a record that can be added to but not quietly altered, and routes work to the smallest model that will do it. Every product in our line inherits all of that on the day it is created.
Mnemos is in active development and is not sold on its own. Client work today is delivered on ZigmaData and on Kyros, powered by Kaman — both in production. We are explicit about the difference because a client is entitled to know which parts of a proposal are proven and which are ours.
Why build the layer first
Governance by construction, not by policy document.
The hard, defensible part of an AI product is not the model and rarely the interface. It is evidence, authority, approval and cost control, and it is the part that has to be right before anything reaches production. Building it once means each additional product is mostly domain experience instead of a fresh governance problem, and it means a client who adopts a second product inherits a control model they have already reviewed.
Each product is mostly domain
Not a fresh governance problem, argued from scratch by a team that has never met a risk committee.
The second review is a delta
A client adopting a second product inherits a control model they have already reviewed.
It cannot be skipped
A governance check a product can route around is decorative. Here there is no route around.
Governance primitives
Six things every product inherits on day one.
- Governed memory
- What the organisation knows, held in scopes that never silently merge.
- Evidence and retrieval
- The checks that decide whether an answer may be given at all.
- Authority and policy
- Agents registered as principals alongside people, under one entitlement model.
- Execution
- Approval gates, isolated execution, autonomy granted in steps instead of at once.
- Record
- Additions only, chained so any deletion or reordering is detectable.
- Routing and cost
- The smallest model that does the job, and a budget that can decline.
Control planes
How the platform is run, paid for, pointed at a model and kept honest.
These are not products and none of them is sold separately. Any organisation adopting anything built on Mnemos gets all four.
Mission Control
Live command across the estate: what is running, what is queued, what failed and why, and which approvals are waiting. Every operator command is itself an audited event, so the layer that fixes things is governed like everything else.
Credits
Spend metered per team, per use case and per agent, routed to the cheapest model that can do the job safely, with a hard brake that refuses work rather than reporting an overrun afterwards. Both the attribution and the refusal are recorded.
Control Room
A private model fabric with one screen to control it: which model runs, what it may spend, and a receipt proving the budget held. It turns a pile of AI infrastructure into something a small team can operate.
Validation Lab
Define what good looks like for a governed capability, then keep checking it: grounding and refusal behaviour, policy enforcement, evidence completeness, and drift against the behaviour that was signed off.
The design-partner route
Not a purchase order. Direct access to the people building it, influence over what comes next, and an honest account of the limitations — in writing, before they start. We take small numbers at a time.