01 · MODELS
Proprietary small foundation models
Compact models built for specific domains — faster, cheaper to run and deployable where large hosted models can't go.
PRODUCTKey4The stack behind every agent.
Nine areas of deep expertise, built across the CoSolve consortium, that let agents read, reason and act inside your own infrastructure.
- capability areas
- 9
- steps in every agent run
- 5
- products
- 4
01 — Capability areas
Nine capabilities, one system.
Combined as each workflow needs, and delivered by one accountable team.
02 · ORCHESTRATION
Enterprise agent orchestration
Coordinating agents, tools and approvals across business systems, with policies enforced at every step.
03 · EXPERTISE
Expert knowledge, encoded
Turning the knowledge of domain experts into executable AI workflows, with verifiable outcomes and full data sovereignty.
04 · DOCUMENTS
Document intelligence
Reading, extracting and reasoning over contracts, forms, records and filings — the evidence agents cite.
PRODUCTKey4KB05 · EDGE
On-device execution
Models and agents that run locally — at the edge, offline, or inside restricted networks.
PRODUCTKey4MB06 · LEARNING
Adaptive learning
Learning software that knows the learner, teaches by asking questions and shows the teacher the whole class.
PRODUCTKey4EDU07 · EXPERIENCE
Personalisation & customer experience
Agents that read catalogue, inventory and behaviour through your existing APIs, build a live profile of each visitor and decide what they see — within the rules you set.
08 · ASSURANCE
Evals, monitoring & explainability
Auditable AI: every output linked to its source with a per-prediction explanation, every deploy gated by evaluations, and live traffic watched around the clock.
09 · DATA
Data & sovereign fine-tuning
Synthetic data and curated datasets, with models fine-tuned inside your private cloud — no calls to a public API.
02 — Products
Ready to deploy.
Packaged products from the stack, running on infrastructure you control.
SMALL MODELSREADY TO DEPLOY
Key4
The small-model platform: a method for making small models expert and checking them against verifiable answers, engineering to run them on ordinary hardware, and its own small language model.
- FOR
- Any organisation with one job that must be done well, inside its walls, and checked before it is relied on
- RUNS ON
- Ordinary hardware, from an office server to a phone
DOCUMENT INTELLIGENCEREADY TO DEPLOY
Key4KB
Answers from your own records, with the page. Each answer quotes the source document and is checked word for word.
- FOR
- Government departments, cooperative banks, universities and boards, legal cells and courts, compliance teams
- RUNS ON
- Your own server
ON-DEVICE EXECUTIONREADY TO DEPLOY · OPEN TO PARTNERS
Key4MB
An AI that uses your phone the way you do — it works in every app and shows every step. Zero tokens, zero cloud, zero data leaving the phone.
- FOR
- Phone makers, carriers and retailers, enterprises and institutions, chipmakers and integrators
- RUNS ON
- Entirely on the phone
ADAPTIVE LEARNINGREADY TO DEPLOY
Key4EDU
Learning that teaches by asking. Its tutor runs on the Key4 model, so its explanations are checked rather than trusted.
- FOR
- Learners preparing for exams, learners who are struggling, teachers with a full classroom
03 — Agents in action
From request to action, with every step on the record.
agent.run / invoice-approval
EXAMPLE> Approve invoice INV-2291 and schedule payment
- Interpretintent: approve_payment
- Plan4 tasks · 3 systems
- Retrieve evidencePO-0417 · INV-2291.pdf
- Apply policyspend limit §4.2 · passed
- Human approvalawaiting finance leadapproved · finance lead
- ActERP · payment scheduled
trace saved · audit-readydata stays on-prem
Evidence linked
Every decision cites the records it rests on.
↳ PO-0417 · INV-2291.pdf
Policy checked
Each step is checked against your rules before the agent acts.
↳ spend limit §4.2 · passed
Human sign-off
The run pauses where you require approval and continues only when a person signs off.
↳ approved · finance lead
Full trace, kept on-prem
The whole run is logged and audit-ready, and never leaves your environment.
↳ trace saved · data stays on-prem
04 — Deployed and delivered
Runs where you need it. Built by the right specialists.
- Your data centreON-PREMISES
- A cloud under your jurisdictionSOVEREIGN CLOUD
- Local executionON-DEVICE & EDGE
- The Key 4 AISmall is not little. Small enough to own.
- Exord Systems
- NeuralcraftAI for regulated industries.
- CoServeEnterprise software delivery, rebuilt around AI.
05 — Put to work
Where these capabilities are put to work.
01 · HEALTHCAREIN PRODUCTION
Healthcare operations automation
Agentic hospital revenue flows, running entirely within customer-controlled infrastructure to protect sensitive patient data.
VALUE
1,000+ hours of manual expert work removed every month, encoded in 6 weeks.
02 · INSURANCEDELIVERED
Member-record workflows inside a private cloud
Extraction, denial classification, summarisation, appeal drafting and policy Q&A over member records that can never leave the insurer's private cloud.
RESULT
Five workflows served by one fine-tuned model the client owns, with zero tokens leaving the private cloud.
03 · RETAIL & E-COMMERCEPILOT
One-to-one personalisation without re-platforming
An agentic layer on a merchant's existing stack reads catalogue, inventory and shopper behaviour, then decides what each visitor sees — recommendations, ranked search, bundles and messaging — within merchant-set rules.
POTENTIAL IMPACT
Higher conversion and order value from one-to-one experiences, tested against a control group — with no overhaul of checkout, order or inventory systems.
FAQ
Questions about the stack.
How are agent decisions explained?
Each action is linked to the evidence retrieved and the policy applied, so reviewers and auditors can trace why it happened.
Do we have to rely on a hosted model and pay per token?
No. Our proprietary small models run on your hardware. You keep the data on-premises, without metered billing.
Where do humans stay in the loop?
You set the critical steps that need human approval. The agent pauses there and proceeds only on sign-off.
Can the agents work with our existing systems?
Yes. Agents act through your existing tools, APIs and business systems — no re-platforming — with your rules, regulations and permissions checked at every step.
How do you know the agents are working correctly?
Every release is gated by evaluations agreed up front. Outputs link to their source with per-prediction explanations, and live traffic is monitored for drift, bias and anomalies.
What does a small foundation model mean for us?
Compact models built for a specific domain: faster and cheaper to run, checked against answers that can be verified, and deployable on hardware you already operate — where large hosted models can't go.
Can models be trained on our data without it leaving our environment?
Yes. Models are fine-tuned inside your private cloud using curated and synthetic datasets, with no calls to a public API.
Can the agents run offline or on devices?
Yes. Models and agents can run locally — at the edge, offline, inside restricted networks, or entirely on a phone.
Contact
Have a workflow worth automating?
Book a 30-minute briefing with our solution team.
EMAIL
hello@cosolve.aiHEADQUARTERS
Meydan Grandstand, Nad Al Sheba, Dubai, U.A.E.