Task decomposition
Break a complex request into smaller, typed subtasks before selecting experts.
Independent AI research
Researching AI for a more useful future.
MorningMind is an AI research company exploring how intelligent systems can help people understand information, solve problems, and make better decisions across fields.
Research-led. Grounded in real-world needs.
Our models · latest research
ORK-V1: MoSLM orchestration.
ORK-V1 is a task-level orchestrator built around MoSLM (Mixture of Small Language Models). It is designed to decompose requests and route each task to suitable specialist models, with a planned release on January 1, 2027.
- ArchitectureMoSLM Mixture of Small Language Models
- Model typeTask-level orchestrator
- Development stagePre-release development
- Target launchJanuary 1, 2027
ORK-V1 capabilities
Sparse expert routing
Use top-k routing to send each subtask to only the most suitable small language models.
Composable specialists
Combine focused experts such as math, code, retrieval QA, and tool calling for multi-skill work.
Confidence-based escalation
Escalate low-confidence subtasks to a larger model instead of using it for every request.
Open expert pool
Represent expert capabilities so models can be added or replaced without retraining the orchestrator.
Quality and efficiency goals
Research training and routing objectives that account for correctness, token cost, latency, and load balance.
ORK-V1 is in pre-release development. Its performance and efficiency targets will be evaluated ahead of the planned launch.
Explore the MoSLM researchExplore ORK-V1 model details and the animated routing demo.
From research to practice
Exploring AI across real-world challenges.
Research and ideas for making intelligent systems more capable and useful.
01
Knowledge systems
Explore how AI can help people find, connect, and make sense of complex information.
Make knowledge useful02
Specialised AI
Research efficient models and orchestration approaches for different tasks and contexts.
Match models to tasks03
Human-centered AI
Study how intelligent tools can support human understanding, judgement, and decision-making.
Keep people in focusOur point of view
Technology should help people do meaningful work.
AI is most useful when it responds to real needs and complements human expertise.
Get to know MorningMindGrounded in real needs
Start with the problems people face, not technology for its own sake.
Useful, understandable AI
Make complex information easier to work with and easier to trust.
People stay at the centre
Build tools to support human judgement, curiosity, and connection.
Security & privacy
Certified where it counts.
Built to meet the security and privacy standards enterprise and healthcare teams require.
ISO 27001
Certified to the internationally recognized standard for information security management within its defined scope.
Active certification · defined scopeSOC 2
Current SOC 2 report covering its defined systems and scope.
Current report issuedGDPR
GDPR obligations and privacy controls reviewed for applicable personal-data processing.
Privacy complianceHIPAA
HIPAA compliance confirmed for applicable healthcare services, with a BAA available for those services.
Healthcare servicesCertification and report scopes apply to defined systems and services. Contact us to confirm applicability to your use case.
Let’s shape what’s next
Let’s move AI research forward.
Are you a researcher, organisation, or potential partner exploring the future of AI?
Get in touch