# More room for human intention. A computing environment that gathers capabilities around the work people want to do, preserves what remains useful and lets a new result become the beginning of something else. Visitor question: What kind of computing is Radiolaria trying to make possible? Genre: Vision. Website explanation · Updated 2026-10-02. ## Delegate work. Keep authorship. People bring intention, values, responsibility, acceptance of consequences and authorship. Computing participants contribute reasoning, search, design, calculation, alternatives and bounded execution. Their results can extend what a person can do while the owners' decisions remain identifiable. This is a human direction for organising work, not an exhaustive runtime contract or a claim about model personhood. ## Human AI symbiosis Human attention should not be the glue between applications. A person should be able to think about a problem without also carrying every fragment of its context between disconnected tools. Some knowledge is difficult to express as a specification at the beginning. A researcher notices a promising discrepancy. A craftsperson recognises a useful shape. Someone knows how their community works and sees a service it needs. Conversation with a capable assistant can help that knowledge take form: a question, a plan, a candidate, something that can be examined. The ambition is not to replace human authorship. It is to give more people the means to turn what they know, imagine and care about into work they can examine and build on. For that partnership to be useful, its contributions must remain distinguishable. An assistant can propose. A program can calculate. A source can supply material. A person or organisation still owns the decisions that belong to them. Radiolaria’s principle, **Capability ≠ authority**, gives the collaboration a structure in which greater computational ability does not silently become greater power over another participant. ## Programming the intention Radiolaria explores a different object of programming: a continuing intention. What should become true, under which conditions, on what evidence, and whose decisions matter? Models, ordinary code, solvers, services, databases, QPUs and devices can contribute bounded work around that intention. The purpose can remain while the composition of work changes. In this architectural view, the computer is the composition that forms around the work; no single participant defines the whole. When the world changes, affected work can be recomputed while useful results remain. Information and permission have their own lifetimes. A missing capability may be retrieved and qualified, or separately created, examined and admitted. Greater capability still does not become greater authority. The [recorded cases](/understand/one-kernel/) and [implemented growth mechanisms](/understand/growth/) make parts of this picture concrete. The wider ambition is to make these recurring relationships—intention, current state, provenance, time, computation and ownership—a common basis for many kinds of work. A shop, a subscription service and a scientific instrument could then differ in their domain knowledge while sharing a way to compose capabilities, evidence and permission around what people intend. ### One pizzeria, a changing day *An imagined application, not a recorded demonstration.* Pepperoni runs out: availability changes. Rain makes the old delivery estimate less reliable. A courier becomes unavailable: a delivery branch needs a different resource. “No orders after 21:00” changes the owner’s permission boundary. A robot courier would add an actuator; a better route planner would add a different kind of computation. Each new capability needs compatible integration and admission before use. The continuing intention remains: **run today’s pizzeria under its owner’s conditions**. With the relevant capabilities in place, changing stock, delivery conditions or opening policy becomes part of operating the business. It need not require a new application for every change. ## Work across devices A screen is a place to interact, not the boundary of a task. Imagine reviewing a piece of work at a desk, continuing a conversation in another room and using an available display when a larger view helps. The task brings its relevant context and retained results. Devices contribute the capabilities they can provide. A lost resource changes the available work without necessarily ending the intention behind it. This is a scene from a proposed native ecosystem. It needs discovery, compatible interfaces, media transport, device integration and decisions about access. The published reference does not supply a finished adapter for every television or appliance. Its narrower foundation is already worth examining. In [Workspace](/demonstrations/workspace/), a changed resource and an expiring permission have different lifetimes from useful photo work. The system retains what remains applicable and performs the work that the change requires. That distinction opens a way to think about continuity beyond one process or screen. ## Reusable capabilities A new pizzeria is a new business, with its own recipes, stock, prices, people and promises. It is not a new computational universe. Ordering, making an offer, checking availability and arranging delivery are recurring classes of work. In the envisaged ecosystem, a capability can be built for such a class and used across compatible settings. Local data and policy stay local. One owner does not become another owner’s subordinate because their systems exchange information. A manufacturer could publish supported device capabilities; a domain developer could build a reusable way to organise a recurring task. There are three levels of reuse, each with a different reach: 1. **Within a task:** retain an applicable result and selectively recompute the affected part. Workspace gives a recorded example of this relationship. 2. **Between tasks:** use an existing capability or qualified history again, after checking whether it fits the new task and current conditions. Past acceptance informs a new decision; it does not carry old permission forward. 3. **Across organisations and devices:** make shared capabilities available in a proposed native ecosystem. This wider horizon still requires infrastructure, compatibility work and adoption. Creating a shared capability, integrating and admitting it in an owner’s installation, configuring local data and policy, and executing a particular task are separate kinds of work. The aim is to move more recurring effort into a useful common foundation. ## Computing without repetition Progress is not only computing faster. It is also avoiding work that no longer needs to be done. A result with a known origin and applicable conditions can survive the operation that created it. A small change may call for a small amount of new work. A capability that has already been built and examined can become a starting point for another application. Useful history can tell us where closer examination is needed, as well as where a bounded shortcut is justified. The economic idea is straightforward: **shared capability development + local data, policy and configuration + compatibility checks + execution and maintenance**. Repeated use can distribute the cost of the common part. It does not remove the costs of local responsibility, integration or continued operation. The [reference material](/research/) supports particular relationships: retained results, selective recomputation and qualified reuse. It is not a general benchmark of energy or financial savings. Those measurements would need a stated workload, comparison and environment. The ambition remains meaningful before such a universal number exists: spend more computation on what is newly needed, and preserve more of what has already become useful. ## Build the way it runs You extend Radiolaria. Radiolaria extends what you can do. Creating a capability is itself continuing work. An intention becomes a brief; the brief becomes a design; implementation produces a candidate; examination finds what works and what needs repair. A decision admits the result into a new role. That result can then participate in the next task. The [Workshop](/workshop/) makes this cycle visible. Humans and assistants have different roles, and their work leaves concrete artifacts for the next participant. Corrections belong to the history rather than disappearing behind a final success label. This is a practical correspondence between creation and use. It is also a direction for further automation. The current human and assistant workflow does not amount to a fully integrated autonomous Needle Factory. It shows how to continue development while keeping proposal, evidence and acceptance distinct. ## The next use Examples make a foundation understandable. They cannot exhaust what other people may do with it. A domain specialist brings the knowledge that makes a new application worthwhile: which conditions matter, which sources deserve attention, what a useful result looks like and who must decide. The next step is to connect that knowledge to the mechanisms already available, identify the missing work and make a bounded result that can be examined. Radiolaria offers an architecture, a working reference and a way to continue. Its larger future depends on people who recognise a task worth bringing to that foundation. [Understand the system](/understand/) · [Explore a possible application with your assistant](/read-with-ai/) · [Start a concrete development task](/build/) · [Contact the author](/about/#contact) ## Sources and scope - [Intent and continuing work](https://github.com/AAkhtanin/hedgehog-os/blob/800f38c3407b3f4ec93dc943c977ffeecb614881/docs/showcase/radiolaria_os_v01/chapters/01_intent.md) — publication explanation; ref `800f38c3407b3f4ec93dc943c977ffeecb614881`; path `docs/showcase/radiolaria_os_v01/chapters/01_intent.md`. Source date: 2026-09-29. Access: HTTP_200; complete chapter text read; checked 2026-10-02. - [Time, memory and continuation](https://github.com/AAkhtanin/hedgehog-os/blob/800f38c3407b3f4ec93dc943c977ffeecb614881/docs/showcase/radiolaria_os_v01/chapters/03_time_memory.md) — publication explanation; ref `800f38c3407b3f4ec93dc943c977ffeecb614881`; path `docs/showcase/radiolaria_os_v01/chapters/03_time_memory.md`. Source date: 2026-09-29. Access: HTTP_200; complete chapter text read; checked 2026-10-02. - [Experience, choice and finite expenditure](https://github.com/AAkhtanin/hedgehog-os/blob/800f38c3407b3f4ec93dc943c977ffeecb614881/docs/showcase/radiolaria_os_v01/chapters/05_experience.md) — publication explanation; ref `800f38c3407b3f4ec93dc943c977ffeecb614881`; path `docs/showcase/radiolaria_os_v01/chapters/05_experience.md`. Source date: 2026-09-29. Access: HTTP_200; complete chapter text read; checked 2026-10-02. - [Evidence and method](https://github.com/AAkhtanin/hedgehog-os/blob/800f38c3407b3f4ec93dc943c977ffeecb614881/docs/showcase/radiolaria_os_v01/chapters/06_method.md) — publication explanation; ref `800f38c3407b3f4ec93dc943c977ffeecb614881`; path `docs/showcase/radiolaria_os_v01/chapters/06_method.md`. Source date: 2026-09-29. Access: HTTP_200; complete chapter text read; checked 2026-10-02. - [Integration and authorship](https://github.com/AAkhtanin/hedgehog-os/blob/800f38c3407b3f4ec93dc943c977ffeecb614881/docs/showcase/radiolaria_os_v01/chapters/11_integration.md) — publication explanation; ref `800f38c3407b3f4ec93dc943c977ffeecb614881`; path `docs/showcase/radiolaria_os_v01/chapters/11_integration.md`. Source date: 2026-09-29. Access: HTTP_200; complete chapter text read; checked 2026-10-02. - [Public Work Assistant Handoff](/downloads/RADIOLARIA_PUBLIC_WORK_HANDOFF_EN_V01_20261001.txt) — author supplied engineering context explanatory companion; ref `public snapshot 2026-10-01`; path `public/downloads/RADIOLARIA_PUBLIC_WORK_HANDOFF_EN_V01_20261001.txt`. Source date: 2026-10-01. Access: complete supplied text read; expected bytes and SHA-256 verified locally by task intake; checked 2026-10-02. This page is a website explanation, updated 2026-10-02. Its explanation does not create a new execution result or change the status of its sources. ## Continue reading - [Work that keeps its meaning.](/understand/) - [Ephemeral Computer](/demonstrations/workspace/) - [Bring a domain. Build a capability.](/build/) - [Radiolaria is built the way it runs.](/workshop/)