Understand the work.
Know the boundaries.

The operating system for AI agents that act in physical places. Aspinos connects observations from cameras and other sources with a site's instructions, responsibilities and history, so agents can prepare and coordinate work.

A camera shows part of a place. Aspinos brings that observation together with the relevant area, operating rule and responsible person. The resulting work can include an inspection, a response proposal or a handover, with the evidence and any uncertainty kept alongside it.

The current prototype uses a defined set of built-in agents to check observations, apply configured context, prepare tasks and retain records. Their work follows explicit rules and authority limits. It is not an open platform for arbitrary agents, and an agent's explanation is not proof that an action succeeded.

The working prototype can answer supported questions from the records a user is authorised to see, such as what changed, what needs attention and which evidence supports a case. It links the answer to those records. Website questions use authored examples; they do not analyse your camera footage.

People responsible for physical places: school staff, mosque and community teams, hotel operators, and teams working in industry, construction, retail and maritime settings. Each place needs its own operating instructions and evaluation. The examples on this website do not identify commissioned customer deployments.

Sharoze Ali, co-founder and CTO, leads perception, camera ingestion and edge runtime work. Yusuf Baykal Bozkurt, co-founder and CEO, leads product vision, strategy, partnerships and market development. Yusuf is the founder of Beonbrand Inc. The company story includes their backgrounds and LinkedIn profiles.

The prototype supports bounded camera and source intake alongside simulated camera, access, notification, environmental and building-system connections. Compatibility must be tested with the specific device, permissions and network. Registering a connection does not establish that equipment is connected or that a physical action was delivered.

Identify the site and responsible people, add and test sources, place them on the map, then configure analysis areas, privacy and operating instructions. The prototype separates draft, testing, review and publication. A reviewed version applies to its selected site; a saved draft does not commission equipment.

Operate brings cases, ownership, evidence and the next task together. Monitor provides camera layouts, source health and retained event-time evidence. Admin contains the full Command experience, site configuration, people and connections. The local team demonstration shares one persistent, permission-scoped site world across all three.

The deployment choice depends on the sources, models and workload. Aspinos has run bounded CPU experiments, and GPU capacity is part of the development and evaluation plan. A site assessment must establish processing needs and measured performance; there is no blanket claim that every camera can run on an existing computer.

Consider a scoped evaluation when observations, site instructions and follow-up work are split across cameras and operational tools. Start with one defined workflow, authorised source data and a named team who can review outcomes. Device compatibility, privacy, processing needs and usefulness must be demonstrated at that site.

In the protected prototype intake, published privacy masks redact image pixels before processing and storage. Viewing and exports use the protected captures, which retain their mask version. A later change does not silently rewrite historical footage. The actual camera path and privacy configuration must be verified before a site evaluation.

No. An analysis zone defines where an observation is relevant, such as an entrance or a restricted area. A privacy mask hides pixels. Excluding an area from analysis does not hide that area in footage.

No. Facial recognition, demographic inference and inferred intent are outside the product. A visible observation can justify a check; it cannot establish who a person is, what they believe or what they intend to do.

Agents run the work. People hold the authority. The runtime requires a named authorised person for consequential actions and enforces additional approval requirements where configured. Administration access does not itself grant operational approval authority. Approval, delivery, acknowledgement and the checked outcome remain separate.

The missing input stays visible. A lost camera does not mean an area is clear, and an unjudged event remains a task for a person. Completing one response task does not erase a separate evidence review or closure requirement.

The website combines authored scenes, retained presentation media and interactive examples to explain the product. These are not proof of live detection or a customer deployment. The working prototype separately retains source records and capture classifications: live, recorded, staged or simulated. Presentation never upgrades their evidence.

The local prototype demonstrates shared case work, monitoring, configuration, permissions, protected captures and retained history. Some supported source paths can accept supervised input; equipment integrations require their own acceptance checks. A completed software demonstration does not establish field accuracy, a commissioned pilot or production readiness.

Aspinos is a venture in development and is not yet separately incorporated. The current software supports a shared team demonstration with fictional sites; the available evidence does not establish a commissioned customer deployment. A real pilot requires agreed scope, a contracting entity, permission to use source data and on-site evaluation.

No. Current site instructions and previous records provide context; they are not automatically trained model weights. Aspinos plans to develop and evaluate its perception and reasoning capabilities using appropriately authorised data. Any training, version change or stronger performance claim needs its own evidence and review.

Start with one place and one useful workflow. Agree the responsible team, access permissions, source data, privacy boundaries and what a successful evaluation would demonstrate. The deployment page helps prepare that conversation. No site access, pilot participation or delivery date is assumed before agreement.