Specialist work
Match bounded work to an eligible model, configuration, inference provider and infrastructure; compare the attributed results.
Agentz · Vision
Your memory. Your tools. Your continuity.
Your intelligence, memory and tools remain with you as your work evolves.
Mission
Provide your continuing Cloud Computer: coordinate specialist work, retain accepted memory in Cortex and disclose scoped context through Nexus.
One of Skyzai’s seven product strategies, designed to contribute to outcomes for Me (individuals), We (organisations) and They (machine intelligence).
Rosetta agent row · Axiology
What matters, to whom, under which consequences, at whose cost?
Agentz connects your priorities and constraints to specialist work, reviewable plans and the continuing memory of a case.
Objectives
- Resume the right case with its reasons and sources.
- Give each specialist only the context its work needs.
- Keep accepted memory recoverable and proposals within your scope.
Strategy
Match each bounded task to an eligible model, configuration, inference provider and infrastructure. Coordinate specialists and independent reviews around Cortex memory, Nexus grants, device continuity and inspectable usage; qualify fallback before replacement.
Cortex memory
Retain accepted knowledge with its sources, corrections and recovery needs.
Nexus connections
Connect permitted context across products, people and devices.
KPIs
Proposed measures of progress
Measure useful outcomes, recovery and export fidelity, permission leaks, replica failures, stale context, full interaction latency and correction-inclusive cost.
These measures describe how to evaluate this product’s contribution. Each needs a defined scope, observation period and evidence.
- Targets
- Not set. Thresholds remain to be agreed for the proposed measures.
- Performance observations
- No measured results are recorded here for these proposed measures.
Agentz measurement definitions
Proposed definitions. Period: the next declared acceptance cycle, with its start/end and selected account/case population recorded before measurement. Performance observation: Not recorded. Target: Unset. Actual accountable assignees and service evidence remain to be supplied.Resume fidelity
Replacement qualification
Context isolation
Recovery and export fidelity
Useful completion
Continuing-work cost and latency
Launch evidence
Dated measurements of product launches, live integrations and published customer outcomes have not been recorded here. Missing evidence is not a measured zero.
- Released services
- Not recorded
- Live product integrations
- Not recorded
- Published customer outcomes
- Not recorded
Historical declaration · . Owner direction, 18 October 2026. An owner-specified zero-release baseline was recorded for planning. It is retained as source history and is not a measured product result.
Agents
The Rosetta agent row connects this product’s field of inquiry to occupations, skills and responsibilities. The occupations below describe specialist work; different people, models and tools may carry these capabilities within an explicit assignment.
Rosetta agent row. Product L4 · Axiology
Responsibility. Reviewable plans and unsigned proposals
Reasoning approach. Abductive planning: propose a plausible way forward, compare alternatives and examine whose needs, costs and consequences each proposal serves.
Work coordinator
Skills and responsibilities. Qualifies task routes and replacements within privacy, budget and deadline limits; brings results together for independent review.
Memory steward
Skills and responsibilities. Maintains provenance and revisions, with portable export and redundant storage as delivery targets.
Context curator
Skills and responsibilities. Scopes each view by purpose, access and expiry, without granting permission to spend or act.
Specialist agents bring domain knowledge, evidence evaluation, planning, memory lifecycle, secure systems, privacy and recovery skills; appointments and operating competence need evidence.Specialist inputs, skills, outputs and limits
Work coordinator and routing specialist
Memory steward
Context curator
Independent reviewer
Usage reviewer
These are roles and skills in development, with a defined context and an output to inspect. Their performance needs evidence and evaluation; a role description does not give an agent permission to act.
See it in a piece of work.
Local interactive example · commissioning an analysis
Continue the work when the worker changes.
A team compares two methods for an independent analysis. The original worker stops partway through. Explore how Agentz could select a compatible replacement while Cortex retains accepted records and Nexus limits the next disclosure.
Fictional records and qualification checks. Nothing is sent, stored persistently or decrypted; no model runs and no money moves. Reload resets the case. This is separate from the workshop procurement example.
Pinned commitment
Analysis agreement · version 1
Compare the two analysis methods using the shared brief and public sources.
USD 40 for the analysis; no purchase or payment permission
A named team reviewer checks scope, source attribution and unresolved questions.
Cortex · accepted memory
Team brief v1 · accepted public source v1
Source v1: “Method A uses public material.”
Accepted source revision is current in this example. A correction creates no interview permission.
Proposal: none. Delivery and acceptance unresolved.
Nexus · next permitted view
Grant analysis-context-v1 · active
Purpose: compare the analysis methods.
- team-brief-v1
- public-method-account-v1
Private personnel records remain excluded. Use for this assignment; no provider retention or training permitted. End of this fictional assignment; expiry can be simulated below.
Task, model, configuration and provider
- Current worker/provider
- Original worker · Provider A
- Model/version
- Example analysis model · version 1
- Configuration
- Read-only sources; no tools that send, purchase or pay
- Stop condition
- Return an attributed comparison or report insufficient evidence; no sending, purchasing or paying.
- Qualification
- Original fictional assignment. These are illustrative fixtures, not tested real providers.
- Usage
- USD 0.12 simulated prior usage; remaining usage unknown. Interruption, refusal and replacement do not erase prior usage or authorise payment. The analysis ceiling remains USD 40.
The original worker has a permitted brief and source revision 1. No result has been accepted.Inspect the local history (0)
Replacing a worker preserves the purpose, accepted versions, agreement and resource obligations. It adds no grant. Storage encryption, independent replicas, export and actual cross-device recovery each require separate implementation and measured evidence.
Inspect memory and disclosure responsibilities · Me · We · They
One job, with clear boundaries.
- Starts with
- Your intentions, permitted evidence and forecasts, accepted memory, budgets and scope.
- Produces
- Attributed specialist results, reviewable proposals and continuing case memory, with acceptance, sources and revisions preserved.
Agentz does not sign or acquire authority through integration. Keys are intended to remain under your control; context grants, work mandates and payment authorisation are separate. Redundant storage and private cloud computation are implementation targets.
Cortex, Nexus and your continuing case
Agentz sits at the centre of APU and MurMur, Circle and LIQUID, Reality Futures and Menexus. The Atman metaphor describes continuity throughout Skyzai; it makes no claim of consciousness or authority.
Cortex · memory and redundant storage
Preserve the records you accept, with their sources and revisions. Redundant encrypted copies, independent recovery and portable export are targets that need measured proof.
Nexus · connections and permissions
Connect only the context each task may use. Purpose, expiry and revocation remain explicit; a connection does not grant permission to act or spend.
APU · your interface
Speak, type and navigate through APU. Its voice and text interface presents the context Nexus permits; useful accepted memory continues in Cortex.
Your computer goes with you
Agentz is designed to route each bounded task to an eligible model, model configuration, inference provider and infrastructure. A replacement needs compatible task performance, source format, privacy, licensing, budget and deadline. The local example demonstrates the decision boundaries; it connects no provider.
APU is the interface through phones, internet-connected earbuds and other devices. Open encrypted neo-cloud infrastructure, high bandwidth, low latency and digital-currency payments are design targets. They are not established by this website.
Your stored memory is intended to use keys under your control. Private cloud AI also requires verified execution and scoped key release: storage encryption alone does not protect plaintext during ordinary inference.
Capabilities and current proof.
As reviewed on , these are selected capabilities traced to their product owner. The proof level describes a plan, source observation or bounded test; it does not establish a released service.
Continuing case work and memory boundary · Source observed
Case code binds work, feedback and retained state across product contributions.
Current limit: Local case storage and generated drafts are not the intended durable redundant Agentz memory.
Bounded local review and learning · Bounded local proof
Device brief and local bounded review/feedback loop passed documented local checks.
Current limit: Device work and synthetic provider checks do not prove autonomous multi-agent service or live model inference.
Authenticated case synchronization · Hosted source; release unaccepted
Flutter outbox/CAS and Firebase callable/Firestore source have local emulator and unit evidence.
Current limit: No production deployment or two signed-in device acceptance was established.
Cortex memory and retrieval within Agentz · Source observed
The owner puts Cortex within Agentz; dated source and a separate proof package show bounded text/retrieval interfaces.
Current limit: Historical Cortex product labels and a text package do not establish production memory or search.
Bounded text lifecycle · Bounded local proof
A dependency-free Dart package and two synthetic Flutter hosts exercise deterministic routing and text lifecycle.
Current limit: No real model, provider quota, production extraction or independent-team adoption is proven.
Encrypted redundant memory · Proposed
User-controlled encrypted replicas and exact restore are owner-directed Cloud Computer requirements.
Current limit: No replica count, recovery, failover, confidential inference or key recovery is implemented by the local receipt carrier.
Purpose-scoped connections · Source observed
Existing Connect repositories and consent UI form source candidates for Nexus connection work.
Current limit: A renamed ownership boundary does not prove independently admitted Nexus grants or cross-product consent.
Specialist work and model routing · Bounded local proof
The local host uses one bounded Agentz review step; a separate package demonstrates synthetic deterministic model routing.
Current limit: This does not establish many-agent orchestration, live model service, independent consensus or global budget enforcement.
Attested cloud workers and usage · Proposed
Cloud Computer plan specifies isolated execution, attestation, latency and metered usage.
Current limit: No worker measurement, task-key release, mobile latency profile or currency settlement is implemented.
Product plan, responsibilities and sources
Direction and delivery
Skyzai’s vision is an emergent Marketsphere. Its mission is to develop SPECTRE as an open operating system for it. Me, We and They describe the participant objectives, and each of the seven products is a strategy with its own KPIs and Agents. This nested product plan explains its direction, objectives, delivery work, measures and responsibilities; customer outcomes remain to be demonstrated.
- Vision
- Your intelligence, memory and tools remain with you across devices and products, under your control.
- Mission
- Provide your continuing Cloud Computer: coordinate specialist work, retain accepted memory in Cortex and disclose scoped context through Nexus.
- Objectives
- Resume the correct case, preserve sources and disagreement, recover accepted memory and keep every agent within admitted scope, budget and authority.
- Strategies
- Match each bounded task to an eligible model, configuration, inference provider and infrastructure. Coordinate specialists and independent reviews around Cortex memory, Nexus grants, device continuity and inspectable usage; qualify fallback before replacement.
- Measures & triggers
- Measure useful outcomes, recovery and export fidelity, permission leaks, replica failures, stale context, full interaction latency and correction-inclusive cost.
- Agents & skills
- Specialist agents bring domain knowledge, evidence evaluation, planning, memory lifecycle, secure systems, privacy and recovery skills; appointments and operating competence need evidence.
Plan basis and open decisions
Current owner direction defines product responsibilities. Retained proposals inform delivery plans and engineering work. Signed charters, current operating mandates, staffing and measured baselines retain their separate standing.
- Prove a complete case journey across devices, specialists and recovery paths before claiming a connected personal cloud.
- Open encrypted infrastructure, user-controlled keys, measured low latency, independent replicas and digital-currency payments are targets, not operating services.
These working documents require access to the corresponding private GitHub repositories.
- Adopted owner product direction · 2026-10-13
- Reviewed product proposal and retained delivery plan · 2026-10-09
- Current Agentz Cloud Computer VMOSK-A · 2026-10-13
- Corresponding Rosetta L4 row · 2026-10-09
Rosetta agent row.
Selected product correspondence. Agentz retains the L4 product home and unsigned-proposal function while housing Cortex, Nexus and the intended redundant memory capability. The Atman metaphor describes continuity across Skyzai; it is not a claim of consciousness or authority.
- Product L4 · Axiology
- Reviewable plans and unsigned proposals
What matters, to whom, under which consequences, at whose cost? Agentz connects your priorities and constraints to specialist work, reviewable plans and the continuing memory of a case.
How the Rosetta agent row is used
Skyzai presents seven products: APU, Circle, ReFu, Agentz, Menexus, LIQUID and MurMur. Agentz is the continuing centre, with Cortex memory and Nexus connections. APU is the interface. Menexus means ensurance, insurance and assurance. Product roles preserve the underlying Rosetta coordinates. Emergentism is the framework across the whole. The Spectre network is a component within the wider Marketsphere programme; Permanent Composability supports independent reuse across the products.
The Rosetta agent row relates this product’s inquiry, reasoning and specialist skills to Emergentism. Several models, people or tools may contribute to the work. The philosophical correspondence, product responsibility and agent dispatch contract remain distinct; this page neither establishes deployed performance nor grants a model permission to act. This product direction does not migrate runtime contracts or amend signed source history.
Explore the design.
Connected capabilities in this website home. These are designs with distinct service responsibilities.
These pages describe the intended product and its delivery work. They do not open a signed-in app or establish operating availability.