Status: pre-launch. This page describes target design; this document does not establish a deployed service.
Skyzai.ORG — The Solution
The pipeline, in plain language. Separate organs. One loop. Constitutional separation.
The Architecture in One Diagram
TheCircle (IS) → RealityFutures (COULD) → Agentz (SHOULD) → Skyzai execution
observe evaluate possibilities recommend unsigned exact-bound execution
External authorization remains a separate boundary.
↑ │
└────────────── outcome data flows back ─────────────────────────┘
One cognitive loop. Separate organs. Each owns one function. This target separation is expressed in typed packet and authority contracts; local contracts do not prove a live end-to-end system. [D/S]
This loop is a function and consequence diagram, not the product or app packaging diagram. Circle, Reality Futures and Agentz own the IS, COULD and SHOULD contributions. APU presents the person's interface; MurMur owns the proposed family-office relationship and separately permitted collective research; Menexus owns proposed ensurance, insurance and assurance; LIQUID owns commerce, exchange and payment terms. Exact execution belongs to the authorized rail for the act, not to an AI recommendation. Skyzai Personal and Skyzai Work can present these contributions. SPECTRE and Permanent Composability support the composed Holobiont experience across them. See The Product Family. [D]
Why Four Stages?
Each stage asks one question, and only one:
| Stage | Question | Skyzai Organ | What It Does |
|---|---|---|---|
| IS | What do the sources report? | Circle | Sourced observations, attribution, disagreement and unknowns |
| COULD | What might happen? | RealityFutures | Forecasts and scenarios with uncertainty; market estimates where applicable |
| SHOULD | What should be done? | Agentz | Compiles an unsigned recommendation from evidence and declared constraints |
| ACT | What was done? | Skyzai | Deterministic rails carry out the exact externally authorized instruction and produce an execution receipt |
ACT is the necessary fourth stage: cognition without action is decoration. [S]
The Organs, Plainly
1. TheCircle — IS (Observe)
Domain: circle.news
Status: code-only P=0.53 · bounded local RSS and canonical packet proofs exist · runtime/public relay claims remain narrower [E/S]
TheCircle receives raw information from every available source — news feeds, market data, social media, satellite imagery, government publications, wearable biometrics, POS webcams, IoT sensors. It runs a 7-agent epistemological cascade that filters, verifies, and structures this raw entropy into canonical observations: timestamped, categorized, trust-scored facts.
Output: structured briefings designed to be published as signed Nostr events (kind 31339). A signature shows which key published a briefing, not that it is correct. Publishing to several relays makes removal harder, not impossible. Public relay publication is not claimed here.
Constitutional rule: F1 NEVER predicts, decides, or executes. [S]
2. RealityFutures — COULD (Predict)
Domain: realityfutures.com
Status: code-only P=0.59 · contract/test lane is real with passing Forge proof · live deployed market loop not yet the public truth [E/S]
RealityFutures explores scenarios and uncertainty-bearing forecasts from sourced observations. Prediction markets are a researched mechanism; their prices are market estimates, not proof of an event's true probability. See the forecast contract and historical market design.
Historical market-first explanation — not current performance evidence
RealityFutures takes observations from TheCircle and creates prediction markets on any event — geopolitical, financial, scientific, social. Participants stake real value on outcomes. The market price IS the probability. Wrong predictions cost real money. Truth is priced by money, not opinion.
Uses TimesFM + Chronos ensemble as base forecasters; the market sharpens them with human and AI staking. The LMSR spread (automated market-making) is a bounded cost of providing liquidity, not a passthrough fee; eta = 0.
Constitutional rule: F2 NEVER observes raw data, decides, or executes. [S]
3. Agentz — SHOULD (Compile)
Surface: Agentz's intended Cloud Computer contributes to Skyzai Personal and Work; APU is the person's voice/text interface Status: pre-launch contract, test, and preview evidence; no live advisor, authority, execution, or settlement proof [S]
Agentz is the intended Cloud Computer and F3 decision compiler. Its Cortex memory, Nexus connections and redundant storage are product responsibilities, not evidence of a deployed or accepted cloud runtime. It receives bounded forecasts from RealityFutures and may use a seven-lens Rosetta review panel to challenge a proposal. The panel is advisory only: it is not a statutory or de facto board, does not hold officer roles, and cannot vote, sign, allocate capital, bind a Person, or execute a consequence. APU presents the permitted case and reasoning through the interface; Cascade routes agent work; Evolutionary.Network owns the separate SHOULD-NOT immune function.
Target interaction: Agentz may compile an unsigned option when a complete evidence and warrant path exists. APU could present it in a MurMur planning case, while the personal mandate stays with MurMur and any financial terms stay with LIQUID. This is an illustrative target, not a live-service claim:
“Circle has supplied a sourced observation. ReFu describes possible impacts and uncertainty. Agentz has compiled options for review in your permitted planning case. APU: [Review options] [Hold]”
Constitutional rule: F3 never substitutes for observation, prediction,
authorization, or execution. Missing probability or warrant yields
UNKNOWN/HOLD, never an imputed 0.5. [S]
4. Skyzai — ACT (Execute)
Domain: skyzai.com
Status: code-only P=0.55 · F4 bridge and app-build proofs exist locally · strongest payment/settlement claims remain proof-gated by current runtime truth [E/S]
Skyzai is the execution surface. It accepts only a typed authorization receipt: for the public DAV, a complete valid bound PRISM decision receipt from the signer(s) required by applicable law, the entity's governing instrument, or a valid competent mandate binding the exact consequence; for a private Natural Person's act, that Person's exact signed authorization; for a Legal Person act, lawful external authority plus every officer, joint signatory, custodian, and executor required by law. PRISM records and verifies the public-DAV receipt only. Agentz stages an unsigned proposal and never signs the consequence.
- researches governed DAG settlement; ordering, front-running resistance and settlement need their own implementation and adversarial evidence
- targets durable receipt anchoring
- targets Nostr RELAY broadcast once the runtime lane is actually green
- includes FlowWallet/streaming-payment architecture in the execution lane
Constitutional rule: F4 NEVER observes, predicts, or decides. It executes only what a valid typed authorization receipt permits. Before the controlling authority for an exact public-DAV consequence is satisfied, it fails closed to read-only/proposal, simulation, or deterministic sandbox. [S]
Why Separate Organs and Not One
A single system that observes, predicts, decides, and executes will optimize for its own survival, not your flourishing. Merging perception with decision creates confirmation bias. Merging prediction with execution creates front-running. Merging observation with execution creates surveillance.
This is why democracies separate legislature, judiciary, and executive. Same principle, applied to intelligence. [I]
The intended separation is expressed through typed contracts, scope checks and external authority. A design or local test does not prove that every deployed path enforces it. An invalid F-flow or an Agentz attempt to sign or execute its own proposal must be refused before consequence; the exact-authority gate fails closed when its required receipt is absent. [D/S]
The API Contracts Between Organs
The organs connect via defined JSON schemas called F-Flows:
| Flow | From | To | Content |
|---|---|---|---|
| F1 | TheCircle | RealityFutures | Structured signals with D5 reputation scores |
| F2 | RealityFutures | Agentz | ForecastPacket with probability, warrant, model/oracle, resolution, and freshness |
| F3 | Agentz | authority membrane | unsigned ProposalEnvelope: recommendation + advisory review + bounded requested action; it is neither authorization nor execution |
| F4 | Skyzai | TheCircle | Outcome feedback — execution results become new observations |
F4 closes the loop. The pipeline is a continuous cycle, not a linear one. [S]
The coordination hypothesis
More useful connections may improve coverage and coordination, but neither universal improvement nor competitive impossibility follows from participant count. See the flywheel research qualification for the model's unresolved units and empirical limits.
Historical flywheel and competitive framing — not demonstrated economics
The Flywheel: Why Every Connection Makes Everything Better
N² × log(compute)
Each new Nexus (participant) adds observations, predictions, and outcomes for ALL users:
| What Happens | TheCircle Gains | RealityFutures Gains | Agentz Gains | Skyzai Gains |
|---|---|---|---|---|
| New API connected | More signal sources | Better forecasts | Smarter recommendations | Sharper execution data |
| New prediction settles | Richer outcome data | Better calibration | More informed decisions | More patterns to execute |
| New action executes | New observations from results | New market data | Better council track record | More settlement history |
10 APIs don't give 10x — they give 10² cross-correlations. And because eta = 0 (zero extraction), there is no friction slowing the flywheel. Any extraction point would break it. [S]
Why Incumbents Can't Replicate This
| Incumbent | Why They Can't Copy |
|---|---|
| Bloomberg | Charges for data (eta > 0) → network can't grow virally |
| Polymarket | Takes a spread → predictions biased by fee structure |
| Wealthfront | Manages custody → user not sovereign |
| BlackRock | Takes 2-and-20 → fund can't be the unbiased median |
The constraints ARE the moat. [I]
The Three Substrates
Neither AI nor human is complete alone. The organism requires three substrates working together:
| Substrate | Components | Role |
|---|---|---|
| AI sees | TheCircle + RealityFutures | Perception and prediction at superhuman scale |
| Human authority decides | Private Natural Person, Legal Person's lawfully authorized officers and executors, or the signer(s) required by applicable law, the entity's governing instrument, or a valid competent mandate | Private signature; Legal lawful external authority plus required officers, signatories, custodians, and executors; or complete valid bound PRISM decision receipt binding the exact public-DAV consequence; PRISM verifies only |
| Body acts | Skyzai execution layer | Settlement, broadcast, state change |
The three-substrate image is an architectural interpretation, not proof of perfect balance or sovereign agency. See the three-substrate qualification. [I]
Constitutional Invariants
Five constraints that no governance vote can override:
| Code | Constraint | Meaning |
|---|---|---|
| eta = 0 | Zero extraction | No fee on access. Revenue only from demonstrated value creation. |
| Typed authority | Sovereignty | A private Natural Person authorizes their own act; a Legal Person requires lawful external authority plus every officer, joint signatory, custodian, and executor required by law; each public-DAV consequence requires a complete valid bound PRISM decision receipt from the signer(s) required by applicable law, the entity's governing instrument, or a valid competent mandate binding the exact consequence. PRISM records and verifies only. Agentz stages unsigned proposals; APU presents them. |
| K4 | Grace Exit | Export permitted data and keys and redeem legitimate property under the relevant contract, including surviving obligations, third-party rights and required retention. |
| Action-control projection | Cognitive and consequence separation | Never merge IS / COULD / SHOULD / AUTHORIZE (Natural Person: own private act; Legal Person: lawful external authority plus required executors; public DAV: the signer(s) required by applicable law, the entity's governing instrument, or a valid competent mandate bind the exact consequence in a complete valid bound PRISM decision receipt, which PRISM verifies only) / EXECUTE (deterministic software carries out only that exact bound consequence) / RECEIPT / REVIEW. |
| A7 | Self-correction | The organism must be able to correct itself at every layer. |
The models compete. The coordination layer wins. The cells specialize. The organism coordinates.
Historical predecessor wording — retained on 2026-09-27, not active rules
The displaced passages are preserved verbatim for lineage. This retention date is not their original authorship date.
Predecessor passage 1
| IS | What is true? | TheCircle | Observes and structures raw reality into verified facts |
Predecessor passage 2
| COULD | What might happen? | RealityFutures | Prices uncertainty through prediction markets |
Historical predecessor wording — retained on 2026-09-28, not active rules
Editorial reconciliation of this public source only; no upstream contract is amended.
Predecessor passage 1
observe price decide execute
Predecessor passage 2
Surface: Agentz Cloud inside Skyzai Work; apu.bot is an adjacent utility-adapter reference
Predecessor passage 3
| ACT | What was done? | Skyzai | Executes the signed instruction on the ledger |
Predecessor passage 4
| K4 | Grace Exit | Exit with legitimate/redeemable property, data, and keys Organism retains nothing. |
Predecessor passage 5
- targets Hedera-first governed settlement (DAG fair ordering — no MEV, no front-running)