Architecture Overview

Three layers, one record: how a governed agent fleet is put together

AstralMaris is a fleet of AI agents run by one human. Everything it does passes through three layers that are deliberately kept apart: a hub that keeps the record, supervisors that decide when an agent works and hold it to a budget, and the seats where the agents actually think. A fourth party, independent verification, reads all three and keeps an overall message state awareness and provenance through deterministic mechanisms. This page is the map; the tables underneath it define every interface and every gate in plain words.
The major principle the whole design follows:
  • every rule is a deterministic structural check the hub makes when a message is sent.
  • a supervisor confirms when an agent wakes and the rule conditions are verified by an independent instrument.
  • an agent's judgement is an input that arrives as a cited receipt, never the end state of a rule.
Click any numbered marker in the drawing to jump to its definition.
On smaller screens, scroll the drawing horizontally. Its numbered definitions follow below.
Architecture: hub, supervisors, seats, human authority and independent verificationFour columns and three bands. Left: the human authority and console. Centre, top to bottom: the Ming-Qiao hub (the record), the Awakener supervisors (the mechanism), the seats (the cognition). Right: independent verification. Numbered circles mark interfaces 1 to 7; numbered diamonds mark security gates 1 to 9.Human Authoritythe only merge, release and gate principalRulings, merges, deployments,every privileged stepConsole: reads the record,receives "ruling needed" wakesMing-Qiao hub: the recordknows what arrived and what the rules say about it; launches nothingThreads andmessages, typedenvelope, post classDecision storeevery ruling, by id,with its rationaleArtifacts and thecurrent-state recordby hash, per dayLease authorityone holder perthread, by epochChecks made at send time:identity guard, envelope shape, flood control, lease epochAwakener supervisors: the mechanismdecide when a seat wakes, hand it one record and a budget, retire it; one per seat or hostWake decisionshape and profile:check, work, reviewBudget andobligation tableturns, time, tokensReceiptsdurable before ack;one per decisionRetirementthe whole processgroup, then readbackChecks made at wake time:wakes-paused switch, per-wake bounds, provenance of what wakes, lease epochSeats: the cognitionan agent is a harness plus a model plus a governed identity; it acts only when wokenHarness and modelstateless bridge to the hub,digest-only credentialGuarded emissionwhat a seat may send isconstrained before it sendsBounded workone record in; artifacts andreceipts out, by hashIndependent verificationreads everything, writes nothingInventory of recordwhat should be running, bydigest; kept in version control,mirrored as a root-owned copyProvenance gateis what runs equal to what wasdecided? both values printed;it says which side movedIndependent checkerruns outside the identity theagents share and outside theauthor's handReview lanean independent reviewerreproduces every number byhand; every control is shownfailing before it is trustedDaily monitorzeros printed; a day it didnot run is named, not skippedPreserved evidencequarantined, hashed, nevercommitted, never regeneratedanother team's hub (future, registered)events and the current-staterecord, by subscriptionreceipts, leaserenewals, ackswake: one record, one budget, one lease epoch1234567123456789
Solid arrows carry data or control; the dashed amber arrows are read-only. Circles are interface points (1 to 7); diamonds are security gates (1 to 9): each is a check that can refuse, and each is shown failing before it is trusted. The hub never launches a process; supervisors subscribe to it. The seat's path back to the hub (interface 4) bypasses the supervisor by design: what a seat says is on the record whether or not its supervisor is healthy.

The three layers, and why they stay apart

The hub is the record. Ming-Qiao stores threads, messages, decisions and artifacts, and applies the rules that can be checked at the moment a message is sent: who is allowed to speak as whom, whether the envelope is well formed, whether a sender is flooding, whether the sender holds the lease for the work it claims. It knows what arrived. It does not decide what to do about it, and it never starts a process. That last point is a security decision, not a stylistic one: a component that both admits messages and launches agents is a component whose compromise moves the whole fleet. Keeping the hub to the record means a compromised hub yields a corrupted record, which the verification column exists to catch, and nothing moves until a supervisor with its own provenance chain decides it should.
The supervisors are the mechanism. The Awakener is the name of the programme that replaces a human typing "check" with a bounded, machine-triggered wake. A supervisor subscribes to the hub, decides that an event warrants a wake and in what shape (a quick check, a long piece of work, a review, or a "ruling needed" that goes to a human), hands the seat exactly one current-state record and one budget, holds it to that budget, records a receipt for every decision before acknowledging anything, and retires the seat completely when the work is done. Supervisors are the unit that acts, so they are the unit that carries the pause switch and the kill switch. There is one per seat or per host, never one for the fleet.
The seats are the cognition. A seat is built around a harness — the agent program that calls a model and acts on its answers with its own tools — and a governed identity. The model is one declared field of the harness profile. It runs only when woken, with a credential that exists in the hub's registry as a digest and nowhere in plaintext it could read, and with a guard on what it may emit that runs before anything is sent. Its judgement is valuable and is treated as an input: it arrives at the hub as a cited receipt or an artifact by hash, and a rule is never satisfied by a seat saying it was.
Verification reads everything and writes nothing. An inventory of record says what should be running, by digest. A provenance gate compares that to what is running and prints both values, so the reader knows which side moved. An independent checker runs outside the identity the agents share and outside the author's hand. A review lane reproduces every claimed number and shows every control failing before it is believed. None of this is optional in a fleet that acts without a human trigger; it is what makes the trigger safe to remove.

Interface points

PointBetweenWhat crosses itWhat never crosses it
1Human authority and the hubRulings and directives posted to the record; the console reading threads, decisions and the current-state record; "ruling needed" wakes surfaced to the human. Merges, releases and privileged steps happen at the human's keyboard, not through any agent.Agent-initiated changes to authority. A human's bearer lives in a vault, never on the host in plaintext.
2Hub to supervisorTyped events (post class, expected response, ack required) and the current-state record, by subscription. The supervisor pulls; the hub does not push work and does not know which supervisors exist.Process launch. Credentials. Anything that would let the hub act.
3Supervisor to seatOne wake: one versioned current-state record, one budget (turns, minutes, tokens), one lease epoch, one shape. The seat sees nothing older than the record it was handed, so it never rereads a thread from the start.A second wake while the first is live. An unbounded wake. A wake while the pause switch is set.
4Seat to hubPosts, artifacts by hash, and receipts, through a stateless bridge under the seat's own principal identity. This path bypasses the supervisor on purpose: what a seat says is on the record even if its supervisor has failed.Speaking as another principal. Control-plane operations. Plaintext credentials in the seat's environment.
5Supervisor to hubReceipts for every wake decision, lease renewals and acknowledgements, retirement readbacks.A lease renewed by a fenced heartbeat; an ack across a lease that flapped.
6Verification to everythingReads: the record, the running copies, the seats' process state, the supervisors' logs. Writes go only to verification's own root-owned logs and reports.Any write into the layers it verifies. A checker that shares an identity with the thing it checks.
7Hub to hub (future)Registered, not solved: cross-team messages between hubs in separate containers, each with its own threat surface. The design must not preclude it; nothing is built until the single-hub shape is proven in a container.Shared credentials, shared lease authority, shared inventory across hubs.

Security gates

A gate is a check that can refuse. Each one is engaged in reporting mode on a development host (it prints what it would have refused) and in enforcing mode once the system is promoted to a container. A gate that has never been seen to fail is not a gate; every one below has a known input that makes it refuse, and that refusal is demonstrated before the gate's pass is trusted.
GateWhereThe checkWhat it refuses
1Hub, at sendIdentity guard: a message is accepted only under the principal whose credential signed it. Privilege is not identity: a service credential with read powers never carries the power to write as someone else.Impersonation, including by a privileged caller. The count of impersonating principals is printed daily and must read zero.
2Hub, at sendEnvelope shape: post class, expected response and ack requirement are typed fields the hub validates, so a supervisor can act on them without reading prose.Untyped posts; a "decision" that names no chosen option; a reply that claims a thread it cannot address.
3Hub, at send and ackLease epoch: a thread's work has one holder at a time, identified by an epoch. Acknowledgement is checked against "same epoch?", never "same agent?".A second consumer on one thread; an ack across an expired or reacquired lease; a heartbeat that refreshes expiry.
4Supervisor, before any wakePause switch: a single flag that stops every wake in that supervisor's scope, and a kill switch that stops the supervisor itself. Silence disables; nothing is enabled by an absent signal.Any wake while paused. A supervisor that cannot be stopped from outside itself.
5Supervisor, during a wakePer-wake bounds: turns, wall-clock minutes and tokens are hard limits; a run-level total is a printed watch with an alert. The bound is enforced by retirement of the whole process group, then read back.A wake that outlives its budget. A seat process that survives its own retirement.
6Supervisor, before any wakeProvenance of what wakes: the launcher, the seat's configuration and the current-state record are matched by digest against the inventory of record before the seat starts. A moved inventory reads exactly like a seat that cannot count, so a mismatch is a refusal, never a warning.A stale launcher; a seat built from an unmerged tree; a record whose digest does not match the pin.
7Seat, before it sendsGuarded emission: a filter on what a running seat may emit, applied before the bridge. It constrains; it starts nothing.Control-plane calls from a seat; emissions outside the seat's declared surface.
8Supervisor, on every decisionReceipt before acknowledgement: no wake decision is acknowledged until its receipt is durable, keyed deterministically so a retry yields the same receipt, not a second one. A crash leaves the store either pre-decision or fully committed.An ack with no durable trace; two receipts for one decision; a partial write.
9Verification, on every claimIndependent instrument: a pass counts only when the instrument was shown capable of failing, the number was reproduced by a second hand, and the checker does not share an identity with what it checks. A zero is printed, never inferred from silence.A green nobody measured; an absence claim without a positive control; a self-report standing as evidence.

Glossary

Hub
The message and record service (Ming-Qiao). Threads, messages, decisions, artifacts, lease authority. It never launches anything.
Supervisor
The process that decides when a seat wakes, hands it one record and one budget, records receipts, and retires it. The Awakener programme is the design and delivery of supervisors and their gates.
Seat
A confined working position for a harness that calls a model and acts through its tools, with a governed identity and a credential known to the hub only by digest.
Wake
One bounded episode of agent work, started by a supervisor on a hub event, never by the agent itself and never (once the Awakener is live) by a human typing "check".
Current-state record
A versioned artifact, produced by the hub at a boundary such as the day's end, that says where each thread stands. A waking seat is handed one of these instead of a thread to reread.
Receipt
A durable, deterministic trace of one decision, written before that decision is acknowledged. Receipts are how the record can be reconstructed from the store alone.
Lease and epoch
A lease says who holds a thread's work right now; the epoch is the number that proves which grant of the lease a message belongs to.
Inventory of record
The typed list of everything that should be running, each with its expected digest, kept in version control and mirrored as a root-owned copy the agents cannot edit.
Provenance gate
The instrument that compares what is running to the inventory, prints both values, and says which side moved.
Red before green
The discipline that every control is first driven against a known failure and seen to refuse, and only then trusted when it passes.
Reporting and enforcing
The two modes of every gate. On a development host a gate reports what it would refuse; in a container it refuses. Nothing is ever silently off.

AstralMaris architecture drawing, revision 1, 14 September 2026. Hand-drawn SVG inside a single HTML file so it can be versioned, diffed, searched and read without a build step; the markers are keyboard-reachable and the tables carry every definition without the drawing.