an Oracle Vision
METHOD MOIRAI✦ ✦ ✦
Governed multi-model orchestration. One brief. Every medium. Every cost visible.
Multiple AI stages review your brief in sequence, with quality checks & a saved archive — no single model writes the final output alone.
Part 1 · The Flow
Before any run starts, the Workshop draws out your actual intent through a conversational intake. As you describe what you need, MOIRAI builds a structured Cauldron beside the chat — title, summary, deliverables, approach, constraints, feasibility.
Talk or type. The Oracle voice reads back each Workshop reply so you can have a fully spoken intake conversation.
The Cauldron surfaces on the right column once enough context is gathered. You review it, edit any field, & either Approve or reject before a weave can begin. Approval is required — the seal gate enforces it.
Submitting fires the three-fate orchestration pipeline. A streaming feed shows each stage completing in real time — cipher glyphs reveal each character as the AI writes. The run is archived on completion.
Creative outputs — image, video, voice, music, sound effects — ship alongside the written synthesis in the same run.
After the weave, the Atropos QA pass grades the output. State machine: draft → pending_approval → approved → pending_review → sealed. Sealed runs are immutable — export-ready.
Every run lands in /archive. Export as branded HTML, Markdown, or PDF-ready. Soft delete → trash → permanent. Full search.
✦ · ✦
“Even the thread begins somewhere: what is spun is not yet fate, only possibility.”
✦ · ✦
Part 2 · Orchestration Tiers
Each tier runs the same three-fate chain — Clotho (gather), Lachesis (measure), Eunoia (refine) — with increasing isolation & adversarial pressure.
Sequential pass: Clotho → Lachesis → Eunoia. Faster cycles, lighter cost. Each stage sees the prior output.
Blind parallel isolates — each fate runs without seeing the others — then two reconciliation passes, Eunoia final refinement, optional adversarial critique. Reduces model confirmation bias.
All three providers run in parallel then synthesized. Historical mode; Tier 2 is generally stronger.
✦ · ✦
“Every run is a question posed to three minds. The Seal is their answer, reconciled.”
✦ · ✦
Part 3 · Creative Outputs
A single run can generate written synthesis alongside any combination of creative outputs. Plan access controls which media are available.
Part 4 · Tools
Part 5 · Governance
Every run emits structured audit events — model routing decisions, token share fractions, creative COGS, escalation flags. Operators view these via the internal usage observability endpoint & governance alerts dashboard.
The escalation halt system parks runs that contain escalation-flagged content for operator review before delivery. Accept continues; revise marks the run for rework.
Billing is Stripe-backed with a nightly cron reconciler. Usage limits, unseal caps, & plan-gated features are enforced server-side — not by frontend flags.
Part 6 · External API
Use the external API when you need the pipeline inside your own application — batch processing, automated workflows, or programmatic delivery to your clients. The same three-stage weave runs server-to-server, with identical governance gates & audit trail as a manual run. Nothing changes about what runs; only how you trigger it.
Server-to-server access via POST /api/v1/generate — synchronous or async with a callback URL. API key auth via Authorization: Bearer. Keys are rotated server-side without a code deploy.
POST /api/v1/generate
Authorization: Bearer <MOIRAI_API_KEY>
{
"prompt": "string",
"moiraiTier": "tier1" | "tier2" | "legacy",
"collaborationMode": "synthesize" | "compare" | "debate"
}
// Sync response
{ "ok": true, "output": "...", "moiraiTier": "tier1", "durationMs": 12400 }
// Streaming (SSE)
POST /api/v1/generate/stream → text/event-streamTiming-safe key comparison. MOIRAI_API_KEY is scoped per-account & never logged.
✦ · ✦
“Bring the strand to the Loom. What you receive will be what was already there — made clear.”
✦ · ✦
Open MOIRAI, describe what you need in the Workshop, & review the plan before you run. Every run is saved, exportable, & traceable.