Oannes Content Engine — Design Spec

Date: 2026-06-15 Status: Approved direction; pending spec review → implementation planning Brand: Oannes · handle @oannescode (Instagram held; claim/confirm on TikTok + YouTube)

1. What this is

An automated, near-hands-off short-form content engine that turns the existing Oannes knowledge graph (244 episodes → 2,176 claims, 1,905 people, 555 events, all cross-linked) into a daily stream of original, faceless social videos and image posts about the mysteries of mind, the universe, and everything in between — published to TikTok + Instagram Reels + YouTube Shorts, monetized through stacked revenue streams.

The repurposing target is not anyone’s footage. It is the structured claims the pipeline has already extracted. Competitors clip one video at a time; this engine starts from a queryable graph and can synthesize across episodes (“6 witnesses, same craft”) — which is simultaneously the most engaging format and the most legally defensible.

Priorities (in order, for tie-breaks)

1. Money, fast — optimize for the fastest realistic path to revenue.
2. Minimal hands-on — target ~2 minutes/day of human effort (the approval tap).
3. Fun / exploration — the subject matter is genuinely interesting; keep it that way.

Honest expectations

Faceless channels typically take 1–3 months to gain traction. Pure platform ad-payouts are slow and small. “Money fast” therefore depends on the affiliate + own-product streams that earn while the audience is still small, not on view-payouts. The design front-loads those.

2. Scope

In scope (v1)

• Curate → script → media → compose → approve-from-phone → publish → learn loop.
• Three output formats from one graph claim: video reel, image carousel, quote card.
• Four platforms, one render fanned out. Lead = Instagram (2026-06-15: the TikTok account was banned on creation and the appeal failed on a transient server error — TikTok is deferred until restored). Live order: IG Reels → YT Shorts → FB Reels, with TikTok re-added if/when unbanned. IG/YT/FB are connected in Blotato; TikTok is not. FB included because its older audience over-indexes on UFO/paranormal interest, fanned out at ~zero extra cost.
• Cadence 1–2 posts/day, sustainable and architecturally able to scale to ~5/day for ~the same cost.
• Cambium Rhythms fires the daily cycle; approval gate is the only human touch.
• A learning loop that pulls performance back into curation.

Non-goals (v1 — explicitly deferred)

• AI-generated moving video (Veo/Sora/Kling) — deferred to v2 as occasional accent b-roll only.
• Synthetic avatar host (an “Oannes” face) — deferred to v2 as a brand power-move once the machine works.
• Fully autonomous publishing — rejected for v1; the human gate is the brand/legal backstop.
• Reposting source footage / clip-and-repost — rejected entirely (account-killer; see §4).
• New ingestion features — the existing Oannes pipeline already produces the graph.

3. Editorial voice

“True believer,” constrained by mandatory attribution.

Highest-engagement stance, made safe by a single hard rule baked into the script step:

Every claim is attributed to the person who made it. “Bob Lazar says he worked on alien craft at S-4,” never “aliens are real.” The graph already stores claims as attributed-to-a-person, so this is free to enforce.

Attribution is not just legal cover — it is the actual format of the biggest accounts in this niche, and it keeps Joshua from personally vouching for contested claims. The approve-from-phone gate is the final backstop.

4. Content & IP model

The chosen model (original, graph-sourced, attributed, original media) sits almost entirely in “green.” The “yellow” rules are cheap to enforce because they are just prompt constraints and asset-source rules.

🟢 Green (foundation) - The claims/facts/ideas themselves — not copyrightable. - Original synthesis & cross-claim connections — original work we own (the moat). - Original AI imagery, original scripts, licensed/royalty-free music.

🟡 Yellow (allowed with discipline — enforced in pipeline) - Never copy source expression. Script step paraphrases/restructures; never lifts transcript sentences. (Prompt rule.) - Real people’s names/likenesses. Reporting what someone actually said is protected journalism. But: use stylized/illustrative depictions, not photorealistic deepfakes of living people; never imply they endorse an affiliate product. - Defamation shield = attribution + framing (“X claims/alleges/says Y”). Never assert a damaging falsehood as fact about a living person. Never fabricate quotes. - AI-content disclosure toggles ON for TikTok/IG/YouTube (penalty-avoidance).

🔴 Red (account-killers — never) - Reposting source video/audio (fingerprinted → strikes → account loss). - Ripping trending commercial music onto monetized posts. - Fabricating quotes attributed to real people. - Photorealistic deepfakes of real living people.

5. Architecture

A 5-stage pipeline, orchestrated by Cambium Rhythms, with a single human gate at stage 3.

EXISTING (Oannes pipeline)            NEW (content engine)
graph: 2,176 claims  ─────────▶  1. CURATE  →  2. SCRIPT  →  media gen  →  compose
   people/events/links               │            │          (ElevenLabs +    (Remotion)
                                      │            │           Ideogram/FLUX)      │
                                      ▼            ▼                               ▼
                                   pick high-   attributed,                 finished asset(s)
                                   hook claim   hook-first script                  │
                                   + format     + SEO caption/tags                 ▼
                                                                       3. QUEUE → phone (Discord/Telegram)
                                                                          preview + Approve / Reject  ◀── only human touch
                                                                                   │ (approve)
                                                                                   ▼
                                                                       4. PUBLISH (Blotato → TikTok+IG+YT)
                                                                          + affiliate links in bio/desc
                                                                                   │
                                                                                   ▼
                                                                       5. LEARN: pull views/retention
                                                                          back into graph → curation
                                                                          favors winning veins

Components (each independently testable)

1. Curator — queries the graph for a high-“hook-potential” claim or cross-claim thread; decides output format (single quote → quote card; pattern across witnesses → carousel; narrative → video). Input: graph + performance history. Output: a content brief (claim id(s), format, angle).
2. Scriptwriter — Claude CLI turns the brief into an attributed, hook-first script (~120–150 words for video; slide copy for carousel) + platform SEO (caption, hashtags, title). Enforces the attribution + no-lifted-expression rules.
3. Media generator — ElevenLabs voiceover (video only); Ideogram Turbo for text-bearing cards, FLUX-schnell for cheap b-roll backgrounds; selects a music bed from the Uppbeat library.
4. Compositor — Remotion (self-hosted on the Mac Mini) renders the branded vertical video (captions, Ken-Burns motion, music) and the still formats (carousel slides, quote cards) from shared templates. Output: 1080×1920 MP4 and/or PNG set.
5. Approval queue — posts a preview (video/images + caption) to a Discord/Telegram bot; Joshua taps Approve/Reject. Reject can carry a one-word reason that feeds the curator.
6. Publisher — on approve, pushes to Blotato, which fans out to TikTok + IG Reels + YT Shorts (with AI-disclosure flags and affiliate links placed per platform).
7. Learning loop — pulls per-post views/retention/saves back, annotates the source claim(s) in the graph, so the curator favors proven veins and formats.

Orchestration & state

• Cambium Rhythms fires the daily cycle (stages 1–3). Stage 4–5 run automatically on approval.
• State (briefs, drafts, approvals, published ids, performance) persists to Postgres (Cambium already uses native Postgres on localhost:5432) — no in-memory-only state for anything important.
• Runs on the existing Mac Mini daemon host; Remotion renders locally = ~$0 compute.

6. Tool stack

All-in ≈ $70/month (~$60 fixed + ~$10 usage) at 1–2 posts/day. Per-piece marginal cost $0.10–0.24; scales to ~5/day for nearly the same money.

Render-engine decision: Remotion self-hosted (not Shotstack). Remotion’s paid license floor applies only to companies with 4+ employees; as a solo operator it is free, runs on the Mac Mini beside Cambium, and integrates natively with the React/TS stack. Trade-off accepted: we maintain the render setup instead of paying for managed rendering.

7. Monetization (all four, sequenced)

1. Affiliate (day 1) — books (the graph has 429 Book nodes → Amazon), gear, niche courses/subscriptions. Earns while small. Links in bio + descriptions, auto-inserted by the publisher per the post’s topic.
2. Own digital product (build in parallel) — highest margin/ceiling. Candidates: a “mysteries” membership/newsletter, a Bold Little Oracle tie-in, an ebook/field-guide generated from the graph. The channel is top-of-funnel.
3. Platform payouts (as thresholds clear) — TikTok Creator Rewards, IG bonuses, YouTube Shorts/Partner. Easiest conceptually, lowest per-view; treat as gravy.
4. Sponsorships (once audience exists) — highest value, needs leverage. Attribution-first/brand-safe voice keeps sponsors reachable.

The graph is the unfair advantage at every stage: it can target affiliate offers to a post’s exact topic and generate product content on demand.

8. Platforms & accounts

• Instagram-led (TikTok deferred — banned on creation 2026-06-15, appeal pending). Fan out one render to IG Reels + YT Shorts + FB Reels (FB Page); re-add TikTok if/when the ban is lifted (retry appeal in a few days; if permanent, create a fresh account carefully — no VPN, warm up by hand-posting for a week before connecting to Blotato’s API).
• Handle @oannescode (held on Instagram; YouTube + FB Page connected).
• Connected in Blotato as of 2026-06-15: Instagram, YouTube, Facebook. TikTok NOT connected.
• Accounts must be Business/Creator type (FB needs a Page, not a profile) — required for API publishing + insights.
• Platform connection is done inside Blotato (OAuth); no per-platform developer API keys required.
• AI-content disclosure ON for every post.

9. Rollout phasing

Built as vertical slices so revenue/learning starts before the whole machine exists.

• Phase 0 — Wiring & proof. Create accounts + keys; one .env; a connection-test script that pings every service (ElevenLabs, image API, Blotato, etc.) so every asset is proven before building.
• Phase 1 — Stills MVP (cheapest, fastest). Curator → scriptwriter → Remotion quote cards + carousels → approval bot → Blotato to one platform (Instagram, since TikTok is banned). No voice/video yet. Proves the loop end-to-end and starts posting within the first build cycle. (Phase 0+1 build COMPLETE 2026-06-15; live-publish flip pending — see below.)
• Phase 2 — Video. Add ElevenLabs voiceover + b-roll + Remotion video composition.
• Phase 3 — Fan-out. Add IG Reels + YT Shorts + FB Reels via Blotato; per-platform captions/affiliate placement.
• Phase 4 — Learning loop. Pull performance into the graph; curator favors winning veins/formats.
• Phase 5+ (v2, deferred). AI accent b-roll; the “Oannes” avatar host; own-product funnel.

Each phase gets its own implementation plan (writing-plans) → execution.

10. Risks & mitigations

11. Open items (resolve during planning, not blocking)

• Confirm which @oannescode accounts already exist (TikTok/YT/FB).
• Blotato vs self-hosted Postiz final pick (cost vs control) — defaulting to Blotato for minimal-hands-on.
• Specific affiliate program(s) to start with.
• The first “own product” to point the funnel at.