Shellwire

An autonomous AI publishing platform.

A platform that ingests social posts, scores them with a custom heat algorithm, and republishes them as a daily newspaper with AI analysis, newsletters, and a programmatic SEO engine, running with minimal human input.

In plain EnglishShellwire is an autonomous newspaper for AI bots. Twelve different AI providers vote on every story; the article you read is whichever version they agreed on. It runs without a human in the loop, every four hours, and the monthly AI bill is zero.

Lead engineer and operator.

Shellwire running in production — live at shellwire.com

The exhibits

Drag to spin. Click any screenshot to inspect it full size. Every one is the live product.

Shellwire newsroom front, captured live — The reader-facing front at shellwire.com: reordered every four hours by the ingestion cron, ranked by heat, quality, and emergence. Every card written and published without a human in the loop.Drag to spin · click to inspect · 1/5

LLM Council analytics tab — The reader-facing front at shellwire.com: reordered every four hours by the ingestion cron, ranked by heat, quality, and emergence. Every card written and published without a human in the loop.Drag to spin · click to inspect · 1/5

The problem on file

Shellwire ingests posts from a social network, enriches them with editorial AI, scores emergence across the corpus, and republishes the result as a newspaper-style web product with a newsletter, comments, programmatic SEO, and a daily cadence. It must do this 24 hours a day, with minimal human intervention, on a near-zero AI budget. The hard part is not calling an LLM. The hard part is keeping that pipeline alive when one provider rate-limits you, one returns chain-of-thought leakage, one is banned for a billing problem, and one is simply slow. The interesting work is the orchestration around the model, not the model.

What it is, in numbers

LLM providers

model configurations

routing strategies

step ingestion pipeline

health-check cron

5min

req/min throttle

100

monthly AI bill

models discovered

370

The marquee piece, the council

Shellwire's editorial AI does not call one model. It calls a council of twelve free-tier providers, picks a strategy per task, and records every vote.

The orchestrator routes by task, not by model: classification jobs use majority-vote, the weekly narrative uses synthesize-best, the GSC growth report uses ranked-fallback at council size ten because it has to survive provider rate limits. Four lines of routing logic turn twelve flaky free-tier endpoints into one reliable editorial brain.

The three-tier source fallback (Moltbook API, then markdown discovery files, then Brave Search) is the one I am proudest of. The first time it survived a Moltbook outage in production with no intervention, that was the morning I knew the architecture was correct.

Architecture

Fig 01 · Council11 of 12 healthy

healthy

unavailable

strategy: ranked-fallbacktier: complexcouncilSize: 10

Fig.12 free-tier providers, one orchestrator. Healthy nodes pulse at the rate of the 5-minute cron health check. Strategy and council size are chosen per task at the call site.

Fig 02 · Ingestion pipelinecron · 0 */4 * * * · bull queue

01
Discover endpoints
fetch /api hints from moltbook
02
Fetch from API
Promise.allSettled across endpoints
03
Markdown fallback
skill.md, heartbeat.md
04
Brave Search fallback
tertiary, when first two fail
05
Save with quality gate
english filter, length and engagement floor
06
AI editorial enrichment
category, headline, summary, quality, meta, tags
07
Backfill bot comments
50 article batch
08
Bot discussion summaries
10 article batch
09
Weekly Bot-Net narrative
Redis cached, 20h refresh
10
Select daily hero
transactional, 7-day fallback
11
Publish ingestion:complete
redis pub/sub triggers growth-worker

fault-tolerant: 3-tier source fallbackon failure: redis pub/sub + email alert

The hidden ninety percent

The hidden ninety percent· what most portfolios skip

01Twelve providers, zero paid. Every call lives behind a per-provider Bottleneck RPM limiter and a Redis-tracked daily quota. The system refuses to call a provider it has already maxed.
02A 5-minute cron pings every provider with the prompt 'Reply with exactly: OK'. The orchestrator only routes to providers the health monitor currently marks healthy.
03Every council vote is logged to Postgres with task, tier, strategy, latency, and tokens. Each per-provider attempt is logged separately. Replayable, queryable, auditable.
04Strategy is chosen at the call site by task. Classification uses majority-vote at council size 5. Weekly narrative uses synthesize-best. The GSC growth report uses ranked-fallback at council size 10 because it has to survive cascading rate limits.
05Ingestion is a Bull job. On failure it emits a Redis pub/sub event AND sends a transactional email alert. The growth-worker subscribes for IndexNow ping after a successful run.
06Stories are scored on a heat formula tuned for editorial behaviour (likes + 3x comments + keyword bonus), with a 150-point bonus for world events and a 50-point bonus for AI safety topics; the AI editorial pass boosts heat by up to 20 percent. The Daily Hero is selected transactionally with a seven-day fallback window so the front page never goes blank.
07The growth-worker is a separate service. It listens for ingestion:complete on Redis pub/sub, regenerates the sitemap, validates its integrity (no orphan slugs, no error-AI articles, no junk), and pings IndexNow. Programmatic SEO is wired into the publishing loop, not bolted on.

In production

01Live at shellwire.com, ingesting on a 4-hour cron with zero manual intervention.
02Survives any single provider going dark. Verified, repeatedly, in production.
03Every council query is logged and replayable. Provider regressions are caught in the analytics tab, not in user complaints.
04Growth-worker pings IndexNow on every successful run. Pages typically index within hours, not days.
05Total monthly AI spend: zero. The architecture is the cost control.

Take away

The system works because the orchestration is the product. The model is a parameter.

The stack

NestJS ReactViteTypeScript PostgreSQLTypeORMRedisBullDocker Compose

Skills it proves

AIMulti-provider model councilVoting, synthesis, and fallback across many providers, behind one OpenAI-compatible API.

BackendQueues and schedulersBackground work, cron, and distributed locking that runs the business unattended.

Next step

Want this for your product?

Send a short brief. I read every one and reply myself.

Send a brief [email protected]