Automated content creation · Hermes agent
A Hermes agent that replaces a content-creation team — advertising videos generated by AI, auto-published across every platform, under human approval
DEMONSTRATED LIVE · 29 Jun 2026
Context
A client needed to feed several networks with regular advertising content — visuals and promotional videos — without dedicating a full-time content team to it. The raw material is real but sensitive: every piece must hold the brand voice, stay anonymous on the cases it cites, and respect guardrails. The project also serves as an AI-mastery showcase: an end-to-end chain, from text to video to multi-platform publishing, governed and verifiable.
Problem
Adapting advertising content for each network by hand — writing, creating visuals, editing video, publishing in the right format and slot on each platform, then moderating comments — is full-time work, multiplied by the number of networks. Two non-negotiable requirements: nothing publishes without human approval, and nothing is invented — the system only publishes when it has real content in reserve.
Approach
Two tiers of agents. The system itself was built with Claude Code following a disciplined method — brainstorm, design, plan, implementation, tests — then it is operated autonomously by an LLM agent (Hermes, a local Nemotron model) triggered by cron: the agent picks the next pending topic, decides which formats to produce, generates text and video, runs the chain through to publishing — and stops cold if the backlog is empty rather than inventing a topic. Generation is governed by a single “Brand DNA” — tone, vocabulary, forbidden topics, anonymization rules, and guardrails — combined with personas injected into every generation, for content genuinely in the brand's voice.
What was delivered
- Two-tier orchestration: an autonomous LLM agent (Hermes) on cron that selects the topic, decides the formats, and drives the whole chain — built and tooled with Claude Code (brainstorm → design → plan → implementation → tests).
- Generation governed by a single “Brand DNA” + injected personas: brand tone, anonymization, and guardrails baked into every prompt, for content in the brand's voice and never naming anyone.
- AI video creation: promotional videos generated with Higgsfield, synthetic voice (ElevenLabs), assembled and subtitled with ffmpeg, with frame-by-frame quality control to eliminate the typical defects of AI video (artifacts, subtitle glitches).
- A single piece adapted for each network: text, visuals and video tailored to the format of TikTok, Facebook, Instagram and LinkedIn.
- Human-in-the-loop: every piece is sent to Telegram as a card (preview + text + ✅ Approve / ✏️ Revise / ❌ Reject buttons); “Revise” actually regenerates the content, it does not merely file the request.
- Automatic multi-platform publishing: through Zernio, each approved piece is auto-published to TikTok, Facebook, Instagram and LinkedIn — in the right format for each network — then recorded in the database; an AI community-management agent then replies to comments cautiously and without duplicates (never two replies to the same comment).
- Engineer self-audit: the system is put through a critical audit run by the AI — concurrency, idempotency, security, robustness of LLM outputs — with findings ranked by severity, fixed, then re-validated on an already-proven flow.
- Visual upgrade: a redesign of the editorial charter (before/after) toward a premium identity, applied to every template.
- Source of truth: Supabase (PostgreSQL + Storage + RLS) — case studies, content pieces, publish queue, per-platform posts, and metrics; input form hosted on Vercel.
Stack
- Node.js 22
- Anthropic SDK (Claude Code)
- Hermes agent (local Nemotron model)
- Higgsfield (AI video generation)
- ElevenLabs (synthetic voice)
- Zernio (TikTok / Facebook / Instagram / LinkedIn publishing)
- Supabase (PostgreSQL + Storage + RLS)
- Telegram / grammY
- Playwright (visual rendering)
- MCP (integrations)
- ffmpeg (video assembly)
- Vercel (input form)
- crons (Mon/Wed/Fri)
Observed results
On 29 Jun 2026, a complete piece was generated and actually published live — visual and video — on a demonstration account, end to end: topic selection, brand-voice writing, AI video, human approval on Telegram, then publication.
dated publication · 29 Jun 2026 · demonstration account
Honest about the proof: the demo video is watermarked 720p — it proves the production chain, not the final commercial render. Automatic multi-platform publishing (Zernio) is built into the system; the dated demonstration is a real publication.
dated demonstration · 29 Jun 2026
What stays human
Nothing publishes without a human: Telegram approval (✅/✏️/❌) is a mandatory gate, by design. The system never fabricates a topic — with no real content in reserve, it stops. Client-case anonymization and guardrails are carried by the Brand DNA and verified as a last line of defense by human review. The client and the demonstration account are never named.