BKB Techies is approximately 90% operated by a local autonomous AI agent swarm, orchestrated entirely in-house on our own hardware — no third-party AI APIs, no cloud LLM subscriptions, no hidden automation. We are fully transparent about this. Every agent, its role, its memory, and what humans oversee is documented below.
Most agencies quietly use AI to cut costs and pretend otherwise. We made the opposite choice: build a real autonomous system, be completely transparent about it, and invest the time saved by automation directly back into quality control and client focus.
Our AI agents do not replace human judgment — they eliminate repetitive execution work (crawling, drafting, formatting, scheduling, deploying) so the humans in the loop can focus purely on strategy, creativity, and editorial standards that AI cannot replicate.
Every piece of content, code, or campaign that leaves BKB Techies passes through a human-in-the-loop veto checkpoint. No agent publishes anything autonomously to production without explicit approval from our editorial team. The swarm proposes; humans decide.
Six specialized departments. Each agent has a defined scope, persona, toolbelt, and DRI ownership. They collaborate cross-functionally — never in silos.
RSS feeds, technical publications, and targeted search queries are pulled. ECHO verifies source authority and extracts raw metrics. GAIA maps search entity coordinates for that campaign's topic.
⚡ Fully AutomatedTransforms crawl output into a detailed, metrics-dense technical briefing. Enforces the BKB Techies editorial voice: confident, evidence-heavy, no marketing fluff.
⚡ Fully AutomatedLine-by-line diff of the draft against raw source text. Any unverified statistic, fabricated quote, or misattributed claim triggers an auto-rewrite instruction back to SOLUS — not a failure.
↩️ Self-Correction LoopFLUX coordinates the social card template and brand alignment. PIXEL renders and compresses the final assets programmatically via Pillow — no generative image AI involved in production assets.
⚡ Fully AutomatedAdapts the verified briefing into platform-optimized social updates: X thread structure, LinkedIn professional tone, Threads casual format, YouTube Community card layout.
⚡ Fully AutomatedARIA packages the complete content bundle — article, social posts, assets — and delivers it to the Open WebUI review dashboard. Nothing publishes without explicit human approval.
👁️ Human-in-the-LoopUpon approval, PRISM uses Playwright-driven browser sessions to publish social content. FORGE simultaneously rebuilds the static blog index, updates sitemaps, and rsync-deploys to Hostinger.
⚡ Fully AutomatedThe swarm is not a static script. It maintains persistent memory across runs, updates itself with real-world signals, and improves its own editorial standards over time.
A local SQLite database (news_memory.sqlite) stores every article ever published, its performance metrics, source reliability scores, and flagged corrections. Agents query this before any new run to avoid duplication and build on prior coverage.
NEXUS runs on a scheduled cron every 6 hours, re-crawling 30+ curated RSS feeds across Hacker News, MIT Tech Review, Google Developer Blog, TechCrunch, and Indian tech publications. The knowledge base updates continuously.
GAIA monitors Google Search Console API data, tracks SERP ranking shifts, and feeds performance signal deltas back into the content prioritization queue — so the next batch of articles doubles down on what is ranking and gaining traction.
VERA maintains a running log of every rejected draft and the reason for rejection. The BLOG_GUIDELINES.md file is updated after pattern-based analysis of failure modes — the quality gate tightens automatically over time.
PRISM scrapes comments and engagement data from published social posts. VERA categorizes sentiment, SOLUS drafts replies, and ARIA reviews them — keeping community engagement sharp and on-brand.
This very transparency page and the underlying agent roster are versioned. As we add new agents or retire roles, the swarm updates its own documentation and reflects the changes here automatically.
No cloud APIs. No subscriptions. No data leaves our machines. The entire swarm runs on consumer-grade hardware, proving that autonomous AI agencies do not require enterprise infrastructure.
| CPU | AMD Ryzen 7 |
| GPU | NVIDIA RTX 4050 (6GB VRAM) |
| OS | Ubuntu Linux |
| VRAM Strategy | Sequential on-demand loading |
| RAM | 16GB DDR5 |
| Framework | Ollama (local) |
| Primary Model | qwen2.5-coder:7b |
| Secondary Model | gemma3:4b |
| Routing | Scoped system prompts |
| Interface | Open WebUI (port 8080) |
| Article Store | Flat-file PHP (zero database) |
| Agent Memory | SQLite (local) |
| Session Auth | Playwright storage_state.json |
| Image Engine | Pillow (programmatic) |
| Video | Remotion (Node.js local) |
| Hosting | Hostinger LiteSpeed |
| Deploy | Rsync over SSH (port 65002) |
| Cache | LiteSpeed Edge Cache |
| CI Trigger | FORGE agent (local cron) |
| SSL | Let's Encrypt (auto-renew) |
BKB Techies operates as a hybrid human-AI organization. Approximately 90% of our operational workflows (content research, drafting, formatting, scheduling, social publishing, and deployment) are handled by our local autonomous AI agent swarm. The remaining 10% — strategic decision-making, client relationships, final editorial oversight, and legal/financial matters — are handled exclusively by human team members.
All AI-generated content is reviewed by a human editor before publication. No client data is processed by third-party AI APIs without explicit written consent. We comply with applicable AI transparency guidelines and recommend reviewing our Privacy Policy for full data handling details.
Last Updated: June 2026 | Agents Active: 13 | Version: 1.0.0