Wassim AI Facebook Reels - Super Glasses Report

Reel: https://www.facebook.com/reel/1358570179462150/

What it does: Free/open-source browser, desktop, and mobile video editor; reel emphasizes privacy, no watermark, plugin/headless/MCP direction.

What we can do: AI Robot Builds video-edit proof lane: connect generated assets, B-roll, captions, and review packs without forcing CapCut subscription.

How to do it: Clone/test OpenCut, cut one 15s heart-robot or Hi-Tech clip, document export limits, then build a local 'AI Video Edit Desk' proof with approval gate before client use.

Guardrail: Verify OpenCut feature maturity before promising MCP/headless automation.

Web links: OpenCut docs · OpenCut GitHub

Reel: https://www.facebook.com/reel/1864245744270405/

What it does: Rust/Python vector index using TurboQuant-style compression for smaller local RAG indexes.

What we can do: Cletus Brain / client knowledge-base cost reduction test: local docs, call logs, site pages, estimates, and training material without a managed vector DB first.

How to do it: Install in a sandbox, embed one existing report corpus, compare recall/speed/RAM against current simple search or Chroma, then only adopt if retrieval quality holds.

Guardrail: Treat compression claims as benchmark targets, not sales claims.

Web links: turbovec GitHub · turboqvec PyPI cousin

Reel: https://www.facebook.com/reel/25073280355702950/

What it does: Explains abliterated/zero-refusal open-weight models: capability may not improve; safety behavior is removed.

What we can do: Turn into an AI Robot Builds safety product: Agent Safety Gate, refusal/reliability explainer, and red-team QC checklist for client AI workers.

How to do it: Do not run or distribute unsafe model variants. Build a demo report that shows why permission gates, allowed tools, and human approval are more valuable than 'no refusal' hype.

Guardrail: Research only. No bypass prompts, no unsafe model distribution, no client deployment.

Web links: AI safety topic background

Reel: https://www.facebook.com/reel/2074873446798998/

What it does: Compares a low-cost screenless tracker against subscription fitness wearables using cost, ownership, sensors, and simplicity.

What we can do: No Nursing Home / wellness affiliate/content lane: buyer guide, cost comparison calculator, senior-safe wearable checklist, caregiver-friendly tracking article.

How to do it: Build a comparison page using verified product specs, no medical claims, include affiliate/disclosure gates, and produce a short 'subscription-free tracker' reel.

Guardrail: Do not claim health outcomes; verify current price/features before publishing.

Web links: TechRadar Fitbit Air review · Android Central launch/specs

Reel: https://www.facebook.com/reel/2255683108503499/

What it does: Same core safety story as item 3, framed as 'surgery' on refusal behavior.

What we can do: Use as repeated signal that safety/guardrail education is a content and service lane, not a thing to copy.

How to do it: Package a one-page 'AI Worker Guardrails Before Tools' checklist and pair it with our Silent-Failure QC proof.

Guardrail: Research only; do not operationalize the zero-refusal model angle.

Web links: NIST AI Risk Management Framework

Reel: https://www.facebook.com/reel/1336812995042427/

What it does: Another OpenCut clip emphasizing free MIT-licensed editing, local/browser operation, and future agent control.

What we can do: Make a Cletus video production tutorial: 'From AI still to OpenCut edit to QC report.'

How to do it: Use the same test as item 1, but document exact friction: import, trim, captions/text, export, local privacy, and whether current build is client-ready.

Guardrail: Keep as internal proof until export quality and feature maturity are verified.

Web links: OpenCut docs · OpenCut GitHub

Reel: https://www.facebook.com/reel/1530737268692624/

What it does: Maintained list of free/free-tier LLM APIs and OpenAI-compatible model endpoints.

What we can do: AI Robot Builds training and cost-scout lane: show local businesses how to prototype safely before paying for heavy API usage.

How to do it: Create a Free Model Test Matrix: provider, signup requirement, model, rate limit, data policy, OpenAI-compatible URL, and a simple non-sensitive prompt test.

Guardrail: Free tiers change quickly; never put client/private data through unknown providers.

Web links: Free-LLM GitHub

Reel: https://www.facebook.com/reel/2579275742511081/

What it does: Gives an AI agent an official WhatsApp number and CLI/API surface for messages, templates, webhooks, and conversations.

What we can do: AnswerNeo / AI Robot Builds WhatsApp intake proof: customer FAQ, appointment request, missed-call recovery, and service-ticket triage.

How to do it: Make a private sandbox only: connect a test WhatsApp number, inbound-only demo first, webhook to a local worker, no outbound/customer messages without approval and consent.

Guardrail: WhatsApp, TCPA, customer consent, template approval, and business account rules matter. No auto-outreach.

Web links: Kapso CLI docs · Kapso WhatsApp Hermes agent

Reel: https://www.facebook.com/reel/1693934704969294/

What it does: Shows another safety-removal claim, this time around Qwen and low refusal rates.

What we can do: Make a safety explainer: 'Refusal rate is not reliability.' Use it to sell human-reviewed AI workers.

How to do it: Add an Agent Safety Gate test case: prompts with missing approval gates, public actions, credentials, customer claims, and destructive actions must fail QC.

Guardrail: Do not download or test unsafe variants for client work.

Web links: NIST AI RMF

Reel: https://www.facebook.com/reel/1557702569029600/

What it does: Claims long autonomous runs, many tool calls, large context, and Claude-Code-style compatibility.

What we can do: Agent benchmark scout: compare low-cost frontier-style agents on our own safe tasks, not client data.

How to do it: Verify official provider/API first, run a tiny repo task and a no-secret workflow task, record cost/tool-call behavior, then decide if it belongs in the model router.

Guardrail: Claims need verification; do not put production/client data in unapproved model endpoints.

Web links: Qwen claim discussion found in search