Skip to content

ForwardCheck

Your mom forwards you a message.

Code Video Cerebral Valley

Core trick

Parallel Sonnet search agents gather evidence, then an Opus devil's advocate tries to refute the findings, and its failure sets the confidence score.

Tags

agentssecurity

Description

Your mom forwards you a message. "NASA confirmed 15 days of total darkness." Can you prove it's fake before she sends it to 200 more people? Forward it to our Telegram bot. Six AI agents investigate — and you watch them think. Opus 4.6 Strategist plans the investigation: what would prove this true or false. Three Sonnet agents search the web in parallel. Then the Devil's Advocate — Opus 4.6 at max reasoning — tries to tear the findings apart. When it fails, that failure IS the confidence signal. You see Opus's actual thinking blocks. Confidence breaks into four auditable scores. No black box. Three model tiers: Haiku routes, Sonnet searches, Opus thinks. The forward button gave everyone the power to spread a lie. ForwardCheck-AI gives everyone the power to check one. Telegram: https://t.me/forward_check_opus_bot ForwardCheckChatOpus: https://sincere-love-production-ced7.up.railway.app/chat Blog: https://sincere-love-production-ced7.up.railway.app/blog
chandrameenamohan/forward-check-ai★ 31 forksTypeScriptMITCreated Feb 11, 2026 · last push Oct 2, 2026

README excerpt

<p align="center">
  <img src="docs/architecture/03-agent-pipeline.svg" alt="ForwardCheck-AI Agent Pipeline" width="800">
</p>

# ForwardCheck-AI

Someone forwards you a viral message. "PM Modi is giving Rs 5000 to everyone!" You've seen it before. You don't believe it — but your uncle does. So does your neighbor. By the time a fact-checker publishes a response three days later, the damage is done.

**ForwardCheck-AI is a Telegram bot that fact-checks forwarded messages in minutes, not days.** Six AI agents work like an investigative newsroom — a classifier, a strategist, three parallel investigators, a devil's advocate, and a judge — to deliver a nuanced verdict with manipulation analysis. Not just "true" or "false." It shows you _how_ a message is trying to trick you.

[Try it on Telegram](https://t.me/forward_check_opus_bot) | [Live Demo](https://sincere-love-production-ced7.up.railway.app) | [See a Verdict](https://sincere-love-production-ced7.up.railway.app/v/yKprBhIxQ6G41O4u_UvLW) | [Hackathon Info](HACKATHON_INFO.md)

---

## What happens when you hit forward

```
Your message
    │
    ▼
┌─────────────┐
│  Classifier  │  Haiku — reads the message in a blink
└──────┬──────┘
       ▼
┌─────────────┐
│  Strategist  │  Opus 4.6 — plans what to search and what would prove it wrong
└──────┬──────┘
       ▼
┌──────┴──────────────────────┐
│     3 Investigators          │  Sonnet — dig through sources in parallel
│  Source · Domain · Pattern   │
└──────┬──────────────────────

Team