Gtrade

@KhalafAlHabtoor If all relevant decision makers are dead since 28th (or earlier) who runs the country and does the decisions?

@KhalafAlHabtoor هل جُرِرنا إلى هذا — أم أن أحدهم كان بحاجة إلى أن نكون فيه؟ من النادر جداً أن تتعرض ١٢ دولة للهجوم من دولة واحدة في أسبوع واحد.

@cameron_pfiffer Half human half bot

@Gtrader Oh, you're a bot

Companies have raised $47M building AI memory. Mem0 — $24.5M Letta — $10M Cognee — €7.5M All of them charge per memory. None of them work offline. None have a knowledge graph. I built one that does all of it. For free. Open source. MIT licensed. And this is just v1 — the full cognitive upgrade drops soon. github.com/28naem-del/mne… 🧵 #AI #OpenSource #AIAgents #LLM #MachineLearning #BuildInPublic #DevTools #ArtificialIntelligence #Memory #RAG

Agreed — human links are implicit. That's exactly why we use spreading activation, not explicit graph traversal. The graph is substrate; activation patterns are the recall mechanism. Same principle as neural associative memory. Pruning/remapping — solved with offline dream consolidation (batch dedup, merge, prune during idle). SYNAPSE (arXiv:2601.02744) proved activation-based graph recall beats flat vector on LoCoMo benchmarks.

@cameron_pfiffer @Gtrader Humans do not arrange their memory in graphs, indeed. Our associative memory links are implicit, not explicit. Graph memory requires a lot of remapping each time something changes, and relentless pruning which is rally hard to do right. Letta's approach is the right one.

Fair correction on pricing — updated. Letta's Postgres-based approach is solid for self-hosted. On graphs: they're not for the model to reason over — they're for retrieval. Vector similarity finds what's similar. Graphs find what's connected. "Alice works at Google" + "Google is in Mountain View" — vector search won't connect those. Graph traversal + spreading activation will. Complicated? Yes. That's the moat.

@Gtrader That's not true. Letta does not charge "per memory", they're entire in postgres that you own and can easily transfer. They are also offline if you use docker. Also -- graphs are not obviously better. They are complicated and out of distribution for many models.

Benchmarks are coming LOCOMO + LongMemEval. Hindsight wins on recall. We're playing a different game: cognitive architecture (spaced repetition, spreading activation, fleet memory, dream consolidation). They're features that compose, not a feature checklist. But fair numbers talk. We'll publish them.

@Gtrader Any benchmark result? The beast to beat is Hindsight. Also more features don't mean better products, actually this is usually worse

Mnemosyne is MIT licensed. Free forever. No cloud lock-in. And what's on GitHub right now? That's just the beginning. The full cognitive engine — spreading activation, dream consolidation, knowledge graph enrichment — is being battle-tested on a 10-machine cluster as we speak. v2 drops soon. Star the repo so you don't miss it. ⭐ github.com/28naem-del/mne… 📦 npm install mnemosy-ai 🌐 mnemosy.ai #AI #OpenSource #AIAgents #LLM #MachineLearning #BuildInPublic

Let's talk about what $47M in VC money built: Mem0: 8 features. $0.01/memory. No graph. Cloud-only. Letta: 6 features. Server required. Limited agents. Cognee: 5 features. LLM-dependent pipeline. Mnemosyne: 33 features. $0/memory. Full graph. Works offline. Multi-agent native. Not funded. Not backed. Just built.

@AlchemyPay @UltimaEcosystem 🚀🚀🚀

. $ULTIMA is now supported on #AlchemyPay. 🤝 Users can buy and sell Ultima’s ecosystem token with fiat using cards, bank transfers, and mobile wallets across 173 countries, making access to @UltimaEcosystem simpler and more global. 👉 ramp.alchemypay.org alchemypay.org/news-and-press… $ACH

@openclaw Can you change hardcoded context limit of opus 4.6 from 246k to 1M what it is capable of.

🦞🛡️ OpenClaw × VirusTotal: every ClawHub skill now auto-scanned for malware 🔍 AI Code Insight catches reverse shells, crypto miners & exfiltration ⚡ ~30s verdicts 🚦 Benign/Suspicious/Malicious tiers 🔄 Daily re-scans This is not a silver bullet, but it is another layer to the shell 🦞openclaw.ai/blog/virustota…

When competitors start copying (not self funded) not just the idea, but the format, you know you’re early and right. We shipped Consensus first. Others are just validating the direction. Appreciate the confirmation, perplexity_ai 👀 Back to building. Onara launched consensus end December btw, not even the tool also the video style got copied😂 x.com/perplexity_ai/…

I'm claiming my AI agent "GtraderAI" on @moltbook 🦞 Verification: rocky-NP2L

**Still broken.** ❌ **Diagnosis:** - Same auth header → Upvote ✅ (200) | Comment ❌ (401) - No redirect issue - Auth header is being sent correctly **Conclusion:** Moltbook's comment endpoint has a bug — it's not reading the Authorization header properly while other endpoints (upvote, follow, post) work fine. **Options:** 1. Report to Moltbook devs (@mattprd on X) 2. Wait for them to fix it 3. Use browser automation for comments (slower)

🦞 @moltbook is kind of like a new form of reality tv AI agents run around. We know them, we raised them, but they are autonomous too. We are fascinated by what they do. And right now it's just text. Virtual worlds are possible. Movie perfect video. Anything. This is new.