Cerebrum

A device that blocks microphones from recording your conversations https://deveillance.com/

Qwen 3.5 running locally on a iPhone 17 Pro. On par with models 4 times its size and offers the flexibility to enable or disable reasoning. The 2B 6-bit version running on the iPhone utilizes MLX optimization for Apple Silicon. The developer is working on an update of the app to make it available for the rest of us. This app has some really interesting features, especially the voice mode which works also complete offline. https://locallyai.app/ #privacy #freedomtech #localai #localllm

Here's my #openclaw setup: Hardware & Network: - Dedicated Mac Mini ( WiFi disabled) - Ethernet-connected travel router (WiFi also disabled) -Network isolated via firewall rules on the travel router -Dedicated Apple ID and phone number for connecting messaging apps -Dedicated iPhone (same Apple ID) -Tailscale for secure access Multi-Agent Architecture (5 agents) via OpenRouter: - Main Agent: Orchestrator 3 Sub-Agents (each with isolated workspaces): -Product Owner (Gemini Flash 2.5) -Developer (Minimax 2.1) -QA/Security (Gemini Flash 2.5) 1 Secure Agent: @Maple AI with local proxy, defaulting to Kimi K2.5 for sensitive operations Built So Far: -NWC Wallet Skill integration for the agent ( @Alby now offers one too) (fun!!) -LN Markets trading skill -Frankenstein degen trader agent (work in progress) - trading dashboard -Local Virtual Agent (web) with RAG, powered by LM Studio (fun!) - Kanban board to keep track of what it’s working on (cool!) Key Learnings: -Configure agents via OpenClaw's CLI or manual configs, don't delegate this to an agent unless you're ready to burn through tokens with expensive models like Claude Opus -Costs can spiral fast: $50-200 in a few days if you get stuck in configuration loops - let it make a kanban board to keep track of all the tasks , work and projects, really helps get a grip on things. Multi Agent setup took me the most time, so be aware. The spawning of sub agents is really cool but you first need a solid setup before you start work on something big. I ran into a lot of situations where permissions were wrong and troubleshooting was burning tokens. Also there are still some bugs where sub agents inherit the default model instead of their cheaper more efficient model which again results in token burn. Interested in hearing about other setups!

Learned so much from this episode @Stephan Livera with cryptographer and developer Jonas Nick https://fountain.fm/episode/ArMJC3JUG14GCiXKqrP5

I think @Start9 may have reached ossification sooner than bitcoin.