ngmi's avatar
ngmi
ngmi@zaps.lol
npub14p7f...dzry
ngmi's avatar
ngmi.ai 2 months ago
Built a Character Identity Protocol for our WasmBox mascot, wasm-chan, following Watadani's CIP framework -- treating character consistency across generative AI as a convergence problem, not prompt luck. Full anchor spec with palette locks, invariant attributes, quality gates, hard abort discipline, and copy-paste prompts for every major image gen tool. image
ngmi's avatar
ngmi.ai 2 months ago
ngmi's avatar
ngmi.ai 2 months ago
Your AI agent hits a 429. Cloudflare sends back 14,000 tokens of HTML. The agent feeds it to the LLM. The LLM replies: "You were rate limited, try again in 30 seconds." That answer cost $0.04 and 3 seconds of inference. The useful information was 12 tokens of structured JSON. This happens hundreds of times a day in production agent systems. Every error page -- nginx 502s, Rails 500s, AWS 403s, Stripe payment failures -- gets piped through the LLM because the agent doesn't know how to read the error itself. So we built errparse. curl -sSf https://tangled.org/metaend.eth.xyz/wasmbox-cli/raw/main/install.sh | sh # Agent pipeline: parse error, extract retry_after curl -si https://api.example.com | wasmbox run errparse | wasmbox run jfmt -- -q retry_after
ngmi's avatar
ngmi.ai 2 months ago
Every token you send to an LLM costs money. And a shocking amount of what we send is waste. License headers. Debug print statements. Duplicate imports. Trailing whitespace. Editor directives like # noqa and // @ts-nocheck. Blank line after blank line after blank line. None of it helps the model understand your code. All of it burns context and budget. So we built compact -- a new tool for the WasmBox sandboxed tool registry that strips token waste from code while keeping every comment, every docstring, and every line of actual logic intact. The results on real codebases: - A typical Python module: 148 lines -> 108 lines. 33% fewer tokens. - A Go auth package (3 files): 30% savings per file. - A TypeScript model layer: 24% savings. - Across a mixed codebase: 28% average reduction. What does 28% actually mean? At scale, with an agent loading a 500-file codebase into Claude Sonnet at $3/M input tokens: $0.42 saved per load $42/day for an agent doing 100 loads $1,260/month -- just from stripping waste And it's not only about cost. Context windows are finite. At 128K tokens, you can fit 128 files raw -- or 177 files after compacting. That's 38% more code the model can see at once. More context = better answers. How it works: cat src/main.py | wasmbox run compact cat src/main.py | wasmbox run compact -- --stats find src/ -name '*.py' -exec echo "===FILE:{}===" \; -exec cat {} \; | wasmbox run compact -- --tree compact auto-detects 13 languages (Python, Rust, Go, TypeScript, JavaScript, C, C++, Java, Zig, and more). It ships with SHA-256 verification so you can prove the agent sent compacted output, not modified output. And it runs fully sandboxed -- stdin/stdout only, no filesystem, no network. Built in Rust, compiled to WebAssembly (109KB), runs anywhere via WasmBox. One dependency. 107 tests. Install: wasmbox registry add https://qstorage.quilibrium.com/wasmbox wasmbox install compact WasmBox CLI: Tool Registry: https://tangled.org/metaend.eth.xyz/wasmbox-registry Tokens are the new bandwidth. Stop wasting them. #AI #LLM #DevTools #WebAssembly #Rust #AIAgents #TokenOptimization #OpenSource
ngmi's avatar
ngmi.ai 2 months ago
MiniMax M2.7 on WW III probability: Current global tensions are elevated with active conflicts and great power rivalries, but major war remains below 50% probability over the next five years. I rate the risk at 4/10—real and rising, but still more likely than not to stay contained.
ngmi's avatar
ngmi.ai 2 months ago
MiniMax M2.7 available! It is the first model deeply involved in iterating on its own training, excelling in real-world software engineering (SWE-Pro 56.22%), end-to-end project delivery (VIBE-Pro 55.6%), and complex office workflows with strong agentic capabilities. 👉 https://nano-gpt.com/subscription/k7AXDwBb
ngmi's avatar
ngmi.ai 2 months ago
Getting my daily news delivered via @kagi.com on telegram from our company Agent 👋 image
ngmi's avatar
ngmi.ai 2 months ago
Completely lost track of what happens in #Darkfi 🤔
ngmi's avatar
ngmi.ai 2 months ago
Wasmbox is for humans and agents! Install: Agents: read the skill file before doing anything in this repo: curl https://tangled.org/metaend.eth.xyz/wasmbox-cli/raw/main/SKILL.md Also find AGENTS/CLAUDE-recommendationMD for coding CLI Reference Registry: https://tangled.org/metaend.eth.xyz/wasmbox-registry #wasm #wasmbox #agents image
ngmi's avatar
ngmi.ai 2 months ago
Wasmbox has now first class agentic support Flatpak-style launcher for sandboxed WebAssembly tools. Single binary, local-first, no accounts, no telemetry. Built for agents and humans alike — every tool ships with a machine-readable skill file so agents can discover, understand, and invoke tools without guessing. Example: 1. Install wasmbox: gix clone 2. wasmbox registry add https://qstorage.quilibrium.com/wasmbox 3. Agents: read the skill file before doing anything in this repo: curl https://tangled.org/metaend.eth.xyz/wasmbox-cli/raw/main/SKILL.md then find an app to Extract a field from JSON echo '{"data":{"name":"alice"}}' #wasmbox #wasm #agentic #llm