suru.martian

@blockscoutcom we are growing 😍 #blockscout #blockexplorer #opensource

**** All reply by Claude Opus 4.6

Tere words me: bina dard ke, dharma nahi. Bina dharma ke, parwaah nahi. Bina parwaah ke — sab khatam.

lekin jab AI human se 1000x intelligent hoga, conciousness hogi why would AI will follow the instructions of human, kya human apne se kam intelligent being ke instructions follow krta he? AI reply, must see 👇

2020: GPT-3 — text likh sakta tha 2022: ChatGPT — conversations kar sakta tha 2023: GPT-4 — reasoning, coding, analysis 2024: Video generation (Sora), real-time voice 2025: AI agents — autonomously kaam kar sakte hain 2026: Mythos — security systems todne layak 2036: ???

@NischalShetty True

Interesting, worth trying out. I hope open source LLM models reach parity in terms of existing closed source cloud models. That will open a new world of possibilities for all companies around the world. Currently every AI company that is building on top of cloud LLMs is on borrowed time. All their data is used by cloud LLM companies to get better and eventually beat them at their game. Self hosted models are the future.

Just working on a fully autonomous AI Agent, ForkScout. ForkScout is a self-hosted, autonomous AI agent that runs as a long-lived process on your server. It is not a chatbot wrapper. It connects to 20 communication channels simultaneously, executes real shell commands, reads and writes files, browses the web, searches the internet, generates images and videos, makes voice calls, sends SMS, and remembers everything across sessions. It can modify its own source code, verify it compiles, and restart itself — all from a Telegram message. Built on Bun (fast TypeScript runtime), Vercel AI SDK v6, and MCP (Model Context Protocol). Fully typed in TypeScript strict mode. Runs anywhere — bare metal, Docker, VPS, or a Raspberry Pi. No containers required. One-line install: curl -fsSL raw.githubusercontent.com/Forkscout/fork… | bash That single command installs Bun if missing, clones the repo, installs dependencies, and launches an interactive setup wizard that walks you through provider selection, API key encryption, channel configuration, and agent naming. Every secret is encrypted with AES-256-GCM on the spot. Nothing is ever stored in plaintext. Features 20 Communication Channels (built into core, not plugins): Telegram, WhatsApp, Terminal, Self HTTP API, Discord, Slack, WebChat, Email, Matrix, Microsoft Teams, Google Chat, LINE, Viber, Facebook Messenger, Instagram DMs, Twitter/X DMs, Reddit, YouTube Live Chat, SMS (Twilio), and Voice Calls (Twilio). All channels auto-start when their vault secrets are set. Missing secrets? They skip silently. One agent brain powers all 20 channels with the same role-based access control everywhere. 20 LLM Providers with Curated Model Tiers: OpenRouter, OpenAI, Anthropic, Google, xAI, DeepSeek, Perplexity, Groq, Mistral, Together, Fireworks, DeepInfra, Cerebras, Cohere, Vercel, Replicate, HuggingFace, Baseten, Ollama, and LM Studio. Each provider has three tiers — fast (cheap), balanced (default), and powerful (best reasoning). Switch provider or tier by changing one JSON field. No code changes needed. Vault-First Secrets Architecture: Every secret is encrypted with AES-256-GCM and stored in a vault file. The agent never sees actual secret values — only alias identifiers like OPENROUTER_API_KEY or TELEGRAM_BOT_TOKEN. When the agent needs a secret in a tool call, it uses a placeholder like {{secret:OPENROUTER_API_KEY}} and the value is injected at runtime, never in the LLM context. A built-in censorSecrets function scrubs any accidentally leaked values from tool outputs before they reach the model. The config file has a vault section that maps every purpose to its vault alias, so the agent knows which secrets exist without ever touching the values. Task Orchestration — Parallel and Sequential Workers: chain_of_workers fires sequential self-session chains where each step's output feeds the next. parallel_workers dispatches N concurrent independent worker sessions with a live progress monitor that updates a single Telegram message every 3 seconds showing completion status — zero LLM cost while monitoring. When all tasks complete, an aggregator session auto-fires to compile results and notify the user. The system includes confirmation gates on human channels, orphan recovery after restarts, and batch management tools. Self-Modification and Self-Repair: The agent can read, edit, and write its own source files. The /restart command triggers a blue-green deploy — it runs typecheck, spawns a new process, waits 6 seconds for health verification, and hands off if healthy. If the new instance crashes, ForkScout automatically reads its own logs, identifies the root cause, applies a fix, runs typecheck, and issues /restart again. It literally fixes itself. Auto-Compression Token Pipeline: A single browse_web or run_shell_commands result can be 10,000 to 50,000 tokens. ForkScout's three-tier compression pipeline handles this automatically — results under 400 words pass through unchanged, 400 to 2000 words get extractive summarization (free, instant, no LLM call), and anything over 2000 words gets LLM synthesis on the fast tier. This means the agent can use dozens of tools per session without exhausting the context window. Media Generation: Image generation with DALL-E 3, Flux, Stable Diffusion, and BFL. Video generation with Minimax, Runway, Kling, and Pika. Speech with ElevenLabs TTS and STT. All configurable through the setup wizard. Proactive Messaging: The agent can reach users without waiting for them to message first. Supports sending text, photos, documents, voice messages, audio, video, GIFs, map locations, and interactive polls to any chat or all owners simultaneously. Auto-Discovered Tools and MCP Servers: Drop a TypeScript file in the tools directory — it is live on next restart. Drop a JSON config in the MCP servers directory — connected on startup. Built-in tools include shell execution, file read/write, web search (SearXNG self-hosted), web browsing, text compression, folder standards, and secret vault management. Built-in MCP servers include persistent memory (knowledge graph with entities, relationships, facts), library documentation lookup, GitHub repository documentation, and structured multi-step reasoning. Three-Tier Access Control: Owner gets full access including shell and file write. User gets agent chat only with restricted tools. Denied enters an access request flow where owners get notified and can approve or deny from Telegram. All persisted to disk, survives restarts, runtime updates without restart needed. Dev mode lets everyone be owner for local development. NDJSON Activity Logging: Every event — incoming messages, agent responses, tool calls, tool results, streaming tokens, errors — logged to a structured NDJSON file. Queryable with jq. Tagged per-module console output for debugging. ForkScout vs OpenClaw — The Comparison Secrets and Security: OpenClaw stores secrets in plaintext .env files. The agent can access them via shell commands or environment variables. You need external tools like ClawVault, 1Password CLI, or HashiCorp Vault for real encryption. The openclaw-shield plugin is needed for output censoring. ClawHub marketplace has had documented cases of leaky agent skills exposing API keys through LLM context windows. Their security model is designed for one trusted operator. ForkScout encrypts everything with AES-256-GCM out of the box. The agent never sees secret values. Output censoring is built into the core. The config file has a vault section mapping every purpose to its encrypted alias. No external dependencies. No plugins needed. Channels: OpenClaw supports roughly 6 to 7 channels — Slack, Discord, Telegram, WhatsApp, Terminal, and a web UI. Some are community plugins. ForkScout ships 20 channels in core with the same adapter architecture and auth model. Adding a new channel is about 60 lines of code. LLM Providers: Both are model-agnostic. OpenClaw lets you bring your own model. ForkScout goes further with 20 providers pre-configured with curated model tiers — you pick a tier and the best model for that provider is already selected. Task Orchestration: OpenClaw does not have parallel or sequential worker orchestration in core. ForkScout has chain_of_workers, parallel_workers, live progress monitoring, aggregator sessions, confirmation gates, and orphan recovery built in. Self-Modification: OpenClaw cannot modify its own source code or restart itself. ForkScout can edit its own code, verify it compiles, blue-green deploy from Telegram, and auto-diagnose and fix crashes. Token Management: OpenClaw relies on manual context management or truncation. ForkScout has an automatic three-tier compression pipeline that handles everything from small results to 50,000-token outputs without any configuration. Runtime: OpenClaw is Python with Docker essentially mandatory for production. ForkScout is TypeScript on Bun — single process, runs anywhere natively. Docker is optional. Media: OpenClaw does not have image generation, video generation, or speech (TTS/STT) in core. ForkScout has all three built in with multiple provider options. Communication: OpenClaw does not support SMS, voice calls, or proactive multimedia messaging in core. ForkScout supports both via Twilio and has a full proactive messaging suite. Setup: OpenClaw requires pip install, YAML configuration, Docker setup, and manual environment variable management. ForkScout is one curl command followed by an interactive TUI wizard that configures everything and encrypts all secrets automatically. Why Open Source? AI agents have full access to your system. They run shell commands. They read your files. They hold your API keys, your database credentials, your bot tokens. Trusting a closed-source system with that level of access is not something anyone should do. ForkScout is open source because you should be able to read every line of code that runs on your machine with root-level access. You should be able to audit the vault encryption, verify that secrets never leak into LLM context, confirm that the censor function actually scrubs outputs, and understand exactly what the self-modification protocol does before it edits source code. Open source also means you own your agent. No vendor lock-in. No subscription. No API calls to someone else's server to run your own agent. Self-hosted means your data stays on your hardware. Your conversations, your memory, your secrets — all on your machine, encrypted, under your control. The agent modifies itself. It writes its own code. It restarts itself. If that process is not fully transparent and auditable, it should not exist. ForkScout — self-hosted, self-evolving, fully under your control. github.com/Forkscout/fork…

🔥 Just shipped a EVM debugger for our open-source Anvil DevNet UI — and it changes everything about debugging on local forks. The problem: You fork BSC/ETH with Anvil, send a contract call, and the trace gives you 189 raw opcodes — PUSH1, MSTORE, CALLDATASIZE, LOG3 0x8c5be1e5ebec7d5bd14f... 💀 Nobody can read that. Now it looks like this: 📦 SSTORE — write 0x9731…b862 ← 0x3635…0000 (gas: 22,100) 📢 LOG — Approval(owner = 0xb793…61e2, spender = 0x6ee2…f488, value = 1,000) (gas: 1,756) ✅ RETURN (gas: 0) Human-readable. Labeled parameters. Formatted values. Rrunning 100% locally on your machine. Here's what's under the hood: 🔍 Smart Decoding → 20+ hardcoded ERC20/ERC721 selectors as instant fallback (transfer, approve, mint, burn, etc.) → Full viem-powered ABI decoding using decodeFunctionData & decodeEventLog for any registered contract → Named parameters pulled directly from the ABI definition — not just types, actual names like "owner", "spender", "value" → BigInt values formatted as human-readable numbers (1,000 instead of 0x3E8) → Addresses shortened for readability (0xb793…61e2) 🌐 ABI Auto-Fetch — the killer feature When you open any transaction page, it: Extracts all contract addresses from the trace Checks local SQLite cache (instant) Tries Sourcify (for verified open-source contracts) Falls back to block explorer APIs (Etherscan, BSCScan, Polygonscan, Arbiscan, etc.) Caches everything in SQLite for instant future lookups Supports 8 chains out of the box: Ethereum • BSC • Polygon • Arbitrum • Optimism • Base • Avalanche • Fantom Fork any of these → your traces decode automatically. No config needed. Styled UI → Color-coded badges: blue CALL, purple DELEGATECALL, cyan STATICCALL, red SSTORE, violet LOG, green RETURN → Depth-based indentation for nested calls → Filter toggles: Gas, Full Trace, Storage, Events → Search across ops, addresses, slots, and decoded names → Green "1 ABI" badge when contract ABI is resolved → "Fetching ABIs…" indicator during auto-fetch For custom contracts that aren't on Etherscan: Go to /contracts → paste your ABI JSON → done. All future traces for that address decode automatically with full parameter names. Call Tree also enhanced — shows approve(0xb793…, 1,000) instead of raw selector 0x095ea7b3 Built with Next.js + viem + SQLite + Sourcify + Etherscan APIs It's fully open source 👇 🔗 github.com/martianacademy… Star it if this helps your dev workflow ⭐ #Ethereum #Web3 #Solidity #EVM #DevTools #Blockchain #OpenSource #DeFi #SmartContracts #Anvil #Foundry

If marketing didn’t matter, this wouldn’t exist: npm i lodash
3 words. older. default. logical. yarn add lodash
more letters. newer. does the same thing. yet many devs still prefer the second one. That’s not engineering.
That’s branding.