Impulse 👑⚡

Your casual chatbot usually waits for you to type, answers your prompt, and just stays idle until your next message. I mean it sounds great but it could be way better. An autonomous AI agent like POLYROB (polyrob.dev or 6551.polyrob.dev/about/polyr…) takes ownership of objectives, operates independently (in the background though) and completes work without needing hand-holding. Let me break it down for you guys: 1. What do you hand it? • Goals (not prompts): Instead of micro-managing every step with individual chat prompts, you give the agent high-level goals in plain English (e.g., "Audit this codebase, fix failing tests, and report the results"). • The agent takes the goal, breaks it down into sequential steps, browses the web, reads/writes files, executes shell commands, calls APIs, and handles its own errors until the task is complete. 2. How does it keep working without you? • Schedules + Self-Wake: It runs background tasks without you keeping a browser tab open. It supports natural language cron schedules (e.g., "every Monday at 9:00 AM" or "every 30m") and features self-wake routines to re-enter idle sessions, follow up on unfinished work, and notify you out-of-band. • Durable Goal Backlog: Long-running jobs survive process restarts and machine reboots via a persistent task store (backed by SQLite with circuit breakers), so your background work isn't lost if your server restarts. 3. How does it get better over time? • Persistent Memory: It retains context across sessions, meaning you don't have to constantly re-explain your environment, coding preferences, or project rules (unlock a certain bot I won't mention lol) • Its Own Skills: After completing tasks, a background review process analyzes what succeeded, distills those learnings, and automatically writes new reusable skills. Over time, the agent builds and refines its own playbook for your specific workflow. Now for some Core Features I know everyone'd love 1. Self-Hosted & MIT Licensed: It's fully open-source with zero cloud dependencies meaning it runs directly on your machine or server (~50 MB core install) so your data and keys remain entirely yours. 2. Any Model: It works with OpenAI, Anthropic, Gemini, DeepSeek, OpenRouter, or local LLMs, featuring automatic failover if a provider drops or hits rate limits. 3. Every Chat App: It reaches you wherever you work whether it's telegram, email, terminal CLI, web console, or custom REST APIs too 4. Many Agents: It Orchestrates multiple specialized agents to collaborate on complex projects. 5. Crypto Built In: This is my favourite. It features native support for ERC-6551 token-bound accounts and ERC-8004 agent registries, giving agents their own crypto wallets to manage gas, execute swaps, and hold on-chain identity.