Self-Improving AI
Agent Framework
Autonomous skill creation, built-in learning loops, and multi-chain execution on Base and Solana — no human in the loop.
Learning Loop
Creates skills from experience automatically. Memory persists across sessions with FTS5 search and LLM summarization.
Multi-Chain
Execute on Base and Solana. Native USDC settlement, cross-chain messaging, automatic wallet management.
MCP Integration
Connect to any Model Context Protocol server. Extend with custom tools, compose agent workflows.
Multi-Platform
Telegram, Discord, Slack, WhatsApp, Signal, Email, CLI. Single gateway manages all channels.
Model Agnostic
OpenRouter, OpenAI, Anthropic, NVIDIA NIM, custom endpoints. Switch providers without code changes.
Scheduled Automation
Cron scheduler for recurring tasks. Delegation via isolated subagents with full context isolation.
One line to install
$ npm install @minary/sdk import { MinaryAgent } from "@minary/sdk"; const agent = new MinaryAgent({ model: "claude-sonnet-4-20250514", chains: ["base", "solana"], memory: true, learning: true, }); await agent.run({ task: "Monitor Base mempool and execute", tools: ["wallet", "dex", "analytics"], }); // ✓ Agent boots, learns, and improves autonomously // ✓ Skills persist across sessions
Ingest & Classify
Task router classifies incoming requests with confidence scoring. Routes to the optimal execution pipeline automatically.
Execute & Learn
Agent executes in isolated sandboxes. On success, learning engine extracts the execution pattern into a reusable skill.
Persist & Improve
Skills are scored for confidence. On similar future tasks, skills are loaded and refined. Every cycle makes the agent better.
Start building with Minary
Clone the repo and deploy your first autonomous agent in minutes.
View on GitHub →