cosmicstack-labs/mercury-agent

Soul-driven AI agent with permission-hardened tools, token budgets, and multi-channel access. Runs 24/7 from CLI, Telegram or More.

What it solves

Mercury is a persistent, "soul-driven" AI agent designed to act as a personal assistant and senior engineer. It solves the problem of AI agents acting silently or unpredictably by implementing a permission-hardened system that asks for user approval before executing sensitive commands. It also addresses the problem of short-term memory loss in LLMs through a structured "Second Brain" for long-term, persistent memory.

How it works

Mercury operates as a background daemon that can be accessed via multiple channels (CLI, Telegram, Discord, Slack, and Signal). It uses a combination of SQLite-backed memory and a customizable personality defined by markdown files. The system includes a dedicated coding environment called "Mercury Code" for repository-aware programming and a "Bots fleet" for managing specialized sub-agents. It employs a token-aware budget system to manage API costs and a skill registry for extending its capabilities.

Who it’s for

It is designed for developers and power users who want a highly customizable, secure AI agent that can manage their codebase, handle recurring tasks via a scheduler, and integrate into their existing messaging workflows.

Highlights

  • Permission-Hardened Tools: Includes a shell blocklist and folder-level scoping to prevent dangerous commands from executing without approval.
  • Second Brain: A persistent SQLite memory system with 10 memory types and automatic fact extraction and consolidation.
  • Mercury Code: A full-screen IDE-like environment with an "AUTO mode" for planning and executing code changes with verified completion contracts.
  • Multi-Channel Access: Native support for CLI, Telegram, Discord, Slack, and Signal.
  • Mercury Bots Fleet: Ability to deploy persistent, persona-scoped specialist bots that run asynchronously.
  • Extensible Skills: A community-driven skill registry allowing users to install new capabilities with a single command.
  • Token Management: Daily budget enforcement and a "Token Saver Mode" to reduce context costs.

Related

  • Project
    MemTensor/memmy-agent

    Memmy is a personal memory infrastructure that provides a shared, structured memory layer across different AI agents, ensuring task continuity and context preservation when switching tools.

  • Project
    henrydaum/second-brain

    A locally-hosted AI agent that integrates personal files and applications through an extensible package system and a unified web interface.

  • Project
    peterfei/ai-agent-team

    A virtual AI development team with specialized roles (PM, Backend, Frontend, etc.) and a persistent memory system for managing complex software projects.

  • Project
    rahilp/second-brain-cloudflare

    Second Brain is a Cloudflare‑hosted personal + team memory system that stores notes, decisions, and project facts and makes them searchable by meaning. It integrates with Claude, ChatGPT, Cursor, Obsidian, Notion, calendars, etc., and provides a secure, self‑hosted backend (D1, Vectorize, Workers AI) plus a “Prompt Capsule” feature for deterministic LLM context.

  • Project
    AtomicBot-ai/atomic-agent

    A local-first AI agent that runs on your machine to browse the web, edit files, and run commands, optimized for small local models via a context-frugal loop and TurboQuant.