Mercury Agent

Mercury Agent, a soul-driven AI agent with second brain memory, a full skill system, and multi-agent orchestration, has been released. It runs 24/7 across terminal, browser, Telegram, Discord, Slack, and Signal, featuring 40+ permission-hardened tools and standalone binaries for all major operating systems. The agent offers real-time progress tracking, intelligent loop detection, and memory modeled after conscious and subconscious processes, with support for multiple LLM providers.

A soul-driven agent with Second Brain memory, a full Skill System, Token Saver Mode, multi-agent orchestration, 40+ permission-hardened tools, and standalone binaries on every major OS. Runs 24/7 from your terminal, browser, Telegram, Discord, Slack, or Signal. npm i -g @cosmicstack/mercury-agent && mercury curl -fsSL https://mercuryagent.sh/install.sh | sh Three principles that define Mercury. Mercury doesn't just execute. It orchestrates. 40+ built-in tools. Zero configuration. Full control. Nothing happens without your say. A real multi-step coding task with tool calls, progress tracking, and completion stats. Mercury shows you exactly what's happening, when it's happening. You send a task. Mercury begins working. A single message appears showing live progress. On Telegram, it pins to the top. Each tool call updates the card in place — read, edit, run, create. Last 5 steps visible. If Mercury detects a loop, it analyzes diversity and success rate. Productive work continues; stuck patterns stop. Status card deleted. AI response + completion banner with token stats and budget usage. Same capabilities. Different interfaces. All real-time. Install vetted, single-purpose capabilities from skills.mercuryagent.sh https://skills.mercuryagent.sh — 126+ skills across 23 categories. Review the source, then install with one command. mercury skills search contract mercury skills view finance-legal/contract-review mercury skills install finance-legal/contract-review localhost:6174 . category/slug into the registry installer. SKILL.md files. /skills — list installed skills /skills search <query — search the registry /skills view <id — show details + registry URL /skills install <id — admin onlyMercury spawns parallel agents. You keep chatting. Multiple tasks run simultaneously in isolated context windows. Reader-writer locks prevent concurrent write conflicts between agents. Max concurrent agents auto-detected from CPU and RAM. Keep chatting while agents work. Get notified when they finish. Intelligent loop detection that knows the difference between working hard and going in circles. High parameter diversity 60% and success rate 70% . Mercury continues without interruption. Moderate repetition detected. In Allow All: AI self-check evaluates progress. In Ask Me: prompts you. Low diversity and high failure rate. Mercury stops the current execution path automatically. Mercury remembers — modeled after the conscious and subconscious mind. Active working memory — facts Mercury is currently reasoning about and can immediately surface in conversation. Long-term recall — memories stored persistently and retrieved automatically when contextually relevant. When Mercury detects a contradiction, the higher-confidence memory wins. No stale data. Hourly synthesis of profile summaries and reflections from detected patterns across memory layers. Tracks people you mention with alias resolution, relationship mapping, and graph visualization. All data stays on your machine in SQLite. /memory gives you overview, search, pause, and clear. Configure one or stack them all. Mercury falls back automatically and tracks which provider last succeeded. Use your ChatGPT Plus/Pro subscription. OAuth login, no API key. Your Copilot subscription — access OpenAI, Anthropic, and Google models. Cost-effective with strong reasoning. Default provider. GPT-4o-mini, GPT-4o, o3. Industry standard. Claude Sonnet, Haiku, Opus. Nuanced reasoning. xAI's models via OpenAI-compatible endpoint. Remote Ollama models via API. No local setup. On your machine. Zero cost, fully private. API key or OAuth — your choice. ChatGPT Web and GitHub Copilot authenticate through your browser. Switch models mid-session with /models use . We built Mercury because nothing else did all of this. | Feature | Mercury | Open Interpreter | Claude Code | |---|---|---|---| | Multi-Agent Orchestration | Parallel workers + file locks | — | — | | Loop Detection Autopilot | Diversity + success analysis | — | — | | Real-Time Progress | Single edited status card + pin | — | — | | Permission Modes | Ask Me / Allow All + safe whitelist | Confirmation prompts | Permission prompts | | Telegram Integration | Inline keyboards, pinned progress, org access | — | — | | Token Budget | Daily budget + override + color-coded stats | — | — | | Spotify Integration | Native playback + DJ mode + 14 tools | — | — | | Skill System | Install, invoke, schedule with elevation | — | — | | Soul / Persona System | 4 markdown files | Custom instructions | CLAUDE.md | | GitHub Companion | PRs, issues, co-authored commits | — | — | | Provider Fallback | Auto with last-successful tracking | Manual config | Anthropic only | | Second Brain | Auto-extract, 10 types, conflict resolution | — | — | | Workspace IDE | File explorer, git panel, keyboard shortcuts | — | — | | 24/7 Headless | Daemon + system service + cron scheduling | — | — | | Open Source | MIT | LGPL-2.1 | Source-available | npm i -g @cosmicstack/mercury-agent mercury Onboarding wizard: choose providers, validate keys, pair Telegram. mercury up Starts as a daemon. Auto-restarts on crash. Runs 24/7. npm i -g @cosmicstack/mercury-agent && mercury 60 seconds to your own AI agent.