Mercury Agent

Overview

Mercury Agent is an open-source, soul-driven AI agent that runs as a local background service. It keeps a Second Brain memory in SQLite on your disk, asks for approval before writes or commands, and connects the same agent to the terminal, web dashboard, Telegram, Discord, Slack, and Signal.

Key Features

• Permission-gated actions: reads run without prompts, while writes, shell commands, pushes, and scope changes wait for approval; a shell blocklist blocks commands such as sudo and rm -rf / in any mode.
• Second Brain memory: local SQLite with FTS5 full-text search, 10 memory types, automatic extraction, conflict resolution, hourly consolidation, and /memory controls.
• Always-on runtime: mercury up installs a background service with boot start and crash recovery; schedules, heartbeat checks, proactive notifications, and a durable job queue are supported.
• Mercury Code: a repo-aware coding environment with AUTO mode, a live plan checklist, per-file diffs, a workspace IDE, background sub-agents, and verified completion after a build, test, or typecheck passes.
• Mercury Bots: persistent specialists with their own persona, model, memory, permissions, and daily budget; fleets can delegate tasks and deliver finished files.
• Multi-channel and model flexibility: terminal, web dashboard, Telegram, Discord, Slack, Signal, and 10+ model providers such as Anthropic, OpenAI, DeepSeek, Grok, Ollama, and Mercury Cloud, with automatic fallback.

Use Cases

• Use Mercury Code inside a repository to refactor code, add tests, and only mark work done after a typecheck or test passes.
• Use Mercury Chat as an everyday assistant that answers from local memory and works with about 50 built-in tools.
• Run persistent bots or fleets for scheduled research, writing, and delivery of finished files.
• Start a task in one channel and check its status from another, such as from the terminal to Telegram.

Getting Started

• Install via the one-line installer for macOS or Linux, or the PowerShell installer for Windows; the standalone binary path does not require Node.js.
• Alternatively install with npm i -g @cosmicstack/mercury-agent on Node.js 22 or newer.
• Run mercury to start the setup wizard, which asks for your name, the agent name, Mercury Cloud or your own API keys, and optional channels such as Telegram.
• Run mercury up to install and keep the background service running.
• Use mercury code . to open Mercury Code in the current repository, mercury doctor to change providers, keys, or channels, and mercury status to see what is running.

Deployment & Requirements

• Mercury Desktop runs on macOS (Apple Silicon and Intel), Windows, and Linux; the Mercury runtime also runs as a standalone binary, via npm on Node.js 22+, on Android via Termux, and on cloud VMs.
• The standalone installer downloads a binary for the OS and CPU and verifies its checksum; Node.js is not required on that path.
• npm installation requires Node.js 22 or newer.
• Signal support uses signal-cli; it runs natively on Linux and on macOS with Java 17+.
• Mercury Cloud is optional; local operation uses SQLite memory under ~/.mercury and the agent ships no telemetry.

Before You Adopt

• License: MIT. Review its terms before using, modifying, or distributing the project.
• Mercury Desktop 1.5.0 is not code-signed for Apple or Windows, so first launch may require an extra OS prompt.
• The agent process is not OS-level sandboxed; safety relies on approvals, a blocklist, and folder scopes, with Ask Me as the default.
• Signal support depends on signal-cli; it runs natively on Linux and on macOS with Java 17+.
• Community skills are not audited and should be reviewed before installation.