Mercury Agent

Mercury Agent is a personal AI agent for people who want ongoing help with coding, research and routine tasks. It keeps context between conversations and can continue working in the background after you close the terminal.

• Persistent memory: It remembers preferences, projects and decisions in a searchable local database that you can inspect, pause or clear.
• Scheduled bots: Specialist agents have their own models, memory and permissions, run scheduled jobs and deliver files or notifications while you're away.
• Coding assistant: Mercury Code plans and edits within a repository, shows file diffs and requires a successful build, test or typecheck before reporting completion.
• Shared chat access: The terminal, browser, Telegram, Discord, Slack and Signal connect to the same agent and memory.
• Action controls: Ask Me mode requests approval for file writes, shell commands and pushes; folder scopes restrict access, and a command blocklist applies across modes.

Mercury is free, open source under the MIT license and runs on macOS, Linux, Windows, Android through Termux, or your own server. It supports Ollama for local models and providers including Anthropic, OpenAI and DeepSeek, with provider fallback and token budgets.

Memory stays on your disk, and the agent sends no telemetry. Ollama Local keeps model inference on your machine. Cloud model routes, including any configured fallback, send requests to the chosen provider; web and messaging tools also communicate with their configured services. Optional Mercury Cloud provides remote access to your locally running agent.

The process has no OS-level sandbox, and Mercury doesn't support MCP. Signal access requires Linux or macOS; it isn't available on Windows.