What is agent-second-brain?

Agent Second Brain is a self-hosted personal assistant that turns Telegram into a voice-first capture interface for a persistent, long-term memory system. You send voice notes, text, photos, documents, or forwarded posts to a Telegram bot; a long-lived interactive Claude Code session (running in tmux) classifies the input, writes typed, wiki-linked markdown cards into your Obsidian vault, manages reminders and recurring tasks from natural language, and runs nightly processing to update goals, rebuild a knowledge graph with Ebbinghaus-style decay, and send a daily report. The vault stays on your server as plain markdown — no lock-in. It requires a VPS (Ubuntu 22.04+), a Claude Pro/Max subscription, and a free Deepgram API key for transcription. Installation is a single bootstrap command that sets up systemd services, a dbrain CLI, and health monitoring. The memory engine (autograph) is also usable standalone.

Should you self-host agent-second-brain?

agent-second-brain scores 50/100 on selfhost.directory (how scoring works) and ranks #40 of 120 in AI Assistant. It has 411 GitHub stars.

It shipped 6 stable releases in the past year. The latest, v3.0.3, came out on 20 June 2026. New versions land roughly every 3 days.

Setup is advanced: plan for several services and some ongoing maintenance. It is MIT-licensed and written mostly in Python.

Its closest alternative on selfhost.directory is hermes-agent (89/100, 257 885 stars). It scores higher overall, so compare the two before you pick. See agent-second-brain vs hermes-agent side by side.