A household AI server that boots its brain from SQLite

What it does

Octop is a self-hosted AI assistant platform built for households and small teams. It runs a single Python process that serves a web dashboard, CLI, IM bridges for apps like Feishu and Discord, cron automation, and even remote desktop sessions. Every user’s data lives under a local directory and is rebuilt from an SQLite database on startup, so the entire system is designed to stay on your hardware.

The interesting bit

Instead of the usual microservices sprawl, Octop routes every surface—Web UI, IM channels, and cron—through one in-process HarnessProcessor. The architecture is almost stubbornly monolithic: no external message broker, no queue, just a restart-safe process that reconstructs its state from disk. That simplicity is the point; it makes a multi-agent, multi-user system feel like a household appliance rather than a data-center deployment.

Key highlights

• Multi-user JWT isolation with admin roles, tool approval workflows, shell guardrails, and PII redaction.
• Per-user agents with distinct workspaces, pluggable storage backends (local disk, S3, COS, PostgreSQL), and portable memory via harness-memory.
• 16 MBTI personality templates plus custom prompts, plus an expert library scanned at boot.
• Bidirectional ACP support so Octop can both host agents for IDEs like Zed and delegate coding tasks to external runners such as Claude Code or OpenCode.
• Built-in browser automation via headless Chromium and remote desktop control for Linux, Windows, and macOS.

Verdict

Households and small teams who want a private, multi-user AI hub without managing Kubernetes or cloud bills should look here. If you need elastic scale or a managed SaaS experience, this monolithic, SQLite-backed appliance is not your architecture.