Hivemind puts multi-agent teams in a self-hosted sandbox with 13 chat channels

Hivemind is a self-hosted, open-source platform for running persistent teams of AI agents, installed with a single shell command and accessed through a local web interface at localhost:8080. The README positions it for teams that want agents to keep state across restarts, collaborate in shared chat, and act through external systems rather than operate as isolated chat sessions. The core idea is a workspace layer around agents. Users create an account through a setup wizard, select from more than 150 agent templates across 18 categories, assign tools and skills, and connect a messaging channel. Agents can mention each other in team chat, hand off work, run on heartbeat schedules, and persist memory and workspace state. The project lists 13 integrated channels — Discord, Slack, Telegram, WhatsApp, Signal, Matrix, Mattermost, email, LINE, Feishu/Lark, Google Chat, Microsoft Teams, and iMessage — with a separate bot identity per agent where the channel supports it. A container-first workspace for agent teams The README emphasizes sandboxing as the organizing constraint. Tools, skills, and integrations are described as running inside containers with no access to the host machine. Agent code is said to run in a separate container without database access, credentials are vault-encrypted, workspaces are isolated, network egress can be controlled per agent, and an append-only audit log records tool calls and decisions. For builders evaluating multi-agent systems, that is a clearer starting point than a prompt-only framework: the trust boundary is defined before the tool surface. That tool surface is broad. Hivemind claims more than 40 built-in tools, including shell, files, a Playwright-based browser, Jira, email and Gmail, cloud storage across Drive, S3, Dropbox, OneDrive,