A ChatGPT-looking dashboard that runs your team's agent fleet

What it does Nakama gives your team a web dashboard that looks and feels like ChatGPT, but each conversation runs against a distinct agent with its own prompt files (SOUL, STYLE, INSTRUCTIONS, MEMORY), tools, skills, and MCP support. Agents can also live outside the dashboard entirely, answering on Telegram, WhatsApp, Discord, and Slack. It’s multi-tenant from the start: one server, multiple orgs, invites, roles, and agents isolated from each other.

The interesting bit Most agent frameworks assume one operator on one machine. Nakama flips that — it’s the same class of capability as OpenClaw or Hermes Agent, per the README, but designed as shared company infrastructure. The footprint is modest too: the Docker container idles around 200 MB, so a 1 vCPU / 1 GB box runs the server.

Key highlights

• Per-agent identity: separate prompts, tools, skills, memory, and permissions
• Channels for Telegram, WhatsApp, Discord, and Slack; Composio opens the door to 1,000+ external apps
• Multi-org with invites and roles — agents isolated from each other
• Runs as a single Docker container, or via managed hosting at Nakama Cloud
• Interactive API docs served at the server’s /docs endpoint; MIT licensed

Caveats

• Windows desktop builds are effectively unavailable — the README admits the unsigned MSIX files in older releases aren’t installable downloads.
• The Docker reset script removes every container publishing port 4310, including unrelated ones. Read before running.
• Local development requires Bun, which narrows contributor onboarding slightly.

Verdict Worth a look if your team wants shared, self-hosted agents with real channel integrations instead of everyone running their own single-user setup. Solo hobbyists after a one-machine agent may find it heavier than they need.