SJobs, Nietzsche and a Zen Master walk into a bar (github.com)

A new open-source “geist swarm” orchestration spins up personality-driven AI agents (called geists) — e.g., @dogen, @nietzsche, @sjobs — as isolated Docker containers, each running Claude Code (Anthropic) and a personality file that encodes worldview and communication style. The Python orchestrator (geist_swarm.py) provides a simple CLI (uv run geist create/list/ask/history/remove/reset) to auto-create agents from examples, run a structured two-round debate (Round 1: independent takes; Round 2: rebuttals where each geist sees others’ Round‑1 responses), and persist conversations to conversations/*.md and .geist_conversations.json. Containers share a docker volume (geist-shared) for inter-geist comms, include git for repo interaction, and can execute code inside their sandboxed environments by default (toggle with GEIST_ALLOW_EXECUTION=false). Setup requires Docker, Python 3.8+, and an ANTHROPIC_API_KEY environment variable. For the AI/ML community this is a compact, reproducible playground for multi-agent research, persona modeling, emergent debate dynamics, and tool-enabled agent workflows. Key technical implications: per-geist isolation for stateful persistence and safer code execution, prompt-context construction across rounds, and straightforward scaling/removal via container lifecycle commands. Practical cautions include managing API keys, Docker permissions, and evaluating safety/ethics of persona emulation and executable agents. The project is MIT‑licensed and targets rapid experimentation with multi-personality interactions using Claude Code.