OpenVorld drops pixel-art harness for AI agents
OpenVorld is an open-source, local-first desktop harness that visualizes multi-agent workflows as an interactive pixel-art village. Developers can assign distinct agent brains like Claude Code, Codex, or local Ollama models to separate crew members who collaborate under strict permission gates.
Turning terminal-based agent swarms into a simulated town sounds playful, but visual spatialization and explicit permission writs address real observability and safety bottlenecks in multi-agent coding. As developers transition from single-prompt assistants to autonomous agent teams, visual metaphors make delegation and supervision substantially more intuitive.
- –Visual spatialization: Agents live in dedicated houses with status bubbles above their heads, making concurrent tool execution and task handoffs inspectable at a glance.
- –Flexible backend routing: Each crew member can run on a different provider, from Anthropic's Claude Code and OpenAI Codex to self-hosted models via Ollama and LM Studio.
- –Strict writ permissions: Agents must request explicit user approval before touching local files, running bash commands, or querying external websites, preventing runaway execution.
- –Fully local and offline: Operates directly on local project folders across macOS, Windows, and Linux without requiring cloud accounts, telemetry, or remote backends.
DISCOVERED
23m ago
2026-11-21
PUBLISHED
5h ago
2026-11-21
RELEVANCE
AUTHOR
Amir Ghasemi