Open-source 'Orbital' supports context sharing between AI agents and project-level state management

The open-source project Orbital manages tasks at the project level rather than the session level, solving the issue of context loss when switching between AI agents. The tool stores project state, decisions, lessons learned, task queues, and artifacts in local files (PROJECT_STATE.md, DECISIONS.md, etc.), which are shared by all agents.

• Context Sharing: When hitting usage limits with Claude Code, users can switch to other agents like Codex, Gemini CLI, or Cursor while maintaining the same project context and continuing their work.
• Multi-Agent Support: Supports various LLM providers such as Anthropic and OpenAI, delegating Claude Code, Codex, Gemini CLI, and others as sub-agents via various methods including SDK, app-server, PTY, and ACP.
• Security and Control: API keys are securely stored in the OS keychain, offering per-project budget limits, safeguards against repetition/infinite loops, and autonomy presets such as approve/reject/bypass.
• Task Management: Tasks in the queue are designed to always conclude as either Completed or Blocked, preventing work loss due to agents stopping without reason.