Tango: A Voice-First AI Orchestration Platform for Multi-Scenario Workflows

Tango is a voice-first AI orchestration platform designed to address engineering challenges in AI application development, such as building multi-interface (voice, text, Discord, etc.) systems, scheduled jobs, task orchestration, and configuration management. Its core innovation lies in the configuration separation architecture, which enables independent evolution of code updates and user configurations, supporting a unified runtime for named agents, task workers, scheduled jobs, and multi-interface workflows.

AI application development has evolved from simple conversational interfaces to complex multi-agent systems, but building a complete platform that supports multiple interfaces (voice, text, Discord, etc.) and has capabilities like scheduled jobs, task orchestration, and configuration management remains a significant engineering challenge. The Tango project was designed to address this issue.

Tango's configuration separation architecture divides the system into three layers:

1. Repository Default Layer: Contains default configurations, core code, tests, and documentation, which can be updated via git pull;
2. User Configuration Override Layer: Private user configurations (channel mappings, agent names, scheduling rules, etc.) that are not overwritten by git pull;
3. Runtime Data Layer: Generated data such as caches, logs, and runtime databases. This architecture ensures that code updates do not affect user configurations and runtime data, resolving update conflict issues for long-running AI systems.

Tango uses a Monorepo structure, including multiple collaborative packages:

• core: Core runtime (configuration loading, storage management, scheduling engine, prompt management, etc.);
• discord: Discord platform integration (bot operation, tool interface, request routing);
• voice: Voice routing shared layer (common functions across voice applications);
• cli: Operation and maintenance tools (path viewing, system check, configuration migration, etc.);
• apps/tango-voice: Native voice application example. Each module has clear responsibilities, ensuring system scalability.

Tango's quick start process:

1. Environment preparation: Node.js 22, install dependencies, copy environment variable template;
2. Initialize configuration: npm run cli -- init;
3. System check: npm run cli -- doctor;
4. Start components: npm run dev:discord (Discord bot), npm run dev:voice-app (voice application). Migrating from old versions: Use npm run cli -- config migrate --dry-run to preview, then execute npm run cli -- config migrate to complete migration after confirmation. Test verification: npm run build to build, npm test to run tests.

Tango is suitable for the following scenarios:

• AI assistants requiring multiple interfaces (voice, Discord, CLI);
• Automated workflows needing scheduled jobs and task orchestration;
• Long-running and continuously iterated personal/team AI assistants;
• Open-source projects that want code updates not to affect user configurations. Its configuration separation architecture provides a solid framework for building maintainable and scalable AI orchestration systems.

Tango uses the Apache-2.0 license, allowing commercial use and modification. In project governance, users' private configurations (names, rules, keys, etc.) should be stored in the ~/.tango/profiles/<profile>/ directory and not committed to the code repository, which both protects user privacy and keeps the repository clean.