Francisco "Chico" Cac, a developer operating out of Lisbon, has debuted Somnus, a desktop AI agent designed to eliminate the friction of managing multiple LLM providers. The application, available for Windows and Linux, aggregates access to over 50 modelsβ€”including Claude, GPT, Gemini, DeepSeek, Grok, Qwen, Kimi, and GLMβ€”under a single workspace and prepaid balance. This release marks Chico's first post on DEV.to, signaling a direct challenge to the fragmented landscape of API keys and dashboard hopping that plagues developers working with diverse model families.

Architecture Built on Open Source Foundations

Somnus is not built from scratch but rather leverages the Hermes Agent by Nous Research, which is released under the MIT license. Chico has customized the base with its own identity and integrated a proprietary model gateway. This gateway is OpenAI-compatible, meaning existing codebases can switch to Somnus by simply updating the base URL and API key. The architecture allows developers to change the model field in their API calls to route requests to different providers without altering the underlying toolchain or switching applications mid-task.

Simplified Billing and Cross-Platform Access

The platform addresses the pain point of stacked subscriptions by offering a prepaid credit system starting at $5, with the notable feature that credit does not expire. This model supports both desktop app usage and terminal access via the same account, providing flexibility for developers who work across different environments. Currently, the desktop application is available for Windows and Linux via AppImage and .deb packages, while the macOS version remains in active development. The unified approach aims to let users switch models mid-task without changing tools, a core requirement for modern AI-assisted workflows.

Key Takeaways

  • Somnus aggregates 50+ models including Claude, GPT, and Gemini behind a single OpenAI-compatible gateway.
  • The app is built on the MIT-licensed Hermes Agent by Nous Research, customized for multi-provider routing.
  • Billing uses a non-expiring prepaid credit system starting at $5, eliminating subscription stacking.
  • Windows and Linux support is live; macOS support is currently in progress.
  • Developers can switch models mid-task by changing the model parameter without altering base code.

The Bottom Line

Juggling API keys is the tax we pay for the LLM renaissance, and Somnus is a necessary hack to reclaim developer sanity. By standardizing access through an OpenAI-compatible interface, Chico is proving that the future of AI agents isn't about picking a winner, but about seamless interoperability.

Future Development Roadmap

Chico plans to document the build process in future posts, focusing on the technical challenges of running a multi-provider model gateway. Topics will include routing logic, usage tracking, and billing infrastructure, as well as the less glamorous aspects of launching on a new domain and navigating SEO. The developer is actively soliciting feedback on the specific annoyances developers face when managing multiple LLM providers, indicating a community-driven approach to refining the platform's utility.