The agent landscape is cluttered with heavyweight orchestrators that burn tokens just to decide which tool to call. Enter Routed, a new open-source project from bshea-1 that claims to handle skill routing locally with a hybrid approach that clocks in under 20 milliseconds and costs zero tokens.
The Zero-Token Promise
Most agent frameworks rely on LLM calls to parse intent and select skills, introducing latency and cost at every step. Routed aims to bypass this by performing routing locally, eliminating the need for API calls during the skill selection phase. This approach is critical for agents requiring real-time responsiveness or operating in environments with strict budget constraints.
Hybrid Routing Architecture
The project describes itself as a 'hybrid router,' suggesting a combination of rule-based and lightweight model-based techniques to achieve its speed. While the source material is limited, the emphasis on sub-20ms performance indicates a highly optimized local execution path. The repository is available on GitHub, inviting developers to inspect the implementation details and potentially contribute to the routing logic.
Early Adoption and Community Signal
As of September 6, 2026, the project has just hit Hacker News with a low score of 2 and a single comment. This early signal suggests it is a niche but potentially impactful tool for those frustrated with the overhead of current agent orchestration layers. The community response will likely determine whether Routed gains traction as a standard component in local-first agent stacks.
Key Takeaways
- Routed offers a local, zero-token routing solution for AI agent skills.
- The project claims latency of less than 20 milliseconds.
- It is currently in early stages, with minimal community engagement on Hacker News.
The Bottom Line
If Routed delivers on its zero-token promise, it could fundamentally change how we design low-latency agents by removing the LLM bottleneck from skill selection.