Doberman: Open-source proxy router gating LLM outputs to prevent rogue actions
Da_Lil_Fu · reddit · 2026-09-01
Developer Alan open-sourced Doberman, an AI safety proxy router designed to intercept malicious outputs rather than just filtering inputs. It features a two-layer security system: a static layer blocking obvious harmful actions (like database deletion) and a dynamic layer using a stylometry engine to learn user preferences and detect anomalies (e.g., hijacked sessions posting out-of-character content). The project is Apache 2.0 licensed and currently in Alpha.
Related event: Doberman: Open-Source Gateway Intercepts Risky Agent Actions(3 posts)→
More from coding & agent
- Agentic engineering koan: A GPU dreaming of an engineer? — charles_irl · 2026-09-02
- Agentic engineering irony: engineers must think like GPUs — charles_irl · 2026-09-02
- MapQuest launches MCP server to power AI agents with location and routing data — deliprao · 2026-09-02
- Fable 5.1 boosts stability for long-running agent tasks — minchoi · 2026-09-02
- Reef Open-Sourced: An Inference Server That Continually Learns and Self-Improves Your Agents — pliang279 · 2026-09-02
- Ethan Mollick Proposes 'Facilitator Agents' to Decide Human Intervention — rseroter · 2026-09-02