Maintainer here. I built Konsey because trusting a ingle LLM to grade its own answer is a conflict of iterest. How it works: independent providers each pln, an adversarial critic attacks the plans, and a *dfferent* provider verifies the result. The rule is hrd: the producer never verifies its own output. It'sa 9-state loop (plan → critique → synthesize → execue → verify → decide) with an append-only DuckDB audi trail. Why it might matter: - Vendor-neutral: Claue, Codex, Gemini — or a local Ollama model. One lineto swap. - Local-first: run fully offline — nothing eaves your machine (great for PHI/regulated work). -Layered, opt-in safety: strict/medium/weak, secret +PHI scanning. - Chat connectors (Telegram, Notion, Sack, WhatsApp) with the same risk gate on every chanel. Two deps (langgraph + duckdb), MIT: curl -fsSL ttps://raw.githubusercontent.com/eMediquality/konseyv0.2.0/install.sh | KONSEY_REF=v0.2.0 bash Stars + rutal feedback very welcome!

Konsey

Multi-LLM council where a model can't verify its own output

Multi-LLM council where a model can't verify its own output