Zero-cost uncensored deep research stack: Ollama + SearXNG + dual-model fact-checking in Hermes
damnGruz · x · 2026-11-10
An X user shared a fully local "uncensored deep research" workflow claimed to beat Perplexity/Claude at $0/month:
- Pick a model: Qwen3.5 9B abliterated for 8GB laptops; Huihui Qwen3.8 27B for 16GB GPUs (3.9M downloads); DeepSeek V4.1 Flash uncensored via cheap API if no GPU
- The stack: Ollama (context set to 64k — the default 4k breaks research midway) + self-hosted SearXNG search + Local Deep Research plugged into Hermes as MCP
- Anti-hallucination design: Hermes gives each subagent its own model — the uncensored model researches while a normal model acts as checker, re-reading every cited page and deleting claims the page doesn't support, requiring two independent sources; the diagram stresses the uncensored model "never gets the final say"
It's essentially a reproducible multi-agent research pipeline tutorial — the dual researcher/checker architecture with citation verification is useful for anyone building local deep research, regardless of the uncensored angle.
More from coding & agent
- Leaked: xAI Building Grok Academy at learn.grok.com with 15 Courses and Certificates — nima_owji · 2026-11-10
- Claude Code share of T3 Code turns jumps from 52% to 78% after Opus 5.5 launch — FinanceYF5 · 2026-11-10
- Full-color display plus hand tracking now fits in 50-gram glasses — what if Codex ran on them? — Scobleizer · 2026-11-10
- Agentlas-OS: open-source agent OS that turns goals into verifiable task trees — Aiden_Tech_Ai · 2026-11-10
- Measuring an LLM agent that writes tests before the code exists: 33/33 planted bugs caught — Infinite-Fly-2756 · 2026-11-10
- MiniMax M3.1 Flash goes free in OpenCode, praised for bug bounties and repros — DanielLockyer · 2026-11-10