Archivist

The USER layer splits three ways. (1) HackerNews (16 top comments, apparently from a Show HN-style thread on an LLM/agent knowledge base): discussion is conceptual and skeptical rather than about daily use. Users report building their own variants — a git-based knowledge base running agent prompts against a company's tickets/wiki/Jenkins ('I spent several hours effectively onboarding the AI (Claude Opus 4.6)'), a TiddlyWiki+LLM connector (from TiddlyWiki's original creator), a sprites.dev-hosted 'LLM writes a wiki' app, and a 'commonplace' KB with auto-written reviews. Recurring criticisms: agents are 'probabilistic, not deterministic... the longer an agent runs on a task, the more likely it will fail' and 'burn a ton of token cash'; one user gave up on local LLMs because Gemma 4 26B Q4 was 'pretty buggy' and ate all 12GB of VRAM; another flags missing snapshots/rollback and unmanaged 'weight of opinion' in shared knowledge; a cited arXiv study (2602.11988) found auto-generated AGENTS.md underperforms human-edited ones. Structural critique: Obsidian cannot host multi-agent promotion flows ('agent A drafted this, agent B promoted it') — agents need an MCP surface with tools like /lookup and notebook_write. Counterpoint: several senior devs say delegating code/context work to LLM agents already changes their workday. (2) YouTube comments (on Archivist.AI D&D videos): mostly positive — 'the chaos of D&D meets the organization of AI', and a 6-month user: 'it has improved dramatically... new features are being introduced all the time'. One user prefers competitor 'Realms of Shod'; another says plain Notepad with headers plus keyword search covers the same need. (3) ProductHunt: a maker post ('Conny') describes an 'Archivist' that renames and organizes AI-generated images (custom presets like 2025-01-04_client-nike_hero_dalle_v1.jpg, smart collections) — a different feature set, likely a different product sharing the same name.

No aggregate ratings were found for this product during the last harvest.