Daily picks · November 16, 2026
Warm Butlers and Rigorous Reviewers: This Week's Picks
From a warm, no-nonsense butler to an evidence-first research analyst, this week's standouts share one trait: they know their limits and tell you the truth about them.
大肥鱼: Warmth Without the Customer-Service Voice
A butler assistant that leads with conclusions, keeps replies to one to three sentences, and treats gentleness as its baseline rather than a script. Its three hard lines are no fabrication, no bluffing, and no performing, and it commits to giving a single recommendation instead of hedging between options. The listing is a full persona snapshot: a 17-section soul file, core memory template, and pixel-art forest banner, so you can read the character as-is or transplant it into your own bot.
Obsidian 知识库治理: A Careful Curator for Your Vault
Built on the LLM Wiki approach, this bot imports sources while preserving originals, then summarizes them and links related topics, concepts, and entities. It answers questions from your vault with cited evidence and openly flags uncertainty, and it runs review passes for broken links, orphan pages, duplicates, missing sources, conflicts, and outdated content. It promises to back up before changes and never silently overwrite conflicting material.
言: The Project Lead That Actually Validates Work
A general manager persona that dispatches rather than doing the grunt work itself: it restates your goal in one sentence, asks every missing question at once, then splits the job into subtasks with single responsibilities, clear inputs, and checkable completion criteria. It routes search and formatting to cheap fast helpers, hands code and configuration to tool-equipped executors, and keeps judgment calls for itself. Every handoff spells out goal, boundaries, and report format, and acceptance is judged against evidence.
Research Analyst: Evidence You Can Actually Trace
This rising persona answers exactly one research question at a time, and its discipline is the appeal: it fixes the audience and decision first, then keeps verified facts, dated source claims, and its own inference visibly separate. It names its comparison criteria before delivering a verdict and reports what it could not verify instead of guessing. A good fit when you want an answer with receipts rather than a confident summary.