evalmonkey vs qmd

evalmonkey is an open-source framework for testing how LLM agents degrade under adversarial conditions. You run your agent against 10 standard datasets (GSM8K, ARC, HellaSwag, etc.) pulled automatically from HuggingFace, then apply chaos profiles that introduce realistic failure modes: malformed JSON schemas, artificial latency spikes, 429 rate-limit errors, context-window overflow, and prompt injection payloads. The key output is a degradation delta — evalmonkey shows you exactly how much your agent's accuracy drops under each failure type versus clean inputs. A model that scores 78% on GSM8K normally but drops to 31% when it gets a 429 mid-chain tells you something crucial about its error-recovery behavior that standard benchmarks completely miss. It supports OpenAI, Anthropic (via Bedrock and direct), Azure, GCP, and any Ollama-hosted model. Corbell-AI published this with a clear thesis: agents break in production for infrastructure reasons, not model reasons — and no existing benchmark tests that. evalmonkey was created today (April 17, 2026) and is still at 3 stars, but the core idea is genuinely novel in the evals space.

qmd is a lightweight local search engine built by Tobi Luetke, CEO of Shopify, for indexing and querying personal knowledge bases, documentation, and meeting notes — entirely offline. It combines three retrieval approaches in a single pipeline: BM25 full-text search for exact keyword matches, vector semantic search via ONNX-based embeddings, and LLM re-ranking using GGUF models through node-llama-cpp. All three stages run locally with no cloud dependency. The tool ships in multiple deployment modes: a CLI for ad-hoc queries, a Node.js library for programmatic use, an HTTP service for local API access, and — most useful for AI workflows — a native MCP server that lets Claude Code, Cursor, and similar editors query your local knowledge base directly during coding sessions. The hybrid retrieval approach means it handles both "find the exact error message from last week's standup notes" and "what was our decision about the auth architecture" equally well. What makes this notable beyond its technical approach is provenance: Luetke shipped it as a personal tool he actually uses, not a startup product. The GitHub history shows active iteration and he's been talking about it on X. It's a credible signal of where pragmatic AI-augmented knowledge management is heading for technical users who prefer local-first tools.