Emdash vs evalmonkey

Emdash is a desktop application from Y Combinator's W26 batch that lets developers run multiple AI coding agents simultaneously, each isolated in its own Git worktree. Rather than switching between Claude Code for one task and Codex for another, you launch parallel agents from one interface, review their diffs in one place, and merge the results through a queue that handles the Git complexity automatically. It supports 23 CLI agent providers including Claude Code, Qwen Code, Hermes Agent, Amp, and OpenAI Codex. The remote development story is particularly strong: Emdash connects to remote machines via SSH/SFTP with keychain credential storage, meaning you can run GPU-heavy agents on a beefy remote devbox while managing everything from your laptop. Ticket integration with Linear, GitHub, and Jira means you can drag a ticket directly onto an agent and watch it work — no copy-pasting requirements into a chat window. Built with Electron and TypeScript with SQLite for local storage, Emdash is local-first by design — your code never touches Emdash's servers, only your chosen agent providers. The project is MIT-licensed, open source, and has accumulated 3,700+ commits since its YC batch. At the intersection of the multi-agent workflow boom and the need for developer tooling that actually scales to parallel workstreams, Emdash is one of the more credible attempts at solving a real daily pain.

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.