What problem does it solve? Compiling a literature survey on an AI/ML topic requires curating real papers, organizing a taxonomy, and writing cited prose — a slow manual process that often produces one-off artifacts that are never reused. ## Core Features & Use Cases - Provider-agnostic generation: Works with Anthropic, OpenAI, OpenRouter, Fireworks, or any custom OpenAI-compatible endpoint, with provider auto-detection from environment variables. - Wiki-native output: Writes versioned markdown surveys to <wiki>/derived/surveys/<slug>-vN.md, appends deduplicated bibliography rows to sources.md, and indexes pages into an FTS5 store for later retrieval. - Schema-validated research bundles: Enforces that every cited paper key exists in the bibliography and that no bibliography entries are invented. - Use Case: Ask for a survey on "Reasoning Models" anchored to an arXiv survey or awesome-list; the agent builds a research bundle, then the script generates a 6-10 section cited survey saved permanently in your wiki. ## Quick Start Ask the agent to generate a literature survey on your chosen topic using a public anchor URL, targeting your wiki slug.