Raw sources in.
Refined intelligence out.

Crucible is an autonomous AI research engine. You define topics of interest, and it searches the web, scrapes documents, evaluates their relevance, and organizes everything into a private Obsidian-compatible knowledge vault, all without manual intervention.

All documents are stored locally as markdown files with YAML frontmatter in an Obsidian-compatible vault on your machine. Your data never leaves your device unless you choose a cloud AI provider for processing.

Yes. Crucible supports Ollama for fully private local inference on your own GPU, as well as OpenAI and Anthropic cloud APIs. You can configure different models per pipeline step to balance cost and quality.

Depending on your plan: DuckDuckGo, Brave Search, SearXNG (meta-search across dozens of engines), RSS feeds, arXiv academic papers, government data APIs (FRED, EIA), and direct PDF extraction.

Available on the Pro plan, the Vault Curator is a background AI agent that continuously maintains your vault by recategorizing documents, auditing quality scores, detecting duplicates, surfacing cross-document connections, and flagging stale content.

Only if you want to run AI models locally with Ollama. If you use OpenAI or Anthropic cloud APIs instead, no GPU is required. The research pipeline itself runs on any modern computer.

Yes. The free tier includes manual collection triggers, up to 50 documents per cycle, DuckDuckGo search, a basic vault browser, and a single LLM provider. No credit card required.

Search engines return links. Crucible goes further: it reads the pages, evaluates their quality, extracts key information, stores structured documents in your vault, and builds a knowledge graph of relationships between them. It's research automation, not just search.

AI chatbots are great at answering questions, but they give you a conversation, not a system. Every session starts from scratch, sources are unverifiable, and the output disappears into a chat log you'll never revisit. Crucible is research infrastructure: it autonomously searches real sources, verifies and scores what it finds, stores original documents with full provenance, and builds a permanent knowledge graph that compounds over time. You're not getting a summary from a model's memory. You're building an intelligence asset you own.

AI search tools are a step closer. They cite real sources and give better answers. But they're still one question, one answer, one session. You don't accumulate anything. Crucible runs autonomous research sweeps across dozens of sources simultaneously, stores every document it finds, scores and categorizes them, deduplicates across runs, and builds a knowledge graph that grows with every cycle. It's the difference between asking a librarian a question and building your own library.

Yes. Crucible stores documents as standard markdown with YAML frontmatter, fully compatible with Obsidian. You can point Crucible at an existing vault directory and it will integrate alongside your existing notes.

Research analysts use Crucible for competitive intelligence and market monitoring. Data scientists use it to aggregate training data and track papers. Prediction market traders use it to build information edges from diverse sources. Hobbyists use it to curate knowledge on any topic they're passionate about. Anyone doing serious research across multiple sources benefits from automated collection and organization.

Crucible focuses on the collection and organization layer: gathering raw data from the web into structured, queryable documents. It pairs well with analysis tools like Jupyter, Tableau, or custom scripts. The knowledge graph and structured metadata make it easy to export curated datasets for downstream analysis, modeling, or simulation work.