Langhuan’s first public release established the product’s basic boundary: turning inputs like PDF, DOCX, Markdown, TXT, CSV, and XLSX into retrievable, traceable knowledge structures. It doesn’t generate answers — it hands stable knowledge-processing and retrieval capability to upper-layer applications.
One download, one service
This release collapses the components you need to run into a single deployable service:
- The single-binary entrypoint carries REST, MCP over HTTP, and the Web Console at once.
- Docker Compose provides a one-command experience with PostgreSQL, pgvector, and the tokenizer.
- MIT License, bilingual README, and public docs ship with the project.
- Document ingestion, parsing, chunking, indexing, and retrieval are wired together through async tasks, so HTTP requests don’t wait for long jobs to finish.
Two entry points for retrieval
The REST API is suited to explicit HTTP integration, while MCP over HTTP lets MCP-capable clients discover and call knowledge-base tools directly. Both share the same workspace isolation and retrieval contract, so the same knowledge data doesn’t need two sets of integration logic.
document → parse → chunk → index → REST / MCP retrievalWhere it fits
If you already have a chat app, an agent, or a business system and just want to make document processing and retrieval a dependable foundation, this version is the minimal viable starting point. The upper-layer application keeps the answering, orchestration, and permission experience; Langhuan focuses on “how knowledge is organized, found, and traced.”