- Category
- AI Tools
- Rank
- No. 1632Tools index
Previous survey · No. 1639 ·
- Pricing
- Open Source
- Type
- TOOL
- Builder
- badlogic
- GitHub
- 32 stars
- Date
About
A RAG and LLM information retrieval tool for querying your own document collections.
What it does
Doxie turns curated support material into branded conversational bots you host yourself. An admin imports FAQs, sitemaps, Markdown archives, or forum exports, processes them into searchable embeddings, checks the resulting document segments, then publishes chat or one-off question interfaces.
Why it's ranked here
Doxie covers an unusually complete path from source preparation to public chat. It includes ingestion jobs, document inspection, bot configuration, streaming answers, session replay, and optional reranking. The appeal is breadth, tempered by a deployment stack and source formats that demand hands-on ownership.
What's good
The admin workflow exposes the parts that often stay hidden in retrieval systems. Operators can inspect embedded segments, run similarity searches, follow processing logs, tune prompts and token limits, and view answer debug data. Bot import and export, custom branding, multiple knowledge sources, and persisted conversations make it practical for repeated publishing work.
Tradeoffs
Self-hosting requires Node services, MongoDB, a vector service, deployment configuration, and an OpenAI key. Cohere is another dependency when reranking is enabled. Authentication uses a shared admin token cached in browser storage. Ingestion accepts four defined source shapes, and reprocessing replaces the complete vector collection rather than updating individual records.
How to use it well
Use Doxie when a technical team owns a bounded documentation or support corpus and wants inspectable, branded assistants. Start with one source, process it, test retrieval quality, then tune prompts before combining collections. It does not cover managed hosting, general-purpose data integration, or unrestricted file ingestion.
Technical notes+
README.md describes a Lit and Tailwind single-page client, an Express API, MongoDB persistence, a background processor, and a custom Java nearest-neighbour HTTP service. package.json provides npm builds for CSS, site, server, and JNN, with esbuild and TypeScript tooling. src/app.ts defines the browser routes; src/common/api.ts holds source, bot, job, chat, REST, and chunked-stream contracts; src/pages/admin.ts implements token-gated bot and source administration; src/pages/chat.ts sanitizes rendered Markdown and displays streamed debug payloads; src/server/jobs.ts persists the waiting, running, succeeded, failed, and stopped lifecycle. src/server/rag.ts also contains Chroma-backed retrieval classes, while README.md describes JNN as the vector store, so the supplied files do not establish which backend is wired into the running stack.
Observed
- License
- MIT
- Primary application language
- TypeScript, with a Java nearest-neighbour sidecar
- Packaging and install surface
- npm build scripts plus Docker Compose deployment workflows
- Interfaces
- Web admin UI, public chat UI, one-off answer UI, and HTTP API
- Knowledge source types
- FAQ, sitemap, Markdown ZIP, and Flarum JSON dump
- Hosting model
- Self-hosted Node services with MongoDB and a vector service
- External model services
- OpenAI is required; Cohere reranking is optional
Read from README.md, package.json, src/app.ts, src/appstate.ts, src/pages/bot.ts, src/pages/main.ts, src/common/api.ts, src/pages/chat.ts, src/server/rag.ts, src/utils/i18n.ts, src/pages/index.ts, src/pages/admin.ts, src/server/jobs.ts, src/utils/icons.ts, src/utils/state.ts.
What it can do
Index document collections for search
Document files (PDF, text, etc.) → Searchable document index
Answer questions about document content
Natural language question and document collection → AI-generated answer with source references
Retrieve relevant document passages
Search query or question → List of relevant text excerpts with source documents
Generate summaries of document collections
Set of documents or document collection → AI-generated summary text
Extract specific information from documents
Information request and document collection → Extracted data points with source citations
Compare information across multiple documents
Comparison query and document collection → Comparative analysis with source references
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.