Kit for AI vs Dust
Dust is a platform for building company AI assistants. If you want an all-in-one, private and cheaper AI agents platform that also converts documents, remembers, recalls and searches the web, Kit for AI is the better fit.
What Dust does well
Dust helps teams build custom assistants over their data with a polished builder and connectors.
Why teams choose Kit for AI
- Kit for AI gives you the primitives — convert, knowledge base, memory, skills, search — behind a clean API and MCP, so you build on your terms.
- Local private models and flat pricing instead of hosted LLM pass-through costs.
- Start free; no per-seat enterprise plan required.
How Kit for AI is built
The difference isn't just scope — it's technique. Here's what drives more accurate, cheaper, more private answers.
Hybrid retrieval, not vector-only
Most memory and RAG tools rank passages by vector similarity alone, which misses exact terms and names. Kit for AI fuses dense vector search with keyword search using Reciprocal Rank Fusion, then reranks the shortlist with a cross-encoder — so the passage that actually answers the question lands first. Higher precision means fewer chunks in the prompt, and fewer tokens billed.
Grounded answers that cite — or decline
Every answer is generated only from your retrieved passages and cites the exact source. When your documents don't cover a question, the model says so instead of inventing an answer. No confident hallucinations, no untraceable claims.
Token-efficient ingestion
Sources aren't dumped in raw. PDFs, Office files, HTML, URLs and scanned images run through a conversion ladder — including vision-model OCR — into clean, structured Markdown. Clean input cuts token counts sharply versus raw extraction, and gives the model less noise to reason over.
Private by default
Answers run on our own models, so your documents and memories aren't sent to a third-party LLM. You get the retrieval quality without handing your corpus to OpenAI or Anthropic.
One surface, zero retrieval infra
Convert, knowledge bases, memory (remember/recall), web search and skills are one REST API, one MCP server, and a typed SDK. No vector database to run, no embedding pipeline to maintain, no chunking strategy to tune — it's managed end to end.
Feature comparison
| Feature | Kit for AI | Dust |
|---|---|---|
| Retrieval technique | Hybrid (vector + keyword) fused with RRF, then cross-encoder reranking | Vector similarity |
| Grounded, cited answers | Cites every source; declines when your docs don't cover it | Depends on setup |
| Document → clean Markdown / JSON | PDF, Office, HTML, URLs, images/OCR — token-efficient | Connectors |
| AI OCR for scans & images | Vision-model OCR → structured Markdown | Via connectors |
| Retrieval infrastructure to run | None — no vector DB, embeddings or chunking to manage | Managed builder platform |
| Local / private AI (no third-party LLM) | Runs on our own models | Hosted LLM |
| Developer surface | REST API + MCP server + typed SDK | API only / limited |
| Agent memory (remember / recall) | Built-in, deduplicated, semantic | Separate product |
| All-in-one (convert → KB → memory → skills → search) | One platform, one API | Assistant builder |
| Starting price | Free tier · Pro $9/mo | Per-seat + usage |
Pricing
Dust bills per-seat with usage. Kit for AI's flat $9/mo Pro covers the whole toolkit.
FAQ
- Is Kit for AI a Dust alternative?
- If you want the primitives — convert, knowledge base, memory, skills and web search — behind a clean REST API and MCP rather than a hosted assistant builder, yes. You build on your terms, with cited answers grounded in your documents.
- How does it keep costs down vs Dust?
- Dust bills per-seat plus usage that passes through a hosted LLM. Kit for AI answers on our own local models under a flat $9/mo plan, so cost doesn't scale with every seat and token.
- Is my data private?
- Yes — retrieval and generation run on local models, so your documents and prompts aren't sent to a third-party LLM.
Switch to the all-in-one AI toolkit
Convert, build a knowledge base, remember, recall, run skills and search — one API, private models, from free.