Most teams either trust a closed cloud assistant or wrestle with brittle, ad-hoc document search. This platform treats your documents and web content as a structured, semantic knowledge layer you can query, extend with agents, and automate — without forcing you into a proprietary cloud.
What Sets It Apart
- Unified semantic index across web pages, PDFs, Markdown, Notion, Word and other formats, so searches return passage-level answers with provenance rather than opaque model guesses — useful when you need sourced responses.
- Agent + automation capabilities: create agents with custom knowledge, persona, tools and scheduled tasks to run research workflows or deliver notifications, turning one-off queries into repeatable pipelines.
- Local-first, multi-provider LLM support: works with local models (Llama-family, Qwen, Mistral) and cloud models (OpenAI, Anthropic, Gemini) so you can balance latency, cost and privacy.
- Multi-platform integrations (browser, Obsidian, Emacs, desktop, mobile, WhatsApp) that make the knowledge base accessible where you work, not confined to a single UI.
Who It's For and Trade-offs
Great fit if you want private, self-hosted semantic search and agentic automations over proprietary docs, or if you need to switch between local and cloud LLMs for cost/privacy reasons. It’s also useful for teams that want reproducible research/automation pipelines tied to a document corpus.
Look elsewhere if you need a turn-key, fully managed enterprise support SLA out of the box (the project is open-source and community-driven, and enterprise features typically require additional configuration or a vendor plan). Also, very large-scale deployments may need custom infra and tuning for vector store and indexing costs.
Where It Fits
Compared with single-purpose RAG libraries or closed assistants, this project sits between a developer-oriented RAG stack and a full SaaS assistant: it provides an opinionated app + orchestration layer for retrieval, reasoning, agents and multi-client access, and is designed to be self-hosted or run as a cloud service.
How It Works (high level)
The system ingests documents and webpages, builds embeddings and a semantic index, and routes queries through a retrieval layer before hitting an LLM. Agents are composable — you can give an agent a persona, knowledge subset, tools (HTTP, file access, search), and schedules so it can run tasks autonomously on your behalf.
Honest trade-offs: licensing is AGPL-3.0 (check for commercial constraints), and running advanced pipelines at scale requires attention to vector store, embedding costs, and model choice. For people who need an auditable, portable knowledge/agent stack, this is one of the more complete open-source options available today.
