A semantic knowledge base for your Nextcloud.
Astrolabe Cloud turns your Nextcloud into searchable, retrievable knowledge — indexed by meaning and served back over the search bar you already use and a hosted MCP endpoint. It does the retrieval; your tools do the rest.
Document indexing & retrieval
From your documents to answers by meaning.
Keyword search only finds the words you remember. Astrolabe indexes your content as vector embeddings and retrieves it by meaning, so “the contract we signed in spring” finds the right document even if it never used those words.
- 01
Index
Astrolabe watches your Nextcloud and chunks new and changed content — Notes, Files, Calendar, Deck — as it lands. Each chunk is turned into a vector embedding so it can be matched by meaning.
- 02
Store
Embeddings live in a per-tenant vector index, kept in sync with your Nextcloud. No index to provision, tune, or back up on your side.
- 03
Retrieve
A query is embedded the same way and ranked against the index with hybrid semantic + keyword scoring — so you get conceptual recall and exact-match precision in one result set.
Capabilities
Built for retrieval, ready for agents.
Everything Astrolabe does is in service of one job: getting the right context out of your Nextcloud, fast and accurately — whether a human is searching or an agent is.
- Semantic search across Notes, Files, Calendar, and Deck from one index
- Hybrid ranking — meaning for recall, keywords for precision
- Tunable result count and score thresholds for narrow or exploratory queries
- Citations back to the source note, file, or event
- Retrieval context any MCP client can feed to its own model for RAG-style answers
Nextcloud integration
It lives where your data already does.
You bring your own Nextcloud — we never host or copy your files. Connect it once with an OIDC client, and Astrolabe provisions a managed backend with a per-tenant index that stays in sync as your content changes.
- Results land in the Nextcloud unified search bar you already use.
- One index spans Notes, Files, Calendar, and Deck.
- Per-tenant isolation with secret-managed credentials.
- The Astrolabe app and backend are open source — inspect or self-host.
RAG via MCP
Generation happens in your client.
Astrolabe ships a hosted MCP endpoint. Plug in Claude, Cursor, or any MCP client and it can search your knowledge bank and pull back the most relevant passages — the retrievalhalf of retrieval-augmented generation. Your client's model reads that context and writes the answer, with citations back to the source.
Retrieve
AstrolabeYour query is embedded and matched against your per-tenant index; the top passages come back as context.
Augment
Your MCP clientThe client assembles those passages into the prompt it sends to its own model.
Generate
Your modelThe model you already use — and pay for — writes the grounded answer.
A deliberate choice
Why we don't bundle an LLM.
Astrolabe is the retrieval layer, not another chatbot. We deliberately stop at giving your client clean, relevant context — and let you choose the model that reads it. That keeps the product focused and your data path short.
Bring your own model
Use the LLM you already trust and pay for — no second AI bill, no lock-in to ours.
A shorter data path
Your content is embedded and retrieved in the EU; it isn't forwarded to a bundled generation model you didn't pick.
Do one thing well
Retrieval quality is the hard part of RAG. We invest there instead of chasing a general-purpose model.
Composable by design
An open MCP endpoint plugs into whatever agent stack you run, today and as it changes.
EU data sovereignty, by architecture.
Every component that touches your content sits in Europe. Not a compliance checkbox — the actual data path. Here is where each part runs:
Compute — Hetzner
The managed backend and your vector workloads run on Hetzner in the EU (Germany & Finland). No hop through a US hyperscaler.
Embeddings — Mistral AI
Text is embedded by Mistral AI, a European model provider — your content is vectorised in-region, not shipped to a US API.
Vectors — Qdrant Cloud
Embeddings are stored and searched in Qdrant Cloud, hosted in the EU, isolated per tenant.
Your files — your Nextcloud
You bring your own Nextcloud; the source documents never leave the server you already control. We index, we don't host.
Put your Nextcloud knowledge to work.
We're rolling out early-access tenants now. Join the waitlist and we'll be in touch.