A Self-Hosted Research Workspace with Cited Answers
Keep sources, notes, search, cited answers, graph relationships, and evidence-backed learning patterns in one self-hosted workspace.
Research often fragments across a bookmark manager, a folder of files, a notes app, and a chatbot with no access to the evidence you chose to keep.
A self-hosted research workspace can keep the source, your interpretation, and later synthesis close together.
Capture evidence into one Library
Arivu captures web pages, selected passages, files, and quick notes. Library leads with saved and authored material in one searchable content view, while generated concepts and entities remain available separately and in Graph.
Capture does not require an AI provider. Local extraction stores readable material, and full-text Search provides a dependable path back to it.
Write Notes that point to sources
Use Notes for conclusions, questions, comparisons, and drafts. Link each important claim to the source material that shaped it. Backlinks preserve the reverse path so a source can reveal the Notes that depend on it.
Explicit links record your judgment. Local extraction can derive concepts, entities, and sources. Optional embeddings can add similarity relationships when configured. Keeping those relationship types distinct makes the Graph more trustworthy.
Search and Ask solve different problems
Search returns items that match text. It is the right tool for a title, phrase, person, domain, or remembered term.
Ask builds a cited response from saved material. It is useful for questions such as:
- Which saved sources disagree about this decision?
- What evidence have I collected for this argument?
- Which Notes mention a migration risk?
- What did these sources say about a recurring concept?
The citations are the important part. Open them, check that they support the response, and save your own conclusion as a Note.
Ask does not require or call a model provider. Optional providers can improve stored summaries and related enrichment, while the workspace remains usable when external models are unavailable.
Investigate relationships with Graph
Graph shows typed nodes and relationships with provenance and confidence. Focus it around a note, concept, entity, or source rather than treating the complete network as a finished answer.
Use the inspector and list alternative to verify exact relationships. A visible edge is a route for investigation, not proof by itself.
Learn from evidence-backed Insights
Insights can surface emerging themes, recurring connections, forgotten value, knowledge gaps, and serendipitous connections. These patterns are deterministic and cite the material that produced them.
Only supported, current derived material qualifies for analytical Insights. Recommendations are possible next actions, not claims that an analytical pattern exists. That separation keeps suggestions useful without overstating the evidence.
That makes Insights useful for deciding what to read next, which connection deserves an explicit link, or where your research lacks evidence.
Keep the operating model understandable
Arivu’s core runs as a single Go binary with SQLite. Self-hosting keeps the application and database within infrastructure you control, but it also makes deployment, updates, and backups your responsibility.
Use the documented import and export paths, know where the database is stored, and test restoration. Portable data is valuable only when the recovery process works.
On supported hosts, the installer can manage an isolated native capture runtime alongside the core Go application. It can improve weak direct extraction, retain local reader images, and add screenshots or offline HTML without using your browser profile. The core-only direct path remains available independently, and PDF stays an operator choice because of its storage cost.
If retained source, screenshot, offline HTML, PDF, or reader-media assets matter, use installer-managed backups that keep adjacent assets and their manifest with SQLite. Account-level JSON remains useful for portability, but serves a different recovery boundary.
When sharing evidence, publish a revocable, point-in-time secret-link snapshot rather than exposing private Notes or a live workspace. The bookmark disclosure creates an item-only share; authenticated APIs can deliberately include selected screenshot or PDF artifacts.
Who this fits
This model fits an individual researcher, writer, reader, or operator who wants connected source work without making AI mandatory. It can also support people on a private multi-user instance, while remaining centered on personal knowledge rather than team co-editing or delivery planning.
Start with Search and Ask, the Graph guide, and Preservation and Sharing.