Plugin: Parallax - new plugin for academic research - feedback welcome!

[MaxOnAMission]

Disclaimer

Is this project open source? Yes
Is this project completely free? Yes
Is this project vibe-coded beyond the author’s ability to comprehend how it works? No
Community Directory: Parallax

---

Parallax - new plugin for researchers - feedback welcome!

Think sharper before you search, weigh the evidence you find, and keep a trail of why you concluded what you concluded. In your own notes, in your own words.

You know the feeling: forty papers read, a conclusion reached … and six months later you can’t reconstruct why you dismissed the alternative. Or worse: an AI summarised it all so smoothly that you’re no longer sure the thinking was ever yours.

Parallax is a research studio for Obsidian built to guard exactly that. It structures the thinking that happens before and around the searching: your assumptions, your framing, the evidence and what it cannot answer, the beliefs you updated along the way and records every step as artefacts in your own notes. An LLM proposes; you decide, and you hold the pen. Nothing is ever written over your own thinking without asking.

(Why “Parallax”? Astronomers measure the distance of a star by looking at it from two points and watching it shift — depth comes from changing your viewpoint. Research works the same way.)

Most AI plugins for Obsidian answer questions: chat over your vault, summarise papers, generate text. The hard part of research happens before any of that. For example:

• Which assumptions are you smuggling in?
• Which theoretical lens are you (not) using?
• Does the evidence actually support what you’ve come to believe?
• …

Parallax structures that thinking process and records it as artefacts in your own notes. This helps to make sure that six months from now you can still see not just what you concluded, but how and why.

Try it in 60 seconds — free, no API key

1. Install & enable: Community plugins → search for Parallax.
2. Open a note and run Quick search (single provider).
3. Insert real academic references from OpenAlex (~250M scholarly works) straight into your note — free, keyless, no account.

That’s the whole entry fee. Everything beyond it is opt-in: bring any OpenAI-compatible LLM (including a fully local Ollama), or none at all.

What a research session looks like

A session is a plain markdown note. A set of copilots can work on it. Each one in a propose-only mode, they offer suggestions but you decide what to accept and can always edit anything.

Steps:

1. Explore the problem before searching — implicit assumptions and their counter-assumptions, competing definitions, reformulations that would change your whole research programme, and search seeds you’d never find from your original phrasing.
2. Theory — theoretical lenses to think with, each grounded (why it might apply here, what it would predict), including the eliminative list: tempting theories that explain little.
3. Evidence — multi-source literature research (OpenAlex + Semantic Scholar, rank fusion, a transparent rerank) with a graded synthesis: findings labelled strong/mixed/limited, evidence tiers per study design, honest “cannot be answered with this evidence” sections, and clickable [n] citations.
4. Challenge — the sceptic of the suite: pushback on your framing along five dimensions, hardest against your recorded beliefs.
5. Beliefs — the convictions you bring, tracked with status and revision history, and confronted with each new synthesis. Verdicts are proposals; you tick what to adopt.
6. Design — a research agenda out of the synthesis: gaps, sharper questions, fitting study designs. One click opens the next session — the loop closes.
7. Methodological account — the audit trail as a deliverable: question, framing, lenses, search strategy, synthesis and beliefs — assembled deterministically from what actually happened, never written by a model about itself.

The Parallax sidebar is the cockpit: where you are, a recommended next step (never a straitjacket), your artefacts, the evidence gaps in your project, and source↔finding provenance. Sessions group into projects; everything exports as a portable, reproducible bundle (JSON + methodological account + BibTeX).

What Parallax is not

• Not a chatbot. There is no chat window. If you want to talk to your vault, Copilot and Smart Connections do that well.
• Not a reference manager. Already using Zotero Integration or Citations? Keep them — Parallax never manages your library; it feeds it (BibTeX export from its citation register).
• Not a PDF reader or writing tool. PDF++ and your editor own those.

Parallax is the layer those tools don’t cover: the reasoning between question and conclusion.

Why trust the output?

• Graded, not generated-and-gone. Findings carry evidence grades and study-design tiers; claims resting on abstracts say so; a resolving DOI marks existence, never content-truth.
• Traceable. Finding→source links are recorded at generation time; every inserted reference lands in a citation register with occurrence tracking and DOI verification.
• Yours. Copilots propose into review screens with adopt-toggles default-off. Re-running a step over a section you hand-edited asks before replacing. Your edits are leading.

How it works & privacy

• Search is free by default via OpenAlex; Semantic Scholar rides along in multi-source research (an optional free key gets you a faster lane); the Consensus API is available as an optional paid provider.
• AI steps are opt-in: Mistral or any OpenAI-compatible endpoint (OpenAI, Ollama, LM Studio, OpenRouter). With a local Ollama endpoint, all LLM traffic stays on your machine. No LLM key? You still get multi-source search with rank fusion.
• No telemetry, no account, no cloud backend of ours. Keys are stored locally in your vault config and never echoed in error messages. The README has a full table of what leaves your machine, and when.

Works on desktop and mobile (Obsidian 1.5.0+), and the artefacts Parallax writes (section
headings, labels, the methodological account) come in 13 languages — the AI-written prose
follows the language of your question.

Installation

Community plugins → search for Parallax → Install → Enable. That’s it — the free search path needs no configuration at all.

Manual: copy main.js, manifest.json, styles.css from the latest release into .obsidian/plugins/consensus-research/.

Feedback welcome!

I’d love to hear how it works for you, especially:

• Does the graded synthesis hold up in your field? Where does the grading feel too generous or too strict?
• Do the Explore and Challenge steps surface things you hadn’t considered, or just noise?
• How’s the free, keyless search path — good enough to be useful on its own?
• Any bugs or rough edges you run into?

Feel free to open issues on GitHub or reply here. Parallax is free and GPL-3.0. If you find it useful a star on the repo is much appreciated to help others find it!

If you feel really generous you are always welcome buy me a coffee of course

---

GitHub: maxonamission/obsidian-parallax
License: GPL-3.0, free and open source

[MaxOnAMission]

Big update round: your own .bib library, argument maps, and a bridge to qualitative fieldwork (0.16.0)

Since the launch post, Parallax has had a very productive stretch — 0.3.1 → 0.16.0. Rather than list every release, here’s what actually changed for how you work. (Full details: changelog.)

Your reference library is now a living link

Point Parallax at a .bib file in your vault (e.g. a Zotero Better BibTeX auto-export) and:

• Library — insert citation: fuzzy-search your own library (citekey, author, year, title) and insert a formatted reference at the cursor — recorded in the citation register like any search result, no network needed.
• Library — update references (this note / this project): your reference manager fixed a year or title? Parallax shows you exactly which fields would change — per reference, old → new — before anything is written. References you marked verified are never touched.

The .bib file stays strictly read-only, and there’s deliberately no Zotero plugin or API coupling — the file is the interface, so there’s nothing to break when Zotero updates. (If you’ve watched the Zotero-integration plugins go quiet over the years, that design choice is for you.)

Map the argument

A new artefact makes the logical skeleton of your thinking explicit: claims, assumptions, and the supports/attacks relations between them. Map the argument (propose-only) proposes a structure drawn from your question, framing, beliefs, synthesis and challenges — you adopt it element by element. It renders three ways, all native:

• an inline Mermaid map in the note (support solid, attack dotted);
• Argument — open as Canvas for bigger maps — a real .canvas file where re-running keeps your positions and never deletes anything you added yourself;
• a Connections section that turns invisible relations (project hub, follow-up sessions, “shares N sources”) into real wikilinks — so the graph view on a project folder finally shows the project’s shape.

From thinking to fieldwork (qualitative research)

This is the part I’m most excited about. The chain now runs end-to-end:

1. Design interview guide (propose-only) — a topic guide of up to ten questions, each carrying a visible provenance label: which sub-question it serves, which evidence gap it should close, which attacked assumption it probes, or which belief it tests. Export it as a clean fieldwork protocol (plain markdown, date/respondent fill-ins) and take it with you.
2. Record the interview, transcribe it with Voxtral Transcribe (my other plugin), and
3. hand it to Quadro (the QDA plugin) via Export Quadro starter kit: a start codebook generated from your chosen theoretical lenses, extraction templates per research question, and a Data/ intake — file-level hand-off, no runtime coupling.

And because not every question wants the same kind of research: the Research Design copilot now proposes a method fit per question (qualitative / quantitative / mixed, with rationale) — adjustable per question before anything lands.

Also worth knowing

• One session per lens: pick several theoretical lenses and Parallax creates one session per lens, each developing its own framework — parallax by construction.
• Records follow your notes: rename/move a note or project and the citation register + research records repoint automatically; a new cleanup command handles notes that no longer exist (nothing deleted without your confirmation).
• Artefacts land where they belong: review screens now name their landing (“Add to session”, “Add to a new session note”) instead of pasting loose blocks at the cursor.
• Network resilience on mobile: visible retries (“retrying synthesis 2/6…”), a wider retry window, and — when it still fails — a one-line diagnosis telling you whether just the AI endpoint is blocked (DNS filter/VPN) or you’re simply offline.

Everything remains propose-only, free, GPL-3.0, and works keyless (OpenAlex search) — LLM steps stay opt-in, local Ollama included.

Feedback I’d especially value

• If you do interview-based research: does the guide’s provenance labelling match how you actually build topic guides?
• If you keep a .bib library: does the update preview show you the right changes, and anything you miss from the old Zotero plugins?
• Argument maps: useful skeleton or busywork? Where does it break down in your field?

Reply here or open an issue on GitHub. And if Parallax is useful to you, a on the repo helps others find it!