A Puli Consulting knowledge tool
RoamPrompt brings source-grounded AI directly into Roam Research—helping scholars, writers, and knowledge workers read notes, images, academic papers, books, reports, news articles, and scanned clippings, then turn key ideas into structured notes and reviewable atomic cards without leaving their graph.
v0.4.5 · Available now in Roam Depot
From source to system
Preserve source identity, research context, and evidence locations.
Build a QEC note that separates findings, interpretation, and uncertainty.
Approve only candidates that deserve a place in your graph.
Link ideas deliberately without inventing relationships.
Weekly Review
RoamPrompt helps you notice themes in recent quotations, revisit open work, and draft a short synthesis. The command explains exactly what will leave your browser before anything is sent.
Add [[Quotes]] to eligible quote blocks and use Roam’s open [[TODO]] task syntax. No dedicated destination pages are required in advance.
RoamPrompt selects the newest 50 eligible Quote strings and newest 50 open TODO strings created during the previous seven days. [[Wisdom & Quotes]] optionally marks a quote as already curated.
After explicit confirmation, those selected strings go directly to Gemini so it can group quote themes, reproduce tasks for review, and draft the weekly synthesis. That semantic work is not performed locally.
The report replaces the bullet focused when the command starts. Original Quote and TODO blocks stay unchanged: RoamPrompt does not move, delete, complete, or refile them.
Linked references are not migration. Page links in the generated report may appear as linked references, but the source blocks remain in their original locations.
Inside RoamPrompt
Select or reject each candidate. Claims remain paired with epistemic status and evidence locations.
Keep source identity, evidence, method, synthesis, limitations, and open questions together.
Each Zettel carries one coherent point—even when related observations gain meaning together.
Document workflows
Run RoamPrompt: Read PDF on Current Page. RoamPrompt finds the synced attachment, reads it with Gemini after consent, and presents candidate findings with page and section locators for review.
Attach an image-only clipping PDF and identify the target headline when useful. RoamPrompt separates the article from advertisements and neighbouring columns, then distinguishes reported events, attributed statements, and contemporary predictions.
Why RoamPrompt
Read, review, and insert structured knowledge without moving the workflow into a separate AI application.
Process academic papers, books, reports, news articles, and image-only clippings while preserving the correct source type.
Turn key source content into QEC or News Source Notes and reviewable Zettelkasten cards built for reuse and connection.
Keep evidence anchors and epistemic status visible while separating existing graph links from suggested concepts.
From the Slip Box to Responsible AI
This research-grounded series begins with the centuries-old problem of external memory, moves through Niklas Luhmann and atomic notes, and arrives at a practical boundary for AI: accelerate the work while keeping evidence, interpretation, and human judgement visible.
Zettelkasten belongs to a much older family of excerpting, commonplace books, indexes, and card catalogues. Its durable question is not how to store more, but how to keep a thought findable, revisable, and able to meet another thought later.
Historical grounding: Markus Krajewski’s Paper Machines ↗
Read on LinkedIn ↗
Luhmann built two major card collections over decades. Their value lay not in accumulating quotations, but in writing compact propositions and creating relations that could support later comparison, questioning, and selection.
Archive record: Niklas Luhmann Archive ↗
Read on LinkedIn ↗
An atomic note is not merely small. It makes one source-aware proposition clear enough to test, cite, revise, and reuse. Too broad, and links become ambiguous; too fragmented, and the knowledge network loses coherence.
Practical grounding: Andy Matuschak on atomicity ↗
Read on LinkedIn ↗
A graph can help ideas meet, but more links do not automatically create intelligence or wisdom. RoamPrompt’s Read → Structure → Review → Connect workflow accelerates reading while leaving the consequential decisions—meaning, evidence, connection, and permanence—with the user.
Current implementation: RoamPrompt v0.4.5 ↗
Read on LinkedIn ↗The four cards use the dedicated LinkedIn covers from the final publication package. Their illustrations are editorial rather than archival; the fourth uses a conceptual knowledge graph rather than a literal brain map. Each card has a named link slot ready for its public LinkedIn URL.
BYOK today · broader choice ahead
RoamPrompt does not supply a shared key. Create your own Gemini API key with Google, install RoamPrompt from Roam Depot, and enter the key in RoamPrompt’s extension settings. Puli Consulting does not receive that key through this website. Each AI workflow describes the selected content and asks for consent before sending it directly to Gemini; Weekly Review also states its seven-day window and item caps.
Planned development will extend bring-your-own-key support to additional LLM APIs, giving users more choice over model capability, availability, account, quota, and cost. These additional providers are roadmap items and are not part of the current public release.
Google’s official API-key guidance ↗Sign in, accept Google’s terms, and open API Keys.
Create a key in an appropriate Google Cloud project and copy it securely.
Install RoamPrompt through Roam Research → Settings → Roam Depot, then paste the key into RoamPrompt’s Gemini API Key setting.
Never put it in a Roam block, screenshot, or GitHub repository. Review quota and billing; rotate it immediately if exposed.
Documentation
Promote · Research · Co-develop
We welcome conversations with Roam practitioners, researchers, educators, extension developers, and institutional partners interested in promoting RoamPrompt or co-developing responsible, source-grounded knowledge workflows.
