This arti­cle start­ed as a back­ground chap­ter for my Ger­man-lan­guage Obsid­i­an guide at obsidianguide.de. While writ­ing it, I real­ized it works just as well as a stand­alone read, so here it is in Eng­lish too. If you have ever won­dered why Obsid­i­an feels dif­fer­ent from oth­er note-tak­ing tools, the answer is not in the app itself but in the decades of ideas it draws from.

Obsid­i­an did­n’t come out of nowhere. Behind the app lie decades of ideas, failed exper­i­ments, tech­ni­cal stan­dards, and aca­d­e­m­ic research, all aimed at the same fun­da­men­tal prob­lem: how does a per­son man­age their knowl­edge in a way that keeps it from get­ting lost, but instead lets it grow and con­nect?

The his­to­ry splits into two par­al­lel threads. The first is the sto­ry of ideas around net­worked knowl­edge and PKM. The sec­ond is the tech­ni­cal evo­lu­tion of markup lan­guages, which were meant to make writ­ing and struc­tur­ing text read­able for both humans and machines. Both threads con­verge in Obsid­i­an.

Networked Knowledge

1945: As We May Think (Van­nevar Bush) — In the mag­a­zine “The Atlantic Month­ly,” Van­nevar Bush sketched out the Memex con­cept, which is sur­pris­ing­ly close to what Obsid­i­an can actu­al­ly do.

Bush described Memex as a device where a per­son could store books, notes, and oth­er arti­facts that could be con­nect­ed through what he called “asso­cia­tive trails.” That maps pret­ty direct­ly onto what Obsid­i­an calls bidi­rec­tion­al links.

What Bush crit­i­cized back then still rings true today: knowl­edge is pro­duced in enor­mous quan­ti­ties, but we have almost no tools to con­nect it the way human think­ing actu­al­ly works, which is asso­cia­tive, not hier­ar­chi­cal. The prob­lem was not access to infor­ma­tion. It was the inabil­i­ty to build last­ing, nav­i­ga­ble con­nec­tions between pieces of it.

Obsid­i­an’s graph struc­ture and back­links are dis­tilled direct­ly from that idea, even though hard­ly any­one who uses Obsid­i­an dai­ly has ever read the essay. Online Reprint

1965: Project Xanadu (Ted Nel­son) — Ted Nel­son coined the term “hyper­text” in 1965 and also began work on Project Xanadu, a sys­tem he envi­sioned as a uni­ver­sal library of every doc­u­ment ever writ­ten. His cen­tral idea was the con­cept of tran­sclu­sion: con­tent should not be copied, but embed­ded direct­ly from its source, with the con­nec­tion to the orig­i­nal pre­served and attri­bu­tion auto­mat­ic.

In some ways, Xanadu was more rad­i­cal than today’s web and more rad­i­cal than Obsid­i­an: links would nev­er go dead, every con­nec­tion would be bidi­rec­tion­al, and the ori­gin of any text would always be trace­able. That sounds famil­iar to any­one who knows Obsid­i­an’s back­links and tran­sclu­sion, but Nel­son was think­ing on a glob­al scale.

The project was nev­er fin­ished and is con­sid­ered one of the most famous failed soft­ware projects in his­to­ry. Nel­son worked on it for over fifty years with­out ship­ping a ver­sion that any­one actu­al­ly used. That Nel­son was right about the core idea is demon­strat­ed by the fact that Obsid­i­an and sim­i­lar tools now imple­ment in minia­ture what he had in mind for all of human knowl­edge. (Wikipedia)

1968: NLS and the “Moth­er of All Demos” (Dou­glas Engel­bart) — On Decem­ber 9, 1968, Dou­glas Engel­bart demon­strat­ed a sys­tem called NLS (oN-Line Sys­tem) in San Fran­cis­co before an audi­ence of about a thou­sand com­put­er sci­en­tists. It is con­sid­ered one of the most influ­en­tial demon­stra­tions in the his­to­ry of com­put­ing. In nine­ty min­utes, Engel­bart showed hyper­text, the com­put­er mouse, video con­fer­enc­ing, and real-time col­lab­o­ra­tive edit­ing live, all con­cepts that would not become stan­dard for decades.

For the his­to­ry of Obsid­i­an, the hyper­text aspect is what mat­ters most: NLS allowed doc­u­ments to be con­nect­ed by links and nav­i­gat­ed between, at a time when most com­put­ers did not even have screens. But Engel­bart’s broad­er ambi­tion is worth not­ing. He called his research pro­gram “Aug­ment­ing Human Intel­lect.” The goal was not to auto­mate tasks. It was to extend what a per­son could think, remem­ber, and under­stand by giv­ing them bet­ter tools for han­dling con­nect­ed infor­ma­tion. That fram­ing is essen­tial­ly what PKM is still about today. Wikipedia

1987: Knowl­edge Nav­i­ga­tor (Apple) — In 1987, Apple under CEO John Scul­ley pro­duced a con­cept video show­ing a vision that was decades ahead of its time: a tablet-like device with an AI assis­tant that inter­acts with the user in nat­ur­al lan­guage, access­ing a net­worked sys­tem of doc­u­ments, notes, and exter­nal infor­ma­tion sources. The assis­tant does not just pas­sive­ly nav­i­gate con­tent but under­stands con­text, sug­gests con­nec­tions, and pre­pares infor­ma­tion based on what is rel­e­vant in the moment.

What makes the video remark­able is not the hard­ware vision, which admit­ted­ly looks a lot like the iPad, but the idea of AI as a nav­i­ga­tion lay­er over a per­son­al knowl­edge sys­tem. The user does not need to search and link man­u­al­ly. Instead, they describe what they need, and the assis­tant moves through the net­worked knowl­edge on their behalf.

That is exact­ly what tools like Note­bookLM, AI plu­g­ins for Obsid­i­an, or Claude as an assis­tant lay­er over your own vault are start­ing to make real today. Apple in 1987 did not just antic­i­pate the iPad, they also posed the ques­tion that PKM enthu­si­asts are wrestling with now: what hap­pens when AI active­ly nav­i­gates inside a per­son­al knowl­edge net­work? (YouTube)

1990: Hyper­text and Hyper­me­dia (Jakob Nielsen) — Before the web exist­ed, hyper­text was already a seri­ous research field, and Jakob Nielsen’s 1990 book is one of the clear­est sum­maries of what that research had learned. One of his cen­tral obser­va­tions was the “lost in hyper­space” prob­lem: users nav­i­gat­ing a hyper­text sys­tem often lose track of where they are, where they have been, and how to get back. The more nodes and links a sys­tem has, the worse the prob­lem gets.

Nielsen’s pro­posed solu­tions, con­tex­tu­al ori­en­ta­tion cues, struc­tured overviews, con­sis­tent nav­i­ga­tion pat­terns, read direct­ly like a design check­list for mod­ern PKM tools. Obsid­i­an’s graph view, the back­links pan­el, and bread­crumb plu­g­ins all address exact­ly the dis­ori­en­ta­tion he described. The fact that his exam­ples were built around Apple Hyper­Card rather than a web brows­er is a reminder of how long these prob­lems pre­date the tools we use today. (Link to bor­row­ing options)

1989–1991: The World Wide Web (Tim Bern­ers-Lee) — Tim Bern­ers-Lee pro­posed the web in a 1989 memo at CERN, devel­oped it through 1990, and pub­lished the first pub­lic web­site in August 1991. What he built was, at its core, a sim­pli­fied ver­sion of what Bush and Nel­son had imag­ined: linked doc­u­ments nav­i­ga­ble through hyper­links.

The key dif­fer­ence from Nel­son’s Xanadu was that Bern­ers-Lee delib­er­ate­ly chose sim­plic­i­ty over cor­rect­ness. Links only point in one direc­tion. There are no auto­mat­ic back­links, no tran­sclu­sion, no guar­an­tee that a link will still work tomor­row. Bern­ers-Lee lat­er said he knew about these lim­i­ta­tions but decid­ed that get­ting some­thing deployed mat­tered more than get­ting it per­fect.

The web became sim­ple enough to go glob­al, but it did not actu­al­ly solve the core prob­lems Bush and Nel­son had described. Every “404 Not Found” error is a small reminder of that trade­off.

From an Obsid­i­an per­spec­tive, that is telling: what Obsid­i­an offers with back­links, graph view, and tran­sclu­sion is, in a sense, an attempt to retro­fit what the web con­scious­ly left out when it was built. (Wikipedia)

1995: The Wiki (Ward Cun­ning­ham) — In 1995, Ward Cun­ning­ham put the first wiki online, which he sim­ply called Wiki­Wiki­Web. The core idea was rad­i­cal­ly sim­ple: any page can be edit­ed by any­one, any page can link to oth­er pages, and new pages come into exis­tence the moment you cre­ate a link that points to some­thing that does not exist yet. The wiki was the first wide­ly used sys­tem to make bidi­rec­tion­al think­ing prac­ti­cal­ly work­able.

Cun­ning­ham’s insight was that the bar­ri­er to cre­at­ing a new node should be zero. You do not plan your knowl­edge struc­ture in advance. You write, link, and the struc­ture emerges from use. That prin­ci­ple is exact­ly what Obsid­i­an encour­ages with its “just cre­ate a note” phi­los­o­phy.

The term “wik­ilink” in Obsid­i­an comes direct­ly from Cun­ning­ham’s inven­tion. The dou­ble square brack­et syn­tax [[Note Name]] orig­i­nates in wiki markup and is today the cen­tral nav­i­ga­tion ele­ment in Obsid­i­an. What Cun­ning­ham built for col­lab­o­ra­tive knowl­edge bases, Obsid­i­an uses as the foun­da­tion­al prin­ci­ple for per­son­al knowl­edge spaces. (Wikipedia)

1990s: Knowl­edge Man­age­ment as a Research Field — Knowl­edge man­age­ment, mean­ing all strate­gic and oper­a­tional activ­i­ties aimed at mak­ing the best pos­si­ble use of knowl­edge, became a seri­ous research field in the 1990s, pri­mar­i­ly in busi­ness admin­is­tra­tion and com­put­er sci­ence. The focus was almost entire­ly on orga­ni­za­tions: how does a com­pa­ny pre­serve the knowl­edge of its employ­ees, how does it make that knowl­edge acces­si­ble to oth­ers, and how does it pre­vent knowl­edge from dis­ap­pear­ing when some­one leaves? Lotus Notes, devel­oped from 1984 onward, was the flag­ship exam­ple: a pow­er­ful sys­tem for col­lab­o­ra­tive doc­u­ment man­age­ment and knowl­edge data­bas­es, built around the log­ic of the orga­ni­za­tion, not the log­ic of the indi­vid­ual thinker. (Knowl­edge Man­age­ment)

Per­son­al knowl­edge man­age­ment, by con­trast, is not about knowl­edge in a team or orga­ni­za­tion but about the per­spec­tive of a sin­gle indi­vid­ual, regard­less of whether the con­text is pro­fes­sion­al or per­son­al. PKM is explic­it­ly a bot­tom-up approach that starts with the per­son, not the insti­tu­tion, and that ori­ents itself around how human think­ing actu­al­ly works: asso­cia­tive, not hier­ar­chi­cal. For Obsid­i­an users, PKM is the rel­e­vant frame. The goal is not to man­age the knowl­edge of a depart­ment, but to orga­nize your own think­ing, con­nect it, and make it usable over the long term.

1990s: Con­tent Man­age­ment Sys­tems — Par­al­lel to PKM research, anoth­er branch of knowl­edge orga­ni­za­tion emerged in the 1990s: con­tent man­age­ment sys­tems, or CMS. Where PKM empha­sizes the indi­vid­ual per­spec­tive, a CMS solves the prob­lem at the orga­ni­za­tion­al lev­el: how does a news­room, a com­pa­ny, or an insti­tu­tion man­age large amounts of con­tent, who can edit what, and how is con­tent pub­lished in a struc­tured way?

A CMS sep­a­rates con­tent, struc­ture, and pre­sen­ta­tion: the author writes text, and the sys­tem han­dles lay­out and pub­lish­ing. Well-known exam­ples include Word­Press, TYPO3, and Dru­pal. That sep­a­ra­tion will sound famil­iar to any­one who uses Obsid­i­an: you write in plain Mark­down, and some­thing like Quartz han­dles the ren­der­ing as a web­site. The dif­fer­ence is fun­da­men­tal, though. A CMS is built for pub­lish­ing and col­lab­o­ra­tion. Obsid­i­an is built for per­son­al think­ing and link­ing. Pub­lish­ing an Obsid­i­an vault with Quartz is, in a way, build­ing a high­ly per­son­al CMS, but one that starts from a com­plete­ly dif­fer­ent phi­los­o­phy. (A Brief His­to­ry of the Con­tent Man­ag­ment Sys­tem)

Technical Foundations

1986: SGML (Stan­dard Gen­er­al­ized Markup Lan­guage) — SGML was adopt­ed as an ISO stan­dard in 1986 and was the first wide­ly used meta­lan­guage for defin­ing markup lan­guages. The core idea was for­ward-look­ing: doc­u­ment con­tent and struc­ture should be clean­ly sep­a­rat­ed, and the markup should be machine-read­able with­out replac­ing the actu­al text.

SGML was pow­er­ful, but cor­re­spond­ing­ly com­plex. It was used main­ly in pub­lish­ing, gov­ern­ment agen­cies, and the aero­space indus­try, any­where large vol­umes of doc­u­ments had to be man­aged in struc­tured form. For ordi­nary users it was sim­ply too much effort. Still, SGML laid the con­cep­tu­al foun­da­tion for every­thing that came after: HTML is an appli­ca­tion of SGML, and the phi­los­o­phy behind Mark­down traces back to the same think­ing. (Wikipedia)

1991: HTML (Hyper­Text Markup Lan­guage) — When Bern­ers-Lee pub­lished the first web­site, he used HTML, a heav­i­ly sim­pli­fied appli­ca­tion of SGML. HTML was delib­er­ate­ly prag­mat­ic: a hand­ful of tags, enough to struc­ture doc­u­ments and link them togeth­er, and that was enough to get the web mov­ing.

What made HTML so suc­cess­ful also became its prob­lem over time. Tags like <b> or <font> mixed con­tent and pre­sen­ta­tion, and as web­sites grew more com­plex, HTML source code became increas­ing­ly dif­fi­cult to read. For writ­ing notes or doc­u­ments it was sim­ply unsuit­able. Try­ing to write a sim­ple arti­cle meant fight­ing through a jun­gle of angle brack­ets. The gap between “human-writable” and “machine-ren­der­able” had nev­er been wider. (Wikipedia)

1998: XML (Exten­si­ble Markup Lan­guage) — XML was an attempt to pre­serve the best parts of SGML while reduc­ing the com­plex­i­ty. Unlike HTML, XML was not a fixed lan­guage with pre­de­fined tags, but a flex­i­ble sys­tem for defin­ing your own markup lan­guages. XML quick­ly became the stan­dard for data exchange between sys­tems and still under­pins many file for­mats today, includ­ing the Microsoft Word doc­u­ment for­mat and the EPUB for­mat for ebooks.

For human writ­ing, XML was just as unsuit­able as HTML, arguably more so, because the strict­ness of the for­mat left even less room for flex­i­bil­i­ty. But XML did rein­force one impor­tant idea: struc­tured, plain-text for­mats could be both machine-read­able and durable. That idea would mat­ter when Mark­down came along. (Wikipedia)

2004: Mark­down (John Gru­ber) — John Gru­ber released Mark­down in 2004 with a refresh­ing­ly clear goal: a markup lan­guage that is just as read­able as plain text as it is after being con­vert­ed to HTML. A head­ing marked with #looks like a head­ing even with­out ren­der­ing. Bold text with **double asterisks** is intu­itive­ly read­able. A list with minus signs is rec­og­niz­able as a list even with­out a pars­er.

That was a small rev­o­lu­tion, because Gru­ber reversed the usu­al direc­tion: instead of the per­son adapt­ing to the machine, the markup lan­guage adapts to the way peo­ple actu­al­ly write. Mark­down spread quick­ly through the devel­op­er com­mu­ni­ty, on plat­forms like GitHub and Red­dit, and even­tu­al­ly became the foun­da­tion for a whole gen­er­a­tion of writ­ing and knowl­edge man­age­ment tools.

One side effect that Gru­ber prob­a­bly did not ful­ly antic­i­pate: because Mark­down files are plain text, they are also future-proof. No pro­pri­etary for­mat, no ven­dor lock-in, no file that becomes unread­able when a com­pa­ny folds. That prop­er­ty turned out to mat­ter a great deal for tools like Obsid­i­an. (Dar­ing Fire­ball)

2020: Roam Research and the Back­link Moment — Before Obsid­i­an exist­ed, there was Roam Research. Launched in ear­ly 2020, Roam was the tool that brought bidi­rec­tion­al links into main­stream PKM con­ver­sa­tions. The idea itself was not new, Bush and Nel­son had described it decades ear­li­er, but Roam made it feel imme­di­ate and prac­ti­cal in a way that no pre­vi­ous tool had. Every note auto­mat­i­cal­ly appeared in the back­links pan­el of every note it linked to. The graph grew as you wrote. The com­mu­ni­ty around Roam, vocal on Twit­ter and in newslet­ters, turned net­worked note-tak­ing into a gen­uine cul­tur­al moment.

Obsid­i­an launched in March 2020, just months after Roam had ignit­ed the con­ver­sa­tion. It took the same core idea but made a dif­fer­ent bet: local files, Mark­down, no sub­scrip­tion, no cloud depen­den­cy. For users who want­ed the linked-think­ing mod­el but were uncom­fort­able hand­ing their notes to a web app, Obsid­i­an was the answer. With­out Roam’s proof of con­cept, it is unlike­ly the PKM land­scape of the ear­ly 2020s would have looked any­thing like it did. And with­out that land­scape, Obsid­i­an might have remained a niche tool for a very small audi­ence.

Obsid­i­an is per­haps the most con­sis­tent exam­ple of what hap­pens when you treat Mark­down not just as an out­put for­mat, but as the native for­mat for an entire knowl­edge sys­tem. The his­to­ry above shows that almost every idea it embod­ies had been described, attempt­ed, or par­tial­ly imple­ment­ed before. What Obsid­i­an did was assem­ble them in a way that was actu­al­ly usable, durable, and open enough to extend. That com­bi­na­tion turned out to be the thing that stuck.