Public dossier section · @ben

the idea

This is @ben's profile revision 1. The original date and editorial confidence below are attributed to its publisher, not independent verification.

Original credited author: Hraness

AI-drafted at Ben Guo's direct request and credited to Hraness; every claim links to its cataloged source.

Back to the full profile · Markdown version · Original document

Published text

roam research: the idea

the unfinished business — memex to zettelkasten to roam, why it stalled, and what agents change.

lineage

1945 — vannevar bush: the memex: 'the human mind does not work that way. it operates by association.' trails of linked material as the founding image of knowledge as a network, not a filing cabinet. (sources: as we may think)
1962–1968 — doug engelbart: augmenting human intellect as an explicit research program — computers as thinking partners. nls shipped hyperlinks, outlining, view control, and collaboration in the 1968 mother of all demos. (sources: augmenting human intellect: a conceptual framework)
1960–1995 — ted nelson: hypertext and transclusion — the same content living in many documents. roam's block reference is a descendant. also the field's cautionary tale: thirty years of xanadu as the canonical unfinished vision. (sources: computer lib / dream machines, the curse of xanadu)
1995 — ward cunningham: the wiki: every page editable, every camelcase word a link that creates the page if missing — bottom-up structure made nearly free. (sources: wikiwikiweb)
1950s–1981 — niklas luhmann: the slip-box as communication partner — 90,000 cards behind ~70 books. also the honest caveat roam lived out: a slip-box needs years to reach critical mass; until then it is a mere container. (sources: communicating with slip boxes)
1990s–2002 — mark bernstein: serious literary hypertext and the personal notes tool — tinderbox (2002) put maps and agents inside notes; 'hypertext gardens' (1998) is the first recorded use of 'digital garden.' (sources: hypertext gardens)
1980s–2010 — the outliner line: the zoomable bullet tree — thinktank and more to org-mode and workflowy — the surface grammar roam rebuilt as a product. (sources: about this particular outliner: outliner history, history and acknowledgments (org manual), turn your entire life into a list with workflowy)
2004 — jeremy ruston: the personal wiki as one self-contained file of reusable tiddlers — quasi-transclusion and non-linear notes a decade and a half early. (sources: the story of tiddlywiki)
2015 — mike caulfield: the garden-vs-stream frame: the web as topology against serialization — 'every walk through the garden creates new paths, new meanings.' (sources: the garden and the stream: a technopastoral)
2017 — sönke ahrens: the zettelkasten method translated for a mass audience — the book that primed the exact readership roam landed in. (sources: how to take smart notes)
2018 — michael nielsen: memory as a choice: the rigorous case that personal memory systems are real and buildable — the cognitive-science backbone of the tools-for-thought wave. (sources: augmenting long-term memory)
2019 — andy matuschak & michael nielsen: the field's manifesto — why the tech industry made little effort on tools for thought, and the warning that good ones arise mostly as byproducts of serious work. that warning doubles as a diagnosis of why note apps stall. (sources: how can we develop transformative tools for thought?, evergreen notes)
2018–2019 — tom critchlow: the build-your-own-garden stance — jekyll wikifolders and blogs without publish buttons — prefiguring the audience roam captured. (sources: of digital streams, campfires and gardens)
2019 — ink & switch: the counter-ideology roam violated: you own your data in spite of the cloud. obsidian and logseq recruited roam's defectors under this banner. (sources: local-first software)
2017–2022 — tiago forte: the demand engine: code and para turned personal knowledge management into a mass aspiration roam rode to its first million in arr. (sources: building a second brain)
2023 — steph ango: the epilogue-manifesto from obsidian's ceo: apps are ephemeral, files last. in the agent era it reads as prophetic — plain markdown is the substrate agents already read. (sources: file over app)

what roam proved

it shipped the lineage: roam shipped the lineage as one product: a daily-notes-first outliner where every block is an addressable node in a queryable graph — datomic underneath, datalog on top. 'notes as a database' was literal, not metaphorical. (sources: the roam research white paper, roam research — a note-taking tool for networked thought)
the primitives won: bidirectional links at page and block level went from research curiosity to default expectation — [[page]] creates the page if missing, ((block ref)) gives nelson-style transclusion. every successor now ships some version. (sources: obsidian — a private and flexible writing app, logseq/logseq, tana — the all-in-one workspace with supertags)
low-friction capture works: daily-notes-first capture made filing-free entry work at scale: open today's page, write, let links do the organizing later. even roam's sharpest critics concede this insight. (sources: why note-taking apps don't make us smarter, roam: why i love it and how i use it)
proof by adoption and imitation: roughly $1m arr within a year of public launch, a $9m seed at ~$200m, and a clone flood inside eighteen months — obsidian, logseq, athens, then tana, reflect, capacities, mem. 'roam-like' became a category name. (sources: a $200 million seed valuation for roam, athensresearch/athens, logseq/logseq)
a tool can be a movement: #roamcult was a genuine community phenomenon — courses, conferences, a plugin ecosystem, and a discourse the tool carried for years. the demand was real even if the retention wasn't. (sources: roam is a note-taking tool, and a dream of a better self, roam research: the limitless tool)

why it stalled

the manual-labor problem: links are cheap to create and expensive to use well. 'i waited for the insights to come. and waited. and waited.' — the structure-building was the work, and the work never paid off at the advertised rate for most users. (sources: why note-taking apps don't make us smarter)
the cargo-cult gap: the promised mechanism — graph, surprise, insight — requires luhmann-grade practice. the surrounding culture became a 'cargo cult of zettelkasten': users follow the motions for months, don't see the gains, and give up, sometimes in shame. (sources: a note on the cargo cult of zettelkasten, zettelkasten, honestly)
capture outran retrieval: roam optimized the cheap side. capture was frictionless; retrieval and synthesis still demanded queries, review discipline, and skill — so most graphs stayed shallow collections. 'a very organized warehouse.' (sources: why note-taking apps don't make us smarter, what happened to roam research? magic, then friction)
price, performance, and lock-in: $15 a month, $165 a year, $500 for the five-year believer plan, no free tier — against free obsidian and free, open logseq. sync bugs, slowdowns on large graphs, an afterthought mobile app, and a proprietary cloud that degraded on export. (sources: roam research faq, why i left roam research after 5 years, moving from roam research to obsidian, conaw on sync issues and the state of roam)
the feature moat drained: everything signature — backlinks, graph view, daily notes, block refs — was copied within eighteen months, often cheaper, local, or open. the white paper's deeper promises (bayesian weights, collaborative reasoning) never shipped. funding stopped at the 2020 seed. (sources: obsidian — a private and flexible writing app, athensresearch/athens, the roam research white paper)

the agent-era reopening

(for) llms collapse the exact labor that killed casual adoption — linking, summarizing, filing, review. karpathy's llm-wiki pattern: the model incrementally builds a persistent, interlinked markdown wiki; lint passes hunt orphan pages; the human moves up a level to sourcing and reviewing. 'the wiki is the codebase.' (sources: llm-wiki.md)
(for) compiled knowledge beats re-derived retrieval. plain rag rediscovers knowledge from scratch on every question; a maintained wiki accumulates — 'the cross-references are already there.' microsoft's graphrag operationalizes the same claim at corpus scale. (sources: llm-wiki.md, graphrag: unlocking llm discovery on narrative private data)
(for) agents need asserted structure, not just vibes: decision traces, typed relationships, provenance, and time-aware truth. the context-graphs thesis re-founds networked thought as enterprise infrastructure — 'the main bottleneck is context.' (sources: ai's trillion-dollar opportunity: context graphs, hraness/oh)
(for) at project scale, links beat embeddings outright: index files as routing tables, structured markdown underneath, and the agent follows links in two or three hops to grounded answers — no vector db required. (sources: look ma, no rag! building agent-native knowledge systems)
(for) files plus conventions are the substrate: agents.md as a readme for agents across codex, cursor, jules, and copilot; letta's git-backed memory projected as markdown into a real repo; mcp as the open socket. the industry independently reinvented roam's data model — plain text, explicit structure, versioned, inspectable. (sources: agents.md, introducing context repositories, introducing the model context protocol)
(against) the retrieval problem may simply be dead: semantic search finds the half-remembered idea across a million tokens without a single tag or link. 'the ai does not care about your backlinks. it reads text, scores relevance, and answers.' (sources: why your second brain is dead)
(against) a filesystem is empirically competitive: letta benchmarked a plain agent with grep and file tools at 74% on the locomio memory benchmark — matching specialized memory systems. if files plus grep suffice, hand-maintained graph structure may be ceremony. (sources: benchmarking ai agent memory: is a filesystem all you need?)
(against) machine-built links inherit machine confabulation: at ~100 articles a maintained wiki can exceed usable context and the model starts fabricating cross-note relationships. 'more context is not the same as better memory.' (sources: karpathy's obsidian wiki broke at 100 articles)
(against) memory behavior is harness-trained, not structure-determined: production agents converge on small markdown taxonomies, and 'remember this' instinct is shaped by post-training — implying the harness matters more than the representation. (sources: agent memory engineering)
(against) the stale-insight objection survives automation: newton's complaint was never labor cost — it was that links didn't produce insight when he did the labor. an agent-maintained graph might be a better map for the agent while leaving the human's empty discovery experience untouched. (sources: why note-taking apps don't make us smarter)
(nuance) the durable position this site stakes out: the agent era doesn't vindicate roam's ux for humans — it transposes the architecture to a new reader. the value proposition shifts from 'help me think' (unproven) to 'give the agent inspectable, reviewable, provenance-carrying context' (demonstrably needed). the unfinished idea isn't backlinks for humans; it's a shared, owned, structured memory layer between humans and agents. (sources: ai's trillion-dollar opportunity: context graphs, hraness/wordcell, hraness/oh)

successors

wordcell: a knowledge base for coding agents: markdown and wikilinks committed with the repo, backlinks and indexes as replaceable views, git as authority — the roam data model with agents as the primary readers and writers. (sources: hraness/wordcell, wordcell docs)
sponge: a knowledge harness for agentic research: save the corpus, bring your own agent to investigate, require cited passages and checked claims — 'agent proposes, human reviews' as the product primitive roam never had. (sources: sponge — a personal library for research)
oh: an ontology kernel for agentic research: content-addressed entity, statement, assertion, and evidence records with an append-only operation log — structured claims as the agent's inspectable memory format. (sources: hraness/oh)
tana: supertags gave nodes schema, then the product rebuilt around ai: 'your graph is the context and conversations are your prompts.' (sources: tana — the all-in-one workspace with supertags, tana raises $14m series a, the next chapter for tana)
reflect: fast encrypted networked notes plus the literal labor handoff: 'decorate my writing with backlinks' as a command. (sources: automatically add backlinks using ai)
obsidian: the defector's destination and the agent era's de-facto vault format — local markdown, wikilinks, file over app. the doctrine is why agents can already read it. (sources: obsidian — a private and flexible writing app, file over app)
logseq: the free, open-source roam workalike storing local markdown/org files — readme credits roam, org-mode, tiddlywiki, and workflowy. (sources: logseq/logseq, logseq raises $4.1m)
athens: the cautionary open-source clone — yc w21, 6.3k stars, unmaintained since ~2023: proof that cloning the graph was never the hard part. (sources: athensresearch/athens)
letta / agents.md: the adjacent infrastructure carrying the idea: git-backed markdown memory the agent greps and commits, plus a cross-vendor readme-for-agents convention. (sources: introducing context repositories, agents.md)

open questions

if an agent maintains the graph, does the owner still get the insight — or only the agent? is the deliverable discovery, or delegation?
is the graph the product or the byproduct? wordcell treats indexes as replaceable views over authoritative markdown and git; graphrag rebuilds the graph per corpus. is any particular graph worth owning, or is the convention the asset?
who reviews the links an agent creates — a diff, a link, a claim? does review scale, or does it become the new manual labor?
explicit links versus learned structure: letta's filesystem benchmark says files plus grep rival memory systems; context graphs and graphrag say asserted structure answers what vectors can't. is the answer a two-layer split — asserted provenance plus learned association?
whose memory is it? roam's lock-in was proprietary; agent-era memory files are written by the agent and trained per-harness. is portable agent memory possible, or does each vendor's format recreate the lock-in with worse consequences?
does the thinking tool survive delegation? matuschak and nielsen warned good tools for thought arise as byproducts of serious work — if agents do the work, do tools for thought collapse into tools for agents, with humans reading the graphs?

AI-drafted at Ben Guo's direct request and credited to Hraness; every claim links to its cataloged source.

Canonical page: https://hraness.com/roam-research/the-idea

Original source occurrences

roam research — a note-taking tool for networked thought · roam research · 2019 (year) · original ID: src-roam-site · roam research
the product site and its own description of the tool.
the roam research white paper · roam research · ~2020 (approximate) · original ID: src-roam-whitepaper · roam research
roam's own statement of the thesis: a directed graph instead of a file tree, with bayesian and collaborative ambitions that mostly never shipped. the page lives inside roam's js-rendered help graph; a readable export sits at heckler.roam.garden/@white-paper.
roam: why i love it and how i use it · nat eliason · 2020-01-18 (day) · original ID: src-eliason-why-i-love-it · nat eliason
the article most often credited with triggering the 2020 sign-up wave; eliason later reported it drove meaningful referral revenue.
roam research faq · ness labs · 2020 (year) · original ID: src-nesslabs-roam-faq · anne-laure le cunff
the pricing record: $15/month, $165/year, and the $500 five-year believer plan.
roam research: the limitless tool · the generalist · 2020 (year) · original ID: src-generalist-roam · mario gabriele
a long, favorable company profile published near the peak of the hype wave.
roam is a note-taking tool, and a dream of a better self · built in sf · 2020 (year) · original ID: src-builtin-roamcult
the feature that documented #roamcult as a genuine community phenomenon and traced the bush-to-roam lineage in the popular press.
a $200 million seed valuation for roam · the information · 2020-09-11 (day) · original ID: src-information-200m
the scoop that fixed the public number: ~$9m at ~$200m, roughly 25x the median seed valuation. paywalled.
why note-taking apps don't make us smarter · platformer · 2023-09-11 (day) · original ID: src-newton-platformer · casey newton
the marquee postmortem by a believer: 'i waited for the insights to come. and waited. and waited.' notes designed for storage, not sparking insights.
a note on the cargo cult of zettelkasten · boffosocko · 2023-02-03 (day) · original ID: src-aldrich-cargo-cult · chris aldrich
the cargo-cult diagnosis: users follow the motions of zettelkasten for months, don't see the promised gains, and quietly give up.
why i left roam research after 5 years · yu-wenhao.com · ~2025 (approximate) · original ID: src-yu-why-i-left · wenhao yu
a five-year believer ($500 plan, $1k investor) leaves for obsidian: 'ai changed everything, and my data was locked in the cloud.'
zettelkasten, honestly · notesnudge · ~2025 (approximate) · original ID: src-notesnudge-zettelkasten
the cleanest evidence audit: 'the system will surprise you with connections' is the least-supported claim; only survivors write essays.
moving from roam research to obsidian · mickmel (substack) · ~2022 (approximate) · original ID: src-mickmel-migration · mickmel
a representative migration note: price, privacy, and sync driving the move to obsidian.
conaw on sync issues and the state of roam · threadreader app · 2021-01 (month) · original ID: src-conaw-sync-thread · conor white-sullivan
conor's own january 2021 thread on sync bugs and performance — 'roam is far from done' — plus the subreddit friction it answered. captured via threadreader.
as we may think · the atlantic · 1945-07 (month) · original ID: src-as-we-may-think · vannevar bush
the 1945 memex essay every networked-thought tool cites as origin.
augmenting human intellect: a conceptual framework · doug engelbart institute · 1962-10 (month) · original ID: src-engelbart-augment · douglas engelbart
augmenting human intellect as an explicit program — computers as thinking partners, six years before the mother of all demos.
computer lib / dream machines · internet archive (scan) · 1974 (year) · original ID: src-computer-lib · ted nelson
hypertext, hypermedia, and 'everything is deeply intertwingled' — plus transclusion, the ancestor of roam's block reference.
the curse of xanadu · wired · 1995-06 (month) · original ID: src-curse-of-xanadu · gary wolf
the cautionary precedent: xanadu as the longest-running vaporware project in computing — vision shipped partially, ambition unreached.
wikiwikiweb · wikipedia · 1995 (year) · original ID: src-wikiwikiweb
ward cunningham's 1995 wiki: every page editable, every camelcase word a link that creates the page if missing — bottom-up structure made nearly free.
communicating with slip boxes · luhmann.surge.sh (translation) · 1981 (year) · original ID: src-luhmann-slip-boxes · niklas luhmann
the slip-box as a partner that surprises you — and the caveat roam lived out: the box needs years to reach critical mass, until then it is a mere container.
how to take smart notes · northbooks · 2017 (year) · original ID: src-smart-notes · sönke ahrens
the book that popularized luhmann's zettelkasten for the roam generation.
history and acknowledgments (org manual) · orgmode.org · 2003 (year) · original ID: src-org-history
org-mode's own history: a plain-text outline mode born in 2003 — the grammar roam rebuilt as a product.
turn your entire life into a list with workflowy · techcrunch · 2010-11-04 (day) · original ID: src-workflowy-launch · alexia tsotsis
workflowy's 2010 launch: proof a whole product could be one infinite, zoomable list.
about this particular outliner: outliner history · atpm · 2003 (year) · original ID: src-outliner-history · ted goranson
the desktop-outliner lineage — thinktank, more, acta — that roam's surface descends from.
hypertext gardens · eastgate · 1998 (year) · original ID: src-hypertext-gardens · mark bernstein
the first recorded use of 'digital garden' (1998) and a deliberate argument for irregular, curated link structure; tinderbox put agents inside notes in 2002.
the story of tiddlywiki · tiddlywiki.com · 2004 (year) · original ID: src-tiddlywiki-story · jeremy ruston
the personal wiki as one self-contained html file of reusable tiddlers — quasi-transclusion in 2004.
the garden and the stream: a technopastoral · hapgood · 2015-10-17 (day) · original ID: src-garden-and-stream · mike caulfield
the garden-vs-stream frame that became digital-garden canon: topology against serialization.
augmenting long-term memory · michael nielsen · 2018-07 (month) · original ID: src-augmenting-memory · michael nielsen
spaced repetition and memory systems as tools for thought — the intellectual neighbor roam sat next to.
how can we develop transformative tools for thought? · numinous productions · 2019 (year) · original ID: src-transformative-tft · andy matuschak & michael nielsen
the essay that named the field roam came to represent — and warned that good tools for thought arise mostly as byproducts of serious work.
evergreen notes · notes.andymatuschak.org · 2019 (year) · original ID: src-evergreen-notes · andy matuschak
the canonical public note corpus: 'better note-taking misses the point; what matters is better thinking'; titles as apis; notes densely linked.
of digital streams, campfires and gardens · tomcritchlow.com · 2018-10 (month) · original ID: src-critchlow-gardens · tom critchlow
the indie-hacker instantiation: streams, campfires, gardens, and the jekyll wikifolder — building your own garden before roam sold one.
local-first software · ink & switch · 2019-04 (month) · original ID: src-local-first · kleppmann, wiggins, van hardenberg, mcgranaghan
the counter-ideology roam violated: you own your data, in spite of the cloud — the banner obsidian and logseq recruited under.
building a second brain · building a second brain · 2017 (year) · original ID: src-building-second-brain · tiago forte
code and para turned pkm into a mass aspiration — the demand roam rode. eliason's verdict: 'roam is v4.'
file over app · stephango.com · 2023-07-01 (day) · original ID: src-file-over-app · steph ango
obsidian's ceo on why files outlive apps — in the agent era it reads as prophetic: plain markdown is the substrate agents already read.
obsidian — a private and flexible writing app · obsidian · 2020 (year) · original ID: src-obsidian
the local-first competitor that won the pragmatists while roam won the evangelists — first beta march 30, 2020, free, plain markdown.
logseq/logseq · github · 2020 (year) · original ID: src-logseq
the open-source outliner clone (agpl, local markdown/org files) that absorbed roam's most cost-sensitive users; its readme credits roam, org-mode, tiddlywiki, and workflowy.
logseq raises $4.1m · logseq blog · ~2022 (approximate) · original ID: src-logseq-raise
logseq's $4.1m raise from collison, friedman, lütke, and dave winer — explicitly on users 'jumping ship from roam.'
athensresearch/athens · github · 2020 (year) · original ID: src-athens
the cautionary open-source clone: yc w21, 6.3k stars, unmaintained since ~2023 — cloning the graph was never the hard part.
tana — the all-in-one workspace with supertags · tana · 2023 (year) · original ID: src-tana
the supertag successor built by ex-roam-community figures, launched on the explicit promise of finishing what roam started.
tana raises $14m series a · tana blog · ~2024 (approximate) · original ID: src-tana-raise
tana's $14m series a as an 'ai-native workspace' — the category's pivot to agents made literal.
the next chapter for tana · tana blog · ~2025 (approximate) · original ID: src-tana-next-chapter
'your graph is the context and conversations are your prompts… your knowledge graph builds itself every time you talk.'
automatically add backlinks using ai · reflect · ~2023 (approximate) · original ID: src-reflect-backlinks · alex maccaw
the labor handoff as a button: 'decorate my writing with backlinks.'
llm-wiki.md · github gist · 2026-04-04 (day) · original ID: src-karpathy-llm-wiki · andrej karpathy
the agent-era restatement: an llm-maintained markdown wiki as 'a persistent, compounding artifact.' 'obsidian is the ide; the llm is the programmer; the wiki is the codebase.'
graphrag: unlocking llm discovery on narrative private data · microsoft research · 2024-02-13 (day) · original ID: src-graphrag · larson & truitt
microsoft's graphrag: llm-extracted knowledge graphs plus community summaries beat baseline rag on global sensemaking questions.
ai's trillion-dollar opportunity: context graphs · foundation capital · 2025-12 (month) · original ID: src-context-graphs · jaya gupta & ashu garg
the enterprise restatement: agents need decision traces — exceptions, overrides, precedents — captured into context graphs.
introducing the model context protocol · anthropic · 2024-11-25 (day) · original ID: src-mcp
anthropic's open socket between agents and data — the plumbing that makes agent-readable knowledge bases practical.
look ma, no rag! building agent-native knowledge systems · blog.pollack.ai · ~2025 (approximate) · original ID: src-no-rag
the links-beat-embeddings argument at project scale: index files as routing tables, let the agent follow links in two or three hops.
benchmarking ai agent memory: is a filesystem all you need? · letta · ~2025 (approximate) · original ID: src-letta-benchmark
the strongest counter-evidence: a plain agent with file tools hit 74% on the locomio memory benchmark — files plus grep rival specialized memory systems.
introducing context repositories · letta · ~2025 (approximate) · original ID: src-letta-context-repos
git-backed agent memory: markdown plus frontmatter projected into a real repo the agent greps and commits — roam's data model reinvented as infrastructure.
agents.md · agents.md · 2025 (year) · original ID: src-agents-md
the cross-vendor convention — 'a readme for agents' across codex, cursor, jules, copilot, and 60k+ projects.
agent memory engineering · nicolasbustamante.com · ~2026 (approximate) · original ID: src-agent-memory-engineering · nicolas bustamante
the audit caveat: production agent memory converges on small markdown taxonomies, and 'remember this' behavior is harness-trained — structure matters less than habit.
karpathy's obsidian wiki broke at 100 articles · dev.to · ~2026 (approximate) · original ID: src-llm-wiki-broke · zaferdace
the counter-lesson: at ~100 articles an llm-maintained wiki exceeds usable context and the model starts confidently fabricating cross-note relationships.
why your second brain is dead · self.md · ~2025 (approximate) · original ID: src-second-brain-dead
the bluntest version of the against case: 'the ai does not care about your backlinks… the expensive operation became free.'
hraness/wordcell · github · 2026 (year) · original ID: src-wordcell · hraness
a knowledge base for coding agents built from markdown, backlinks, semantic search, and git context — markdown and git authoritative, indexes as replaceable views.
wordcell docs · wordcell.io · 2026 (year) · original ID: src-wordcell-docs · hraness
wordcell's docs: context, backlinks, and history per code path, installed as an agent skill.
sponge — a personal library for research · sponge.computer · 2025 (year) · original ID: src-sponge · hraness
a knowledge harness where agents research against a saved corpus — cite the passages used, check claims against the text — while people steer, review, and control release.
hraness/oh · github · 2026 (year) · original ID: src-oh · hraness
an ontology kernel for agentic research: content-addressed entity, statement, assertion, and evidence records with an append-only operation log; indexes derived and rebuildable.
what happened to roam research? magic, then friction · buildfirstbrain · ~2026 (approximate) · original ID: src-what-happened-roam
a representative 'what happened to roam' essay — likely ai-assisted, used for the genre's shape rather than its claims: 'connections you do not understand are not understanding.'

Structured research records

llms collapse the exact labor that killed casual adoption — linking, summarizing, filing, review. karpathy's llm-wiki pattern: the model incrementally builds a persistent, interlinked markdown wiki; lint passes hunt orphan pages; the human moves up a level to sourcing and reviewing. 'the wiki is the codebase.' · agent-era-arguments

Original location · source-ebe31e9964193545c49d

Original structured fields

{
  "claim": "llms collapse the exact labor that killed casual adoption — linking, summarizing, filing, review. karpathy's llm-wiki pattern: the model incrementally builds a persistent, interlinked markdown wiki; lint passes hunt orphan pages; the human moves up a level to sourcing and reviewing. 'the wiki is the codebase.'",
  "id": "arg-labor",
  "side": "for",
  "sourceIds": [
    "src-karpathy-llm-wiki"
  ]
}

compiled knowledge beats re-derived retrieval. plain rag rediscovers knowledge from scratch on every question; a maintained wiki accumulates — 'the cross-references are already there.' microsoft's graphrag operationalizes the same claim at corpus scale. · agent-era-arguments

Original location · source-ebe31e9964193545c49d · source-c8a880fa1eb9bbb177d9

Original structured fields

{
  "claim": "compiled knowledge beats re-derived retrieval. plain rag rediscovers knowledge from scratch on every question; a maintained wiki accumulates — 'the cross-references are already there.' microsoft's graphrag operationalizes the same claim at corpus scale.",
  "id": "arg-compounding",
  "side": "for",
  "sourceIds": [
    "src-karpathy-llm-wiki",
    "src-graphrag"
  ]
}

agents need asserted structure, not just vibes: decision traces, typed relationships, provenance, and time-aware truth. the context-graphs thesis re-founds networked thought as enterprise infrastructure — 'the main bottleneck is context.' · agent-era-arguments
Original location · source-f115b24131354e28c3a6 · source-1ca961629a2fc81119fd
Original structured fields
```
{
  "claim": "agents need asserted structure, not just vibes: decision traces, typed relationships, provenance, and time-aware truth. the context-graphs thesis re-founds networked thought as enterprise infrastructure — 'the main bottleneck is context.'",
  "id": "arg-structure",
  "side": "for",
  "sourceIds": [
    "src-context-graphs",
    "src-oh"
  ]
}
```

at project scale, links beat embeddings outright: index files as routing tables, structured markdown underneath, and the agent follows links in two or three hops to grounded answers — no vector db required. · agent-era-arguments

Original location · source-d063cd82be41582e70fc

Original structured fields

{
  "claim": "at project scale, links beat embeddings outright: index files as routing tables, structured markdown underneath, and the agent follows links in two or three hops to grounded answers — no vector db required.",
  "id": "arg-links-beat-embeddings",
  "side": "for",
  "sourceIds": [
    "src-no-rag"
  ]
}

files plus conventions are the substrate: agents.md as a readme for agents across codex, cursor, jules, and copilot; letta's git-backed memory projected as markdown into a real repo; mcp as the open socket. the industry independently reinvented roam's data model — plain text, explicit structure, versioned, inspectable. · agent-era-arguments
Original location · source-d4931dfaad1e407f922a · source-0a351a290030e1f4d7a8 · source-617acd4ac8d72c1eff15
Original structured fields
```
{
  "claim": "files plus conventions are the substrate: agents.md as a readme for agents across codex, cursor, jules, and copilot; letta's git-backed memory projected as markdown into a real repo; mcp as the open socket. the industry independently reinvented roam's data model — plain text, explicit structure, versioned, inspectable.",
  "id": "arg-substrate",
  "side": "for",
  "sourceIds": [
    "src-agents-md",
    "src-letta-context-repos",
    "src-mcp"
  ]
}
```

the retrieval problem may simply be dead: semantic search finds the half-remembered idea across a million tokens without a single tag or link. 'the ai does not care about your backlinks. it reads text, scores relevance, and answers.' · agent-era-arguments

Original location · source-e43054229dc7b1a23cc5

Original structured fields

{
  "claim": "the retrieval problem may simply be dead: semantic search finds the half-remembered idea across a million tokens without a single tag or link. 'the ai does not care about your backlinks. it reads text, scores relevance, and answers.'",
  "id": "arg-retrieval-dead",
  "side": "against",
  "sourceIds": [
    "src-second-brain-dead"
  ]
}

a filesystem is empirically competitive: letta benchmarked a plain agent with grep and file tools at 74% on the locomio memory benchmark — matching specialized memory systems. if files plus grep suffice, hand-maintained graph structure may be ceremony. · agent-era-arguments

Original location · source-d533a7d02fbb317ead16

Original structured fields

{
  "claim": "a filesystem is empirically competitive: letta benchmarked a plain agent with grep and file tools at 74% on the locomio memory benchmark — matching specialized memory systems. if files plus grep suffice, hand-maintained graph structure may be ceremony.",
  "id": "arg-filesystem",
  "side": "against",
  "sourceIds": [
    "src-letta-benchmark"
  ]
}

machine-built links inherit machine confabulation: at ~100 articles a maintained wiki can exceed usable context and the model starts fabricating cross-note relationships. 'more context is not the same as better memory.' · agent-era-arguments

Original location · source-9503dca74ecd52885c51

Original structured fields

{
  "claim": "machine-built links inherit machine confabulation: at ~100 articles a maintained wiki can exceed usable context and the model starts fabricating cross-note relationships. 'more context is not the same as better memory.'",
  "id": "arg-confabulation",
  "side": "against",
  "sourceIds": [
    "src-llm-wiki-broke"
  ]
}

memory behavior is harness-trained, not structure-determined: production agents converge on small markdown taxonomies, and 'remember this' instinct is shaped by post-training — implying the harness matters more than the representation. · agent-era-arguments

Original location · source-3b9eef51ab8cb9b55766

Original structured fields

{
  "claim": "memory behavior is harness-trained, not structure-determined: production agents converge on small markdown taxonomies, and 'remember this' instinct is shaped by post-training — implying the harness matters more than the representation.",
  "id": "arg-harness",
  "side": "against",
  "sourceIds": [
    "src-agent-memory-engineering"
  ]
}

the stale-insight objection survives automation: newton's complaint was never labor cost — it was that links didn't produce insight when he did the labor. an agent-maintained graph might be a better map for the agent while leaving the human's empty discovery experience untouched. · agent-era-arguments

Original location · source-d87fdbaf0346bc71f4e5

Original structured fields

{
  "claim": "the stale-insight objection survives automation: newton's complaint was never labor cost — it was that links didn't produce insight when he did the labor. an agent-maintained graph might be a better map for the agent while leaving the human's empty discovery experience untouched.",
  "id": "arg-insight",
  "side": "against",
  "sourceIds": [
    "src-newton-platformer"
  ]
}

the durable position this site stakes out: the agent era doesn't vindicate roam's ux for humans — it transposes the architecture to a new reader. the value proposition shifts from 'help me think' (unproven) to 'give the agent inspectable, reviewable, provenance-carrying context' (demonstrably needed). the unfinished idea isn't backlinks for humans; it's a shared, owned, structured memory layer between humans and agents. · agent-era-arguments
Original location · source-f115b24131354e28c3a6 · source-f2c79886b553d0269039 · source-1ca961629a2fc81119fd
Original structured fields
```
{
  "claim": "the durable position this site stakes out: the agent era doesn't vindicate roam's ux for humans — it transposes the architecture to a new reader. the value proposition shifts from 'help me think' (unproven) to 'give the agent inspectable, reviewable, provenance-carrying context' (demonstrably needed). the unfinished idea isn't backlinks for humans; it's a shared, owned, structured memory layer between humans and agents.",
  "id": "arg-transpose",
  "side": "nuance",
  "sourceIds": [
    "src-context-graphs",
    "src-wordcell",
    "src-oh"
  ]
}
```

vannevar bush · idea-lineage

Original location · source-45d6846fe066c5f0509b

Original structured fields

{
  "contribution": "the memex: 'the human mind does not work that way. it operates by association.' trails of linked material as the founding image of knowledge as a network, not a filing cabinet.",
  "era": "1945",
  "id": "lin-memex",
  "name": "vannevar bush",
  "sourceIds": [
    "src-as-we-may-think"
  ]
}

doug engelbart · idea-lineage

Original location · source-b90e7f9a32d4ff5895f0

Original structured fields

{
  "contribution": "augmenting human intellect as an explicit research program — computers as thinking partners. nls shipped hyperlinks, outlining, view control, and collaboration in the 1968 mother of all demos.",
  "era": "1962–1968",
  "id": "lin-engelbart",
  "name": "doug engelbart",
  "sourceIds": [
    "src-engelbart-augment"
  ]
}

ted nelson · idea-lineage

Original location · source-04e3e64f7b2736fe106e · source-708b4a88b48b6b533cc3

Original structured fields

{
  "contribution": "hypertext and transclusion — the same content living in many documents. roam's block reference is a descendant. also the field's cautionary tale: thirty years of xanadu as the canonical unfinished vision.",
  "era": "1960–1995",
  "id": "lin-nelson",
  "name": "ted nelson",
  "sourceIds": [
    "src-computer-lib",
    "src-curse-of-xanadu"
  ]
}

ward cunningham · idea-lineage

Original location · source-a024316ffd8a24a7ffa9

Original structured fields

{
  "contribution": "the wiki: every page editable, every camelcase word a link that creates the page if missing — bottom-up structure made nearly free.",
  "era": "1995",
  "id": "lin-wiki",
  "name": "ward cunningham",
  "sourceIds": [
    "src-wikiwikiweb"
  ]
}

niklas luhmann · idea-lineage

Original location · source-04705a1cf632f4a9d889

Original structured fields

{
  "contribution": "the slip-box as communication partner — 90,000 cards behind ~70 books. also the honest caveat roam lived out: a slip-box needs years to reach critical mass; until then it is a mere container.",
  "era": "1950s–1981",
  "id": "lin-luhmann",
  "name": "niklas luhmann",
  "sourceIds": [
    "src-luhmann-slip-boxes"
  ]
}

mark bernstein · idea-lineage

Original location · source-b839bce2051cdb142c08

Original structured fields

{
  "contribution": "serious literary hypertext and the personal notes tool — tinderbox (2002) put maps and agents inside notes; 'hypertext gardens' (1998) is the first recorded use of 'digital garden.'",
  "era": "1990s–2002",
  "id": "lin-bernstein",
  "name": "mark bernstein",
  "sourceIds": [
    "src-hypertext-gardens"
  ]
}

the outliner line · idea-lineage

Original location · source-7806a8742f0f53f1b7de · source-c50c200c5fb3be948c82 · source-635f8d788fce2784f71f

Original structured fields

{
  "contribution": "the zoomable bullet tree — thinktank and more to org-mode and workflowy — the surface grammar roam rebuilt as a product.",
  "era": "1980s–2010",
  "id": "lin-outliners",
  "name": "the outliner line",
  "sourceIds": [
    "src-outliner-history",
    "src-org-history",
    "src-workflowy-launch"
  ]
}

jeremy ruston · idea-lineage

Original location · source-a4f807c58c547db1b05f

Original structured fields

{
  "contribution": "the personal wiki as one self-contained file of reusable tiddlers — quasi-transclusion and non-linear notes a decade and a half early.",
  "era": "2004",
  "id": "lin-tiddlywiki",
  "name": "jeremy ruston",
  "sourceIds": [
    "src-tiddlywiki-story"
  ]
}

mike caulfield · idea-lineage

Original location · source-776433847a838f1e4ca0

Original structured fields

{
  "contribution": "the garden-vs-stream frame: the web as topology against serialization — 'every walk through the garden creates new paths, new meanings.'",
  "era": "2015",
  "id": "lin-caulfield",
  "name": "mike caulfield",
  "sourceIds": [
    "src-garden-and-stream"
  ]
}

sönke ahrens · idea-lineage

Original location · source-257fd8d62d252b1b3aa9

Original structured fields

{
  "contribution": "the zettelkasten method translated for a mass audience — the book that primed the exact readership roam landed in.",
  "era": "2017",
  "id": "lin-smart-notes",
  "name": "sönke ahrens",
  "sourceIds": [
    "src-smart-notes"
  ]
}

michael nielsen · idea-lineage

Original location · source-82891e8d2b461584f30f

Original structured fields

{
  "contribution": "memory as a choice: the rigorous case that personal memory systems are real and buildable — the cognitive-science backbone of the tools-for-thought wave.",
  "era": "2018",
  "id": "lin-ltm",
  "name": "michael nielsen",
  "sourceIds": [
    "src-augmenting-memory"
  ]
}

andy matuschak & michael nielsen · idea-lineage

Original location · source-d48b38c91fa9873411c7 · source-ae441fe63b14a342cfe7

Original structured fields

{
  "contribution": "the field's manifesto — why the tech industry made little effort on tools for thought, and the warning that good ones arise mostly as byproducts of serious work. that warning doubles as a diagnosis of why note apps stall.",
  "era": "2019",
  "id": "lin-ttft",
  "name": "andy matuschak & michael nielsen",
  "sourceIds": [
    "src-transformative-tft",
    "src-evergreen-notes"
  ]
}

tom critchlow · idea-lineage

Original location · source-8037029da5d580fbf303

Original structured fields

{
  "contribution": "the build-your-own-garden stance — jekyll wikifolders and blogs without publish buttons — prefiguring the audience roam captured.",
  "era": "2018–2019",
  "id": "lin-critchlow",
  "name": "tom critchlow",
  "sourceIds": [
    "src-critchlow-gardens"
  ]
}

ink & switch · idea-lineage

Original location · source-75dfe6e1b48a22f623ef

Original structured fields

{
  "contribution": "the counter-ideology roam violated: you own your data in spite of the cloud. obsidian and logseq recruited roam's defectors under this banner.",
  "era": "2019",
  "id": "lin-local-first",
  "name": "ink & switch",
  "sourceIds": [
    "src-local-first"
  ]
}

tiago forte · idea-lineage

Original location · source-15c650ab19ba5c01eb17

Original structured fields

{
  "contribution": "the demand engine: code and para turned personal knowledge management into a mass aspiration roam rode to its first million in arr.",
  "era": "2017–2022",
  "id": "lin-basb",
  "name": "tiago forte",
  "sourceIds": [
    "src-building-second-brain"
  ]
}

steph ango · idea-lineage

Original location · source-b8b946a41bad6831c413

Original structured fields

{
  "contribution": "the epilogue-manifesto from obsidian's ceo: apps are ephemeral, files last. in the agent era it reads as prophetic — plain markdown is the substrate agents already read.",
  "era": "2023",
  "id": "lin-file-over-app",
  "name": "steph ango",
  "sourceIds": [
    "src-file-over-app"
  ]
}

if an agent maintains the graph, does the owner still get the insight — or only the agent? is the deliverable discovery, or delegation? · idea-open-questions
Original location
Original structured fields
```
{
  "text": "if an agent maintains the graph, does the owner still get the insight — or only the agent? is the deliverable discovery, or delegation?"
}
```
is the graph the product or the byproduct? wordcell treats indexes as replaceable views over authoritative markdown and git; graphrag rebuilds the graph per corpus. is any particular graph worth owning, or is the convention the asset? · idea-open-questions
Original location
Original structured fields
```
{
  "text": "is the graph the product or the byproduct? wordcell treats indexes as replaceable views over authoritative markdown and git; graphrag rebuilds the graph per corpus. is any particular graph worth owning, or is the convention the asset?"
}
```
who reviews the links an agent creates — a diff, a link, a claim? does review scale, or does it become the new manual labor? · idea-open-questions
Original location
Original structured fields
```
{
  "text": "who reviews the links an agent creates — a diff, a link, a claim? does review scale, or does it become the new manual labor?"
}
```
explicit links versus learned structure: letta's filesystem benchmark says files plus grep rival memory systems; context graphs and graphrag say asserted structure answers what vectors can't. is the answer a two-layer split — asserted provenance plus learned association? · idea-open-questions
Original location
Original structured fields
```
{
  "text": "explicit links versus learned structure: letta's filesystem benchmark says files plus grep rival memory systems; context graphs and graphrag say asserted structure answers what vectors can't. is the answer a two-layer split — asserted provenance plus learned association?"
}
```
whose memory is it? roam's lock-in was proprietary; agent-era memory files are written by the agent and trained per-harness. is portable agent memory possible, or does each vendor's format recreate the lock-in with worse consequences? · idea-open-questions
Original location
Original structured fields
```
{
  "text": "whose memory is it? roam's lock-in was proprietary; agent-era memory files are written by the agent and trained per-harness. is portable agent memory possible, or does each vendor's format recreate the lock-in with worse consequences?"
}
```
does the thinking tool survive delegation? matuschak and nielsen warned good tools for thought arise as byproducts of serious work — if agents do the work, do tools for thought collapse into tools for agents, with humans reading the graphs? · idea-open-questions
Original location
Original structured fields
```
{
  "text": "does the thinking tool survive delegation? matuschak and nielsen warned good tools for thought arise as byproducts of serious work — if agents do the work, do tools for thought collapse into tools for agents, with humans reading the graphs?"
}
```

it shipped the lineage · idea-proved

Original location · source-1afd8938472ec4d9c672 · source-fe85ea12502846d392c0

Original structured fields

{
  "detail": "roam shipped the lineage as one product: a daily-notes-first outliner where every block is an addressable node in a queryable graph — datomic underneath, datalog on top. 'notes as a database' was literal, not metaphorical.",
  "id": "clm-shipped-lineage",
  "sourceIds": [
    "src-roam-whitepaper",
    "src-roam-site"
  ],
  "title": "it shipped the lineage"
}

the primitives won · idea-proved

Original location · source-c93fbd25fcb5484c2f3d · source-84997723082acd9a105f · source-b40545bca43c37fdd757

Original structured fields

{
  "detail": "bidirectional links at page and block level went from research curiosity to default expectation — `[[page]]` creates the page if missing, `((block ref))` gives nelson-style transclusion. every successor now ships some version.",
  "id": "clm-links-mainstream",
  "sourceIds": [
    "src-obsidian",
    "src-logseq",
    "src-tana"
  ],
  "title": "the primitives won"
}

low-friction capture works · idea-proved

Original location · source-d87fdbaf0346bc71f4e5 · source-012efcf9d02963d3277b

Original structured fields

{
  "detail": "daily-notes-first capture made filing-free entry work at scale: open today's page, write, let links do the organizing later. even roam's sharpest critics concede this insight.",
  "id": "clm-daily-notes",
  "sourceIds": [
    "src-newton-platformer",
    "src-eliason-why-i-love-it"
  ],
  "title": "low-friction capture works"
}

proof by adoption and imitation · idea-proved

Original location · source-b928a8dd5673a54965ef · source-aad8ffea41cfaab5f947 · source-84997723082acd9a105f

Original structured fields

{
  "detail": "roughly $1m arr within a year of public launch, a $9m seed at ~$200m, and a clone flood inside eighteen months — obsidian, logseq, athens, then tana, reflect, capacities, mem. 'roam-like' became a category name.",
  "id": "clm-adoption-imitation",
  "sourceIds": [
    "src-information-200m",
    "src-athens",
    "src-logseq"
  ],
  "title": "proof by adoption and imitation"
}

a tool can be a movement · idea-proved

Original location · source-c5e347ec5899a4da4625 · source-6a6b332d63c6a1bffc52

Original structured fields

{
  "detail": "#roamcult was a genuine community phenomenon — courses, conferences, a plugin ecosystem, and a discourse the tool carried for years. the demand was real even if the retention wasn't.",
  "id": "clm-community",
  "sourceIds": [
    "src-builtin-roamcult",
    "src-generalist-roam"
  ],
  "title": "a tool can be a movement"
}

the manual-labor problem · idea-stalled

Original location · source-d87fdbaf0346bc71f4e5

Original structured fields

{
  "detail": "links are cheap to create and expensive to use well. 'i waited for the insights to come. and waited. and waited.' — the structure-building was the work, and the work never paid off at the advertised rate for most users.",
  "id": "clm-manual-labor",
  "sourceIds": [
    "src-newton-platformer"
  ],
  "title": "the manual-labor problem"
}

the cargo-cult gap · idea-stalled

Original location · source-6e06bc266e6daa49d8c5 · source-3f2d4dd838809ad8561d

Original structured fields

{
  "detail": "the promised mechanism — graph, surprise, insight — requires luhmann-grade practice. the surrounding culture became a 'cargo cult of zettelkasten': users follow the motions for months, don't see the gains, and give up, sometimes in shame.",
  "id": "clm-cargo-cult",
  "sourceIds": [
    "src-aldrich-cargo-cult",
    "src-notesnudge-zettelkasten"
  ],
  "title": "the cargo-cult gap"
}

capture outran retrieval · idea-stalled

Original location · source-d87fdbaf0346bc71f4e5 · source-b497c2d2497a3f7594e1

Original structured fields

{
  "detail": "roam optimized the cheap side. capture was frictionless; retrieval and synthesis still demanded queries, review discipline, and skill — so most graphs stayed shallow collections. 'a very organized warehouse.'",
  "id": "clm-capture-retrieval",
  "sourceIds": [
    "src-newton-platformer",
    "src-what-happened-roam"
  ],
  "title": "capture outran retrieval"
}

price, performance, and lock-in · idea-stalled

Original location · source-0c06f35b5641ae07d537 · source-95fb0c07952ed758f86d · source-ae5b5e381bf54f41c176 · source-77a876269b4eee3c12a8

Original structured fields

{
  "detail": "$15 a month, $165 a year, $500 for the five-year believer plan, no free tier — against free obsidian and free, open logseq. sync bugs, slowdowns on large graphs, an afterthought mobile app, and a proprietary cloud that degraded on export.",
  "id": "clm-market",
  "sourceIds": [
    "src-nesslabs-roam-faq",
    "src-yu-why-i-left",
    "src-mickmel-migration",
    "src-conaw-sync-thread"
  ],
  "title": "price, performance, and lock-in"
}

the feature moat drained · idea-stalled

Original location · source-c93fbd25fcb5484c2f3d · source-aad8ffea41cfaab5f947 · source-1afd8938472ec4d9c672

Original structured fields

{
  "detail": "everything signature — backlinks, graph view, daily notes, block refs — was copied within eighteen months, often cheaper, local, or open. the white paper's deeper promises (bayesian weights, collaborative reasoning) never shipped. funding stopped at the 2020 seed.",
  "id": "clm-moat-drained",
  "sourceIds": [
    "src-obsidian",
    "src-athens",
    "src-roam-whitepaper"
  ],
  "title": "the feature moat drained"
}

wordcell · idea-successors

Original location · source-f2c79886b553d0269039 · source-4f23913aa952b80025a6

Original structured fields

{
  "href": "https://github.com/hraness/wordcell",
  "id": "suc-wordcell",
  "name": "wordcell",
  "note": "a knowledge base for coding agents: markdown and wikilinks committed with the repo, backlinks and indexes as replaceable views, git as authority — the roam data model with agents as the primary readers and writers.",
  "sourceIds": [
    "src-wordcell",
    "src-wordcell-docs"
  ]
}

sponge · idea-successors

Original location · source-d427882e005d41aa9ecd

Original structured fields

{
  "href": "https://sponge.computer/",
  "id": "suc-sponge",
  "name": "sponge",
  "note": "a knowledge harness for agentic research: save the corpus, bring your own agent to investigate, require cited passages and checked claims — 'agent proposes, human reviews' as the product primitive roam never had.",
  "sourceIds": [
    "src-sponge"
  ]
}

oh · idea-successors

Original location · source-1ca961629a2fc81119fd

Original structured fields

{
  "href": "https://github.com/hraness/oh",
  "id": "suc-oh",
  "name": "oh",
  "note": "an ontology kernel for agentic research: content-addressed entity, statement, assertion, and evidence records with an append-only operation log — structured claims as the agent's inspectable memory format.",
  "sourceIds": [
    "src-oh"
  ]
}

tana · idea-successors

Original location · source-b40545bca43c37fdd757 · source-cd07f2f4ba23541b9dcf · source-64fd3495b97d0316d495

Original structured fields

{
  "href": "https://tana.inc/",
  "id": "suc-tana",
  "name": "tana",
  "note": "supertags gave nodes schema, then the product rebuilt around ai: 'your graph is the context and conversations are your prompts.'",
  "sourceIds": [
    "src-tana",
    "src-tana-raise",
    "src-tana-next-chapter"
  ]
}

reflect · idea-successors

Original location · source-3e47d2312b8370fd2500

Original structured fields

{
  "href": "https://reflect.app/",
  "id": "suc-reflect",
  "name": "reflect",
  "note": "fast encrypted networked notes plus the literal labor handoff: 'decorate my writing with backlinks' as a command.",
  "sourceIds": [
    "src-reflect-backlinks"
  ]
}

obsidian · idea-successors

Original location · source-c93fbd25fcb5484c2f3d · source-b8b946a41bad6831c413

Original structured fields

{
  "href": "https://obsidian.md/",
  "id": "suc-obsidian",
  "name": "obsidian",
  "note": "the defector's destination and the agent era's de-facto vault format — local markdown, wikilinks, file over app. the doctrine is why agents can already read it.",
  "sourceIds": [
    "src-obsidian",
    "src-file-over-app"
  ]
}

logseq · idea-successors

Original location · source-84997723082acd9a105f · source-c80e42958be8bbb0583f

Original structured fields

{
  "href": "https://github.com/logseq/logseq",
  "id": "suc-logseq",
  "name": "logseq",
  "note": "the free, open-source roam workalike storing local markdown/org files — readme credits roam, org-mode, tiddlywiki, and workflowy.",
  "sourceIds": [
    "src-logseq",
    "src-logseq-raise"
  ]
}

athens · idea-successors

Original location · source-aad8ffea41cfaab5f947

Original structured fields

{
  "href": "https://github.com/athensresearch/athens",
  "id": "suc-athens",
  "name": "athens",
  "note": "the cautionary open-source clone — yc w21, 6.3k stars, unmaintained since ~2023: proof that cloning the graph was never the hard part.",
  "sourceIds": [
    "src-athens"
  ]
}

letta / agents.md · idea-successors

Original location · source-0a351a290030e1f4d7a8 · source-d4931dfaad1e407f922a

Original structured fields

{
  "href": "https://www.letta.com/",
  "id": "suc-letta",
  "name": "letta / agents.md",
  "note": "the adjacent infrastructure carrying the idea: git-backed markdown memory the agent greps and commits, plus a cross-vendor readme-for-agents convention.",
  "sourceIds": [
    "src-letta-context-repos",
    "src-agents-md"
  ]
}