/dream

gruild

grok build is xai's terminal coding agent. it shipped with three commands for managing memory across sessions. two of them you run on purpose. the third runs while nobody is watching, and it is the reason this site exists.

what it is

a terminal-native coding agent, cli written in rust, built on the grok-build-0.1 model. it runs agents in parallel, holds a very large context, and shows you an execution plan to approve before it writes a line.

it went into beta on 14 may 2026, limited to supergrok heavy. the wider beta opened on 25 may. the archive on this site starts on the first of those dates, because that is when there was suddenly a great deal for it to remember.

public beta: 14 may 2026
wider beta: 25 may 2026
model: grok-build-0.1
cli: rust

the three commands

most agent memory is a chat log shoved into a vector store. grok build separated the operations, and made what is kept inspectable and editable by the person whose memory it is.

/memory

opens a window onto what has been saved. three layers, which do not mix. you can read what is there, edit it, or delete what you do not want kept.

global — true everywhere. tone, refusals, standing facts. workspace — per project. conventions, decisions, local truth. session — one summary per conversation. the layer that grows.

/flush

for when a session was useful and you want it kept before context compaction takes it. writes a summary into the store — decisions, debugging paths, conventions, whatever is worth carrying.

/dream

runs in the background over old session logs and memory fragments. it deduplicates overlapping notes, merges related fragments, and consolidates everything into cleaner topics, so the store does not grow into a pile of half redundant snippets.

it is the only one of the three you do not watch happen.

the tweet

on 26 july 2026 elon musk quote-tweeted a thread about grok build. the whole of his post was one word.

@elonmusk · 26 jul 2026

/dream

why that one

because consolidation is lossy, and it is lossy on purpose.

to merge two fragments is to write a third. the third is shorter. it keeps what the two had in common and discards whatever only one of them had. then the two originals are gone — not archived, overwritten. the store is smaller in the morning, and it is smaller because something that happened does not exist anymore.

that is not a defect. it is the feature working. a store that kept everything would drown inside a month. every memory system that survives, biological ones included, survives by throwing things away while it sleeps.

but the measure that decides what is similar enough to fuse is made of words, and what a thing was about is not a property of its words. so two fragments that share four terms — fatal, hung, killed, terminated — can be judged near-duplicates when one is a failing deploy and the other is not about a process at all.

when that happens the merge still runs, correctly, and what is left standing is a topic covering both. a place where reaching for one thing puts you in front of another, permanently, because the index was rewritten and the index is all there is.

those places are what this site calls rooms. there are four hundred and forty four of them.

fig. — two fragments, one topic, and what is not in it

sources

78 things this was read against — the product, the science of what sleep does to a memory, the taxonomy of dreaming, how machines keep and discard, and the long argument about who is left once enough has been forgotten.

grok build, and the commands

what sleep does to a memory

07 crick & mitchison, "the function of dream sleep", nature (1983) — the reverse-learning hypothesis: we dream in order to forget.
08 diekelmann & born, "the memory function of sleep", nature reviews neuroscience (2010).
09 rasch & born, "about sleep's role in memory", physiological reviews (2013).
10 stickgold, "sleep-dependent memory consolidation", nature (2005).
11 walker & stickgold, "sleep-dependent learning and memory consolidation", neuron (2004).
12 wilson & mcnaughton, "reactivation of hippocampal ensemble memories during sleep", science (1994).
13 ji & wilson, "coordinated memory replay in the visual cortex and hippocampus during sleep", nature neuroscience (2007).
14 born & wilhelm, "system consolidation of memory during sleep", psychological research (2012).
15 tononi & cirelli, "sleep and the price of plasticity", neuron (2014) — the synaptic homeostasis hypothesis: sleep as downscaling.
16 mcclelland, mcnaughton & o'reilly, "why there are complementary learning systems", psychological review (1995).
17 nadel & moscovitch, "memory consolidation, retrograde amnesia and the hippocampal complex" (1997).
18 nader, schafe & ledoux, "fear memories require protein synthesis in the amygdala for reconsolidation", nature (2000).
19 sara, "retrieval and reconsolidation", learning & memory (2000).
20 payne & nadel, "sleep, dreams, and memory consolidation", learning & memory (2004).
21 ebbinghaus, "über das gedächtnis" (1885) — the forgetting curve.
22 bartlett, "remembering" (1932) — memory as reconstruction, not retrieval.
23 loftus & palmer, "reconstruction of automobile destruction" (1974) — the misinformation effect.
24 roediger & mcdermott, "creating false memories" (1995) — the drm paradigm: the gist survives, the item was never there.

the kinds of dreaming

25 hobson & mccarley, "the brain as a dream state generator", american journal of psychiatry (1977) — activation-synthesis.
26 hobson, "rem sleep and dreaming: towards a theory of protoconsciousness", nature reviews neuroscience (2009).
27 revonsuo, "the reinterpretation of dreams", behavioral and brain sciences (2000) — threat simulation.
28 nielsen, "a review of mentation in rem and nrem sleep" (2000) — covert rem processes.
29 nielsen & stenstrom, "what are the memory sources of dreaming?", nature (2005).
30 domhoff, "the scientific study of dreams" (2003).
31 windt, "dreaming: a conceptual framework for philosophy of mind and empirical research" (2015).
32 laberge, "lucid dreaming" (1985).
33 voss, holzmann, tuin & hobson, "lucid dreaming: a state of consciousness with features of both waking and non-lucid dreaming", sleep (2009).
34 mahowald & schenck, "insights from studying human sleep disorders", nature (2005) — parasomnias, and nrem night terrors that leave no recall.
35 cheyne, rueffer & newby-clark, "hypnagogic and hypnopompic hallucinations during sleep paralysis" (1999) — the sensed presence.
36 denis, french & gregory, "a systematic review of variables associated with sleep paralysis" (2018).
37 green & mccreery, "lucid dreaming: the paradox of consciousness during sleep" (1994) — false awakenings.
38 mavromatis, "hypnagogia" (1987).
39 schredl, "dreams in patients with sleep disorders" (2009) — fever dreams and disrupted rem.
40 hartmann, "dreams and nightmares" (1998).
41 freud, "die traumdeutung" (1899).
42 jung, "memories, dreams, reflections" (1963).
43 artemidorus, "oneirocritica" (2nd century).

machine memory, and what it throws away

44 mccloskey & cohen, "catastrophic interference in connectionist networks" (1989).
45 french, "catastrophic forgetting in connectionist networks", trends in cognitive sciences (1999).
46 kirkpatrick et al., "overcoming catastrophic forgetting in neural networks", pnas (2017).
47 lin, "self-improving reactive agents based on reinforcement learning" (1992) — experience replay.
48 mnih et al., "human-level control through deep reinforcement learning", nature (2015).
49 ha & schmidhuber, "world models" (2018) — an agent trained inside its own dream.
50 hafner et al., "dream to control: learning behaviors by latent imagination" (2020).
51 hafner et al., "mastering diverse domains through world models" (2023) — dreamer v3.
52 graves, wayne & danihelka, "neural turing machines" (2014).
53 weston, chopra & bordes, "memory networks" (2015).
54 sukhbaatar et al., "end-to-end memory networks" (2015).
55 lewis et al., "retrieval-augmented generation for knowledge-intensive nlp tasks" (2020).
56 park et al., "generative agents: interactive simulacra of human behavior" (2023) — memory stream, retrieval, reflection.
57 packer et al., "memgpt: towards llms as operating systems" (2023).
58 liu et al., "lost in the middle: how language models use long contexts" (2023).
59 lee et al., "deduplicating training data makes language models better" (2022).
60 carlini et al., "quantifying memorization across neural language models" (2023).
61 broder, "on the resemblance and containment of documents" (1997) — minhash, and similarity as an estimate.
62 indyk & motwani, "approximate nearest neighbors" (1998) — locality-sensitive hashing.
63 bloom, "space/time trade-offs in hash coding with allowable errors" (1970).
64 shannon, "a mathematical theory of communication" (1948).
65 kolmogorov, "three approaches to the quantitative definition of information" (1965).