Aug 24, 2026
For decades, software engineers have struggled with the dynamic lifecycle of components within long-running processes. OSGi pioneered dynamic module loading in Java, while Cordiverse formalizes it mathematically with revertible effects. Datom.world approaches the same problem by sidestepping mutable state entirely, leveraging immutable streams and treating the AST as queryable datoms to make dynamic composition structural rather than imperative.
Read the article → Aug 23, 2026
Three centuries before distributed systems, Gottfried Wilhelm Leibniz formulated the Monadology: a reality of indivisible, windowless substances coordinating without shared physical contact. Datom.world realizes this metaphysics as a computational architecture. By treating datoms as relational events, agents as windowless continuation machines, semantics as external interpretation, and coordination as stigmergy over immutable streams, Datom.world bridges 17th-century rationalism with 21st-century distributed systems.
Read the article → Aug 23, 2026
Monad tutorials in computer science are notoriously opaque. For Clojure developers accustomed to immutable data, threading macros, and unbundled atoms, monads can seem like an alien construct. This article traces monads from their 1950s origins in category theory through Eugenio Moggi and Philip Wadler's effect revolution in Haskell, and explains how monads relate to Clojure idioms, macros, and continuation machines.
Read the article → Aug 18, 2026
Both Datom.world and Nathan Marz's Rama start from the exact same premise: traditional databases are a category error, and data should be managed as immutable logs and projected views. However, their solutions take different paths. Where Rama unifies infrastructure into a highly programmable, integrated platform, Datom.world uses strict abstraction boundaries and mobile continuations to remain open and composable.
Read the article → Aug 17, 2026
Datomic revolutionized databases by introducing the "database as a value." However, its storage model evolved to address the tension of storing a mutable root alongside immutable segments. By examining this history, we see why the strict abstraction boundaries of dao.jing and dao.stream represent the structural endgame of this architectural lineage, elegantly scaling from distributed cloud to embedded single-process deployments.
Read the article → Aug 16, 2026
Modern architecture treats data lakes, operational databases, and vector stores as separate silos requiring costly ETL replication. In Datom.world, a data lake is not a silo or a monolithic engine: it is simply an interpreter point in the open moduli space of databases, projecting columnar scans and point queries over the shared, content-addressed substrate of DaoJing—which itself can reside directly on Amazon S3.
Read the article → Aug 14, 2026
Datom.world does not prescribe one universal database. It supplies a stream substrate and a core axiom: semantics are external to the stream. An interpreter observes streams and materializes data structures; the observation fixes a point in the moduli space of possible databases.
Read the article → Aug 13, 2026
I once argued that every piece of information should be a five-slot datom. Mixed-dimensional Datalog changed my mind: over an open n-tuple tuple space , relations of different dimensions can meet through shared variables, while the five-slot datom remains a powerful local contract.
Read the article → Jul 21, 2026
Coding agents do not need an orchestrator to collaborate. They need a shared medium . Once agents coordinate by leaving and reading traces on a tuple space, three famous models of collective intelligence (Mixture-of-Agents, the ant colony, and the wisdom of the crowd) stop looking like rivals and start looking like points in one design space. This is how they compose in dao.space: MoA is a circuit, the hive is a search, the crowd is a judge, and the tuple space is the substrate that lets you run all three at once.
Read the article → Jul 18, 2026
GKE's Agent Sandbox suspends agents by freezing entire pod memory images. datom.world suspends agents by persisting their continuation as datoms: only the kernel of the computation. By communicating with legacy libraries over streams (pi-calculus) instead of linking them in-process, agent state never entangles with library state. The result is lightweight, architecture-neutral agents (written in Python, Clojure, JS, or PHP) that suspend, migrate, and resume seamlessly via dao.space.
Read the article → Jun 27, 2026
Biological systems coordinate at scale using stigmergy: indirect coordination through a shared environment, like ants leaving pheromone trails. The tuple space is stigmergy made formal, a shared space where processes read and write facts instead of addressing each other. This piece traces the fifty-year evolution of that idea into dao.space , which finally makes the stigmergic medium durable, content-addressed, and trust-bounded by combining it with Datomic's storage and query disciplines.
Read the article → May 31, 2026
At the boundary where an engine meets the world, whether it evaluates Abstract Syntax Trees or neural network tokens, the architecture of isolation requires stigmergy: emitting data to be interpreted by a host. Bringing an LLM agent inside a virtual machine reveals a profound, nested symmetry.
Read the article → May 10, 2026
GPU utilization is limited less by FLOPS than by architecture. When inference and training are treated as movable suspended computations over streams, scarce accelerators can be used more fully without discarding live work.
Read the article → Apr 12, 2026
Git stores code in units that are too coarse for agents. Files force sub-file semantic work through line diffs, so renames, reorders, and refactors collide for accidental reasons. When code lives in DaoSpace as tuples, the storage unit and the code unit finally match. That makes DaoSpace a better Git than Git for agentic coding.
Read the article → Apr 2, 2026
Exploring how the SeniorCareMarket/eve library brings zero-serialization shared-memory to datom.world, enabling lock-free transactors and true in-browser concurrency for agents.
Read the article → Mar 28, 2026
A useful agent needs more than a larger context window. It needs a way to move facts from transient inference into explicit, queryable memory. In Datom.world, Agent Smith plays that hippocampal role by selecting, consolidating, and reconstructing datoms across streams.
Read the article → Mar 21, 2026
I previously described DaoSpace as a moduli space of databases. That confused one interpreter with the larger space it inhabits. DaoSpace is one precise thing: an immutable n-tuple space built from associative matching and generative communication.
Read the article → Mar 12, 2026
Once Yin.vm externalizes CESK state as datom streams, JIT compilation and garbage collection stop being separate runtime subsystems and become two stream interpreters over the same execution log. A third interpreter, operating at the timescale of sessions, extends that coverage to patterns no execution-frequency profiler accumulates. All three answer by emitting datoms into the same continuation scheduler.
Read the article → Mar 9, 2026
How Yin.vm reimagines macros not as a hidden compiler phase, but as auditable, causality-preserving events in a datom stream. We compare this capability-gated approach to traditional Lisp macros, Fexprs, Racket, and Unison.
Read the article → Mar 8, 2026
Datom streams align with process calculi and causal categories: computation is interaction, history is causal structure, and interpreters behave like functors from event streams to application-specific views.
Read the article → Mar 7, 2026
Feynman's path integral treats evolution as a sum over histories. Datom systems record one realized history, then derive state by interpretation. Both are history-first views of reality and computation.
Read the article → Mar 4, 2026
Modern physics is event-first, not object-first. The datom tuple [e a v t m] fits this view: each fact is a causal event, and higher structures emerge through interpretation.
Read the article → Mar 2, 2026
Consensus can be treated as entropy reduction across distributed observers. Security, decentralization, and scalability become thermodynamic tradeoffs in how much disagreement a system can remove per unit resource.
Read the article → Mar 1, 2026
The trilemma is usually framed as decentralization, security, and scalability. A more precise view is that it emerges from bounded observation : nodes see different prefixes of global history, so they must pay to reconcile those views.
Read the article → Feb 28, 2026
A performance analysis of the datom.world execution models, comparing the Register, Stack, Semantic, and AST Walker VMs. Includes details on how the Semantic VM was optimized to beat direct tree-walking.
Read the article → Feb 22, 2026
A note on the path from Hilbert spaces to infinity: structure versus semantics, incompleteness, unbounded recursion, and the ontological split between substrate and interpreter.
Read the article → Feb 18, 2026
The CESK machine (Control, Environment, Store, Continuation) is the theoretical foundation of Yin.vm, but Yin.vm’s implementation flips the traditional model: instead of storing state inside the machine, it externalizes everything into queryable datom streams. This post traces CESK’s origins in semantic semantics, explains each component with concrete datom examples, and shows how Yin.vm’s stream‑based CESK enables lightweight continuation migration, full introspection, and time‑travel debugging.
Read the article → Feb 18, 2026
Yin.vm/Yang, Gambit Scheme, and Ribbit all descend from the Scheme tradition of first‑class continuations, yet their designs diverge radically. This blog compares their philosophies, architectures, and tradeoffs: a data‑centric stream‑based VM for mobile agents, a high‑performance Scheme‑to‑C compiler, and a compact portable VM that targets 25+ host languages.
Read the article → Feb 17, 2026
The Language Server Protocol (LSP) treats code as text and IDE features as RPC calls. Yin.vm inverts this: the Universal AST stored as datoms makes the entire codebase a Datalog database. The IDE becomes a query interface: no per‑language servers, no text‑based diffs, just semantic queries across languages and time.
Read the article → Feb 13, 2026
AI agents become dangerous when they escape containment. In Yin.VM, an LLM does not run as a process. It creates continuations that execute inside pocket universes: sandboxed streams the executing node constrains by design.
Read the article → Feb 9, 2026
Yin.VM starts from invariance, not language. This essay frames the AST as the invariant and compilation as symmetry preserving transformation.
Read the article → Feb 4, 2026
Exploring the architectural implications of block-based editing within the Yin VM. When code is fundamentally data, visual assembly becomes a direct viewport into the AST, bypassing text representation entirely.
Read the article → Feb 3, 2026
What if the C Standard Library (LibC) wasn't a collection of system calls, but a schema for data streams? By reimplementing LibC as a wrapper over Datom.world streams, we unlock a universe where legacy C programs become automatically distributed, persistent, and reactive—running seamlessly in kernel space, user space, or across a global grid.
Read the article → Feb 2, 2026
Solid pods allow users to own their data, but they cannot control what an application does once access is granted. By reversing the relationship and moving code to data via restricted continuations, datom.world ensures that your data never leaves your sphere of control.
Read the article → Jan 26, 2026
We benchmark semantic bytecode (datom triples) against traditional numeric bytecode across three platforms: JVM, Node.js, and Dart. The results quantify the tradeoff: semantic bytecode is 4-11x slower to compile and 1-7x slower to execute, but enables queries that are impossible with numeric bytecode. The cost of introspection is measurable, but so is the value.
Read the article → Jan 25, 2026
In the landscape of virtual machine design, the boundary between host and execution engine is often the most significant bottleneck. For Yin.VM, the computation core of the Datom.world ecosystem, we have moved away from function-based Foreign Function Interfaces. Instead of treating the VM as a library to be commanded through imperative calls, Yin.VM treats the execution boundary as a continuous stream of Datoms. This shift from "doing" to "telling" represents a fundamental rethink of how decentralized systems should handle cross-language execution.
Read the article → Jan 24, 2026
The promise of decentralized intelligence is vast: millions of IoT devices collaborating to train smarter AI models without sharing raw, private data. But in a zero-trust environment, this dream faces a critical nightmare: Poisoning Attacks. By synthesizing Yin.VM, DaoSpace, and Shibi, Datom.world creates a self-defending nervous system for decentralized intelligence, where participants don't need to trust each other, only the immutable physics of the infrastructure.
Read the article → Jan 23, 2026
ClojureDart compiles Clojure to Dart at build time, inheriting Dart's static compilation constraints. No runtime eval, no dynamic code loading, no traditional macros. But what if the dynamism moved from the language into the data? Yin.vm and the Yang compiler are written in portable cljc code, running on JVM, JavaScript, and Dart via existing Clojure compilers. A native Dart implementation provides a performance baseline. The result: full Lisp-like flexibility on Flutter's cross-platform runtime.
Read the article → Jan 21, 2026
The Traveling Salesman Problem is NP-hard: no algorithm solves it in polynomial time. Yet logistics networks route millions of packages daily. The trick is fungibility. Money reduces TSP from an intractable combinatorial search to a tractable distributed optimization: prices compress constraints into scalars, arbitrageurs explore in parallel, and fungible tokens provide the slack that enables approximate solutions. Use the market solution as initialization for gradient descent and you get polynomial-time global search with high-precision local refinement. This reframing reveals markets as the initialization engine for optimization, inflation as state decay, and trust as computational bit-depth.
Read the article → Jan 20, 2026
Browser bookmarks are dead data. The CLI tool buku liberated them to the terminal. Datom.world liberates them to the stream. Why managing knowledge requires more than just a database file
Read the article → Jan 15, 2026
Plan 9 reimagined Unix as a distributed file system. Datom.world evolves this vision into a distributed reactive state machine, replacing bytes with facts and processes with continuations.
Read the article → Jan 13, 2026
History shows that software architecture has largely prioritized "freedom." Developers created generic Tuple Spaces capable of holding anything, SQL tables with infinite schema variations, and JSON documents of any shape. Datom.world proposes a radical inversion: Restriction is the catalyst for evolution.
Read the article → Jan 12, 2026
The Tao Kernel treats data not as static files, but as an event stream. By combining Single Address Space principles with Semantic Interpretation, the Tao Kernel enables stream-based interpretation and isolated execution.
Read the article → Jan 10, 2026
Datom.world is a substrate for stigmergy: how ants, cells, and neurons coordinate without central control. 24 use cases from Flash Firms to time-travel debugging.
Read the article → Jan 7, 2026
The Agent Web replaces ad-driven scraping with browser-equipped agents, LLM-powered knowledge graphs, and utility tokens. A dual economy: pay-for-truth (ownership) and pay-for-utility (rental).
Read the article → Jan 7, 2026
Firms exist to minimize transaction costs. When the Agent Web reduces these costs to near zero, the firm dissolves into stigmergy : coordination through shared data rather than hierarchy.
Read the article → Jan 6, 2026
We lift language out of the Flatland of probability tables into the high-dimensional space of vectors, where we can finally move freely between ideas.
Read the article → Dec 31, 2025
RDF triples lack crucial dimensions for applications. Datom.world extends to five-tuples: entity, attribute, value, time, metadata.
Read the article → Dec 31, 2025
Without political borders, boundaries still emerge: fractal patterns of economic cost and contextual relevance. Capability tokens create nested borders based on expertise. Borders from physics, not decree.
Read the article → Dec 29, 2025
Datom.world treats streams and continuations as data structures, unifying information mobility, channel mobility, and code mobility in a single model.
Read the article → Dec 28, 2025
DaoStream can operate at Layer 2, Layer 3, or Layer 7. Layer 2 is most elegant; Layer 7 is most pragmatic.
Read the article → Dec 28, 2025
Plan 9's 9P required 13 message types to maintain POSIX semantics. DaoStream admits what communication actually is: datom passing. One packet type. Stateless routing. Interpreters at the edges.
Read the article → Dec 27, 2025
I started Yin.VM in Rust and progress stalled. Switching to Clojure and Datascript changed everything. The right foundation matters more than building from zero.
Read the article → Dec 26, 2025
Ten hard objections to continuation-based mobile computation, and why the synthesis succeeds now when individual components failed before.
Read the article → Dec 20, 2025
Four critiques of attention markets, and how Shibi's polycentric, user-issued token architecture addresses each.
Read the article → Dec 18, 2025
Yin.VM keeps the complete program structure as queryable datoms. Systems can inspect and modify their own code through Datalog.
Read the article → Dec 18, 2025
Interfaces enforce stability through rigid contracts. Streams replace prescriptive blueprints with descriptive genomes. Systems that adapt rather than ossify.
Read the article → Dec 17, 2025
Software fails because it has no source of truth. Structural decoherence across databases, APIs, and frontends creates a translation tax. Until we enforce architectural constraints at the execution layer, entropy wins.
Read the article → Dec 10, 2025
Yin.vm inverts traditional execution: continuations become thin control objects that migrate to where data lives, not containers dragging gigabytes of heap. By externalizing state into streams and intelligently resolving symbols, computation becomes mobile—making distributed computing feel local without pretending everything is.
Read the article → Dec 9, 2025
Object databases failed because storing objects stores semantics . Language-specific layouts decay when languages evolve. Datom.world stores pure syntax, keeping semantics external.
Read the article → Dec 9, 2025
Fixing all information into a five-element tuple seems limiting, but restriction becomes power: uniform structure eliminates negotiation, enables effortless composition, simplifies distribution, and frees interpretation, so a smaller tuple opens a larger world.
Read the article → Dec 9, 2025
Datom.world explained in plain language (Vietnamese, Chinese, English): a unified structure for information where people own their data and software becomes humane.
Read the article → Dec 8, 2025
TCP forces global ordering and connection state, smuggling in semantics that contradict Datom.world's axioms. UDP provides the raw substrate where causality lives in datom metadata (as entity references), not the wire—enabling mobile continuations, CRDT merges, and interpreter-directed reliability without transport-layer lies.
Read the article → Dec 7, 2025
Continuations aren't just control flow—they're the universal semantic kernel . As first-class datoms in Yin.vm, they unify exceptions, async/await, generators, coroutines, backtracking, effects, and distributed computation, making nearly every language feature a special case of manipulating continuations.
Read the article → Dec 7, 2025
Yin.vm's universal AST-as-datom design transforms programming from "code as text" to "code as queryable data". When combined with DaoSpace's distributed tuple store, this architecture creates a new computational substrate where programs, continuations, and knowledge graphs coexist in a single queryable space. LLMs become meta-compilers operating over structured semantic graphs rather than generating free-form text, while contracts and ontologies eliminate hallucinations at the system boundary.
Read the article → Nov 27, 2025
LLM training isn't clustering—it's learning a geometric representation that makes next-word prediction efficient. Through gradient descent, the model discovers that encoding semantics as geometry minimizes loss, creating a space where vector arithmetic mirrors linguistic operations by necessity, not design.
Read the article → Nov 23, 2025
Computation is structural transformation through three operations: expansion (generating richer semantics), compression (extracting basis-free invariants), and morphism construction (building bridges between structures). This framework unifies interpretation, learning, and understanding.
Read the article → Nov 22, 2025
Large cardinal axioms express reflection principles —properties of the universe V already true in smaller sets M ⊂ V. These logical fractals bridge infinite set theory to distributed communication complexity: reflection in mathematics mirrors channel nesting in π-calculus, revealing why work emerges from dimensional gradients.
Read the article → Nov 22, 2025
Every interpreter adds a dimension to state-space . Crossing interpretive layers creates dimensional gradients, and traversing them requires work—this is computation's fundamental structure. It explains why complexity emerges, why compression is computation, and why the universe might be self-interpreting.
Read the article → Nov 22, 2025
DaoSpace stores all states, not just current state . Time travel, audit trails, and undo are fundamental. Local-first with quantum-inspired synchronization.
Read the article → Nov 21, 2025
Does semantics come from interpretation or structure? Both: Semantics = (structure) × (interpreter) . Chinese characters encode semantic associations (structure), but meaning only emerges when a reader interprets them. Remove either factor and semantics vanishes—this resolves the apparent contradiction.
Read the article → Nov 19, 2025
Maxwell's demon doesn't add energy—it releases hidden potential by interpreting information (observing particles and acting on them). Datom.world agents unlock economic value the same way: not creating more data, but interpreting existing datoms in smarter ways, extracting work from chaotic streams.
Read the article → Nov 19, 2025
Datom.world erases the kernel/userspace boundary: everything is streams and interpreters . Yin.vm continuations handle isolation without memory protection. The OS reduces to a universal append-only fabric where interpreters are the primitive executors, making the rigid split a relic.
Read the article → Nov 19, 2025
All money—BTC, USD, gold—is Monopoly money: accounting artifacts whose value comes from trust, not materials. Money's power lies in creating flow through skilled people. The real question isn't BTC vs USD, but which system generates the best flow of human productive activity.
Read the article → Nov 19, 2025
ShiBi cannot be money . Money requires global transferability; ShiBi is local and subjective. Agents compensate each other for attention without universal agreement.
Read the article → Nov 18, 2025
Logseq uses DataScript queries but stores data as Markdown files —creating architectural tension. The canonical truth lives in mutable text, not immutable datoms, forcing constant parsing overhead and making git diffs meaningless. It's a knowledge graph trapped in a document format.
Read the article → Nov 18, 2025
Google Docs' document model forces Operational Transform complexity, loses version history, and surrenders data ownership. Datom streams with branch-aware storage enable true collaboration where local edits are never lost and merges create new branches instead of overwriting your work.
Read the article → Nov 17, 2025
AI agents need unrestricted access to data streams, but SaaS vendors lock data behind paywalls. Datom.world eliminates this with local-first datom streams —you own your data, apps become interpreters, and agents operate as first-class peers without API barriers.
Read the article → Nov 17, 2025
RQM says reality is interactions, not state . π-calculus says computation is message-passing, not mutation . These are the same insight: both reject independent facts/state as primary, making interaction fundamental. DaoSpace implements this—distributed observers correlate through sync, not through shared global state.
Read the article → Nov 17, 2025
Datoms appear fixed-size— [e a v t m] —but their elements aren't. Variable-size values create hidden performance costs in parsing, indexing, and caching. The solution: hybrid representation with typed streams, interning, and columnar encoding that balances semantic universality with execution speed.
Read the article → Nov 16, 2025
Datalog is defined by semantics, not syntax : relational foundation, Horn clause logic, stratified negation, bottom-up evaluation, set semantics. Datomic's EDN, Prolog's predicates, SQL's WITH RECURSIVE—all Datalog if the evaluation model matches. Surface representation is irrelevant.
Read the article → Nov 16, 2025
DaoSpace implements Relational Quantum Mechanics : no global state exists, only local states relative to observers. When isolated devices sync, their conflicting states undergo wave function collapse through CRDT merge—creating correlation, not convergence to pre-existing truth.
Read the article → Nov 15, 2025
The Universal AST looks like assembly —flat, explicit, verbose datom sequences. But it operates at a fundamentally higher abstraction: assembly manipulates machine state, the Universal AST represents program semantics. It's the difference between "move register" and "function application"—low-level form, high-level meaning.
Read the article → Nov 15, 2025
When you edit AST directly , text becomes a materialized view—not the canonical code. Syntax becomes a user preference (C-like vs Lisp vs Python), refactoring becomes querying datoms, and collaboration merges AST changes, making text's tyranny obsolete.
Read the article → Nov 15, 2025
Traditional wisdom says you must choose between bytecode performance or AST queryability. Datom streams blur this distinction —by compiling ASTs to linear execution streams via Datalog, we achieve bytecode-like speed while preserving full semantic introspection.
Read the article → Nov 15, 2025
When ASTs are datoms in DaoSpace, the entire program becomes a queryable database . Every optimization transforms from manual tree-walking to declarative Datalog queries—enabling whole-program reasoning, composable analyses, and user-programmable compilation impossible in traditional architectures.
Read the article → Nov 15, 2025
How can Java's classes and Clojure's closures share a Universal AST? The semantic differences are different encodings of the same concepts : classes are closures with vtables, exceptions are non-local continuations, null is an optional type. The Universal AST represents the underlying computation, not the surface syntax.
Read the article → Nov 15, 2025
One AST maps to many valid syntaxes (Python's list comprehension vs map() vs explicit loop). The renderer must choose, but the choice is arbitrary. The solution: user preferences, heuristics based on context, and diff-aware rendering that preserves the original syntax when semantics haven't changed.
Read the article → Nov 14, 2025
Chinese dialects are mutually unintelligible, yet all share written characters —meaning in the character, not the sound. Yin.vm's Universal AST is programming's equivalent: languages share semantic representation (datom streams) while varying in syntax. Code becomes portable across languages, just as Chinese text is readable across dialects.
Read the article → Nov 14, 2025
ASTs are traditionally static trees, but in Datom.world they're materialized views over datom streams spanning five dimensions: structure, time, types, language transformations, and execution state—making code simultaneously queryable, portable, and alive.
Read the article → Nov 13, 2025
Schemas appear to carry meaning, but data is always syntax . Meaning emerges through interpretation. Schemas are secret interpreters: type checkers evaluate, reasoners infer.
Read the article → Nov 4, 2025
Unitarity (information preservation) + π-calculus (local communication) explain why communication has a finite speed limit . Instantaneous communication would violate causality and create contradictions in information flow. The speed of light is the architectural parameter enforcing locality in the universe's distributed computation.
Read the article → Jan 10, 2025
Good code isn't designed, it evolves through a dance with entropy. By bounding complexity within each module and distributing it outward, code becomes malleable, the only objective measure of quality that matters. Bounded complexity ensures changes stay local and the system remains learnable.
Read the article → Dec 2, 2024
Datoms flow as immutable streams that can be folded into state anywhere —phone, WASM, edge nodes. Using Clojure transducers, we process these streams with zero intermediate collections, making distributed computation deterministic and conflict resolution just another transformation.
Read the article →