Why I stopped trying to give LLM agents a better memory and instead built Lemmalog, a Datalog engine that maintains an agent's knowledge as analysis state, with provenance, retractions and incremental evaluation, plus what happened when I benchmarked it on LongMemEval and LoCoMo.

In the early days of our PPoSS project, we built Slog, an untyped logic programming language that extended Datalog to be much closer to Scheme. In this post, I want to describe Slog informally, but also describe some of our initial excitements, alongside key challenges we faced in our prototype implementations. Slog was initially designed using MPI and InfiniBand, running on Theta (a now-retired supercomputer at Argonne). We still maintain an old version of the Slog 1.0 repository, but our focus has moved on; the reality is that the codebase had grown to a gnarly mishmash of partial implementations along with a research prototype compiler. We have recently started new efforts to reboot the ideas from our initial Slog project to modern hardware (e.g., GPU clusters) and especially focusing on GPUs. In previous attempts, we had planned to integrate Slog with SMT solving–personally, I am wary of this in light of the recent LLM takeover. However, I am still optimistic on high-performance exact reasoning, especially in a way which enables richly-expressive queries that involve nontrivial logic reasoning.

The art of learning new programming languages is at risk due to AI. Meanwhile, Verse is set to replace Blueprints in Unreal Engine 6. How will we, as a socie...

Prolog library for interfacing with LLMs. Contribute to vagos/llmpl development by creating an account on GitHub.

Lecture presented at the OntoBRIX conference on Ontology and Semantics at the Intersection of Linguistics held in Brixen in June 2026https://grokipedia.com/p...

What is datalog? What can you do with datalog? Getting Started Path Reachability Python Naive Semi-Naive Indexing Lattice SQL recursive common table subexpressions Naive SQL translation Seminaive Lattices Ocaml Naive Rust Magic Set Program Analysis Evaluation Constant Propagation Symbolic Evaluation Reaching Definitions Liveness Points To Available Expressions Intersection Very Busy Expressions Zippers For Program Points Dominators Forall Emulation Doop Datalog Diassembly / Decompilers Bap Resources First Class Sets \& Reflection BitSets Bitset reflection Sort Lists Patricia Tries BDDs Call Meta circular interpreter BogoSort Translating functional programs Lists Dynamic Programming Q learning Mandelbrot Ball and Springs Sudoku Constraint handling Rules (CHR) Backprop Lambda representation Parsing Hilog Equality Saturation Term Rewriting Datalog Modulo Term Rewriting Graph rewriting Graph Algorithms Reachability Shortest Path Spanning Tree Clique Cycle Subgraph Matching Coloring Propagators Boolean Constraint Propagation Difference Logic Finite Domain Macros define ORBODY(h,b1,b2) CLAUSE(h, b1) CLAUSE(h, b2) define ORHEAD(h1,h2,b) CLAUSE(h1,b) CLAUSE(h2,b) define IMPLBODY(h, b1, b1) CLAUSE( h, (b1, b2) ) define IMPLHEAD() Emulating Prolog Need Sets Magic Set First class union find Translating Imperative Programs Iteration Model Checking Timestamping Theorem Proving Skolemization for Existential Heads Goals / Queries Uncurrying Contextual Datalog / Hypothetical Datalog Higher Order Clauses (Harrop) Stack database / Harrop Datalog / Tentative Datalog Existenial Queries Universal Quantifier Geometry Categorical Example Typeclass resolution Borrow Checker Type checking Coinductive or Greatest Fixed Point Datalog DFA Minimization CRDTs MultiSet Semantics Access Control Policies Networks Make Topics Answer Set Programmng Provenance Semi Naive Evaluation Algebraic Data Types Lattices Subsumption Subsumption as a master feature Provenance max min Recurusive Sum and Count Lattices Min/max lattice Maybe/Option lattice Intervals Widening Dyadic lattice Equivalence relations Negation Choice domain Semiring Semantics Probability Datalog+- and the chase Tabling Descriptive Complexity and Least Fixed Point Logic Push based Datalog Incremental / Differential Datalog Backtracking a Datalog Implementations Rel DDlog IncA Formulog Datafrog Ascent Flix dr lojekyl Datafun QL Souffle intrinsic functors floats Souffle proofs Aggregates User Defined Functors ADTs Contexts are King field accessors Vectors Use ADT instead of autoinc() Record Packing Macros Components Choice Domain Negation Souffle source EPR datalog Resources include prelude.ml class(slotname : f(x,y) , ) :- building souffle emscripten

Simple Shallow Embedding of Seminaive Datalog into SQLite

Warren's Abstract Machine: A Tutorial Reconstruction by Hassan Ait-Kaci - a-yiorgos/wambook

I’m pleased to announce the release of Incremental (well commented mli here), a powerful library for building self-adjusting computations, i.e., computations...

I’ve been working on improving how we schedule maintenance in Akamai’s cloud infrastructure, especially disruptive maintenance on hypervisor hosts serving hundreds of thousands of guest VMs. The problem is fairly complex, with competing constraints such as capacity, customer disruption SLAs, and concurrency limits across hosts, racks, and datacenters.

This is one part of an idea that’s been haunting me for around 18 years now: can we take the Logic Programming model as in Prolog or Kanren, and remove the concept of “logic variable” from it, ending up with a fully point-free logic programming? The advantage of Logic Programming for malleable software is that it is a very simple way of getting a personal database that you can not just search but do arbitrary inferences on. It can be ridiculously easy to type data into a logic database. Though ...

Event-Based Modelling andReifications in Turtle 1.2

Exploring the world of constraint solvers with very simple examples.

Contribute to alexpdp7/prolog-asciidoc development by creating an account on GitHub.