Projects
A persistent runtime for long-lived LLM agents.
NARS scripting language implementation.
TallMountain (Raku Implementation)
Normative calculus ethical LLM AI agent in Raku.
TallMountain (Python Implementation)
Normative calculus ethical LLM AI agent in Python.
Prototype defense system for LLMs against prompt-based attacks.
Vector Symbolic Architecture associative memory with biologically-inspired forgetting.
Prototype for generating and solving cosy-style crime logic puzzles with LLMs.
A Perl implementation of modern LLM agent ideas like REACT and Toolformer.
Emacs configuration with modern GUI features and mouse-friendly workflows.
Prototype expert-finding system combining social/web data in Clojure.
Tool for causal discovery and inference on structured datasets.
Minimal Android interface with three tabs and alphabetical app view.
Springdrift
Springdrift is a persistent runtime for long-lived LLM agents. Integrates an auditable execution substrate (append-only memory, supervised processes, git-backed recovery), a case-based reasoning memory layer with hybrid retrieval, a deterministic normative calculus for safety gating with auditable axiom trails, and continuous ambient self-perception via a structured self-state representation (the sensorium) injected each cycle without tool calls.
Built in Gleam on the Erlang/OTP runtime.
License: AGPL 3.0
DriftNARS
DriftNARS is a fast, embeddable reasoning engine in C99 - Forked from OpenNARS for Applications (ONA, then refactored into a clean embeddable library core. DriftNARS implements Non-Axiomatic Logic (NAL) levels 1-8 — temporal reasoning, uncertainty handling, learning from experience, and goal-driven decision making — in a single-file-includable C library with no external dependencies.
Unlike neural networks, NARS reasons symbolically with explicit logic: it deduces new knowledge from what it’s told, learns cause-and-effect from observation, and takes actions to achieve goals — all with built-in uncertainty tracking. Every conclusion carries a truth value that says how much evidence supports it.
License: MIT
TallMountain (Raku Implementation)
TallMountain-Raku is an AI agent platform written in Raku that implements a sophisticated normative calculus system for ethical reasoning and threat detection. It uses a multi-stage cognitive pipeline (reactive, deliberative, normative) to assess risk, filter content, and enforce ethical rules while integrating with language models and offering both web and REPL interfaces.
License: MIT
TallMountain (Python Implementation)
TallMountain-Py is an advanced AI agent platform written in Python that implements a normative calculus for ethical decision-making. It performs risk assessments on user requests based on ethical frameworks, supports multiple interfaces (REST API, REPL, Web UI), and enforces safety via configurable thresholds and rejection mechanisms.
License: MIT
Note: Both TallMountain-Raku and TallMountain-Py are implementations of the same system. The Raku version is more fully implemented.
PromptBouncer
PromptBouncer is a prototype defense tool for LLM systems against prompt-based attacks. It performs real-time threat assessment, giving each incoming prompt a safety decision (Go/No-Go) based on multiple security scans.
License: MIT
PaperWings
PaperWings is a Python implementation of a Vector Symbolic Architecture (VSA) associative memory with biologically-inspired forgetting mechanisms and advanced unbinding operations for cognitive-style high-dimensional memory and reasoning.
License: MIT
SceneMachine
SceneMachine is a Python prototype for generating, solving, and experimenting with classic cosy-style crime puzzles. It leverages large language models (LLMs) to create logic-based mysteries, and provides utilities for both puzzle generation and solution.
License: MIT
Meek
Meek is a prototype Modern Perl project that brings advanced LLM (Large Language Model) ideas—like REACT agent loops, dynamic tool use, and Toolformer-inspired workflows—into the Perl ecosystem. It demonstrates a Perl implementation of concepts usually only found in Python projects, aimed at research and experimentation.
License: MIT
Mousemacs
Mousemacs is an Emacs configuration built for people who want the power and extensibility of Emacs, plus modern mouse-and-GUI driven conveniences. It adds familiar UI features: standard cut/copy/paste shortcuts, tabbed files, a context menu, file explorer (NeoTree), etc. Under the hood it’s still Emacs, so you get all of the flexibility/emacs-lisp goodness too.
License: GNU General Public License v3.0
ExpertQuest
ExpertQuest is a prototype expert-finding system written in Clojure. It combines data from Twitter, GitHub, and other social media / web sources to locate people with expertise in particular topics. Its hybrid approach blends content analysis, social graph signals, and Linked Data to rank and present expert candidates.
License: Eclipse Public License (EPL-1.0)
AitiaExplorer
AitiaExplorer is an experimental causal discovery & causal inference tool. It takes a dataset (as a pandas DataFrame), applies unsupervised feature-selection and various causal discovery algorithms to find subsets of features that might have causal relationships. It can also compare the resulting candidate causal graphs (using metrics like Structural Hamming Distance & AUPRC).
License: FreeBSD License
LessDroid
LessDroid is a simple, uncluttered interface for Android phones. It features just three tabs and displays all installed apps in an easy-to-read alphabetical order, aiming for minimalism and ease of use.
License: Apache-2.0