Ideas, implemented.

agent-guard

Cedar-powered per-tool-call authorization for AI agents — tamper-evident audit, RFC 8693 delegation, AuthZEN PDP.

ai-agents / audit-log / authentication / authorization / authzen

agent-passport

Identity, delegation, and trust for autonomous AI agents.

ai-agents / algorand / delegation / javascript / reputation

alfred

Local-first research agent runtime: HTTP/SSE turn loop, SQLite+JSONL persistence, 25+ MCP workers (search, plotting, evidence DAG, image generation).

agent / golang / htmx / llm / mcp

avqa

Adaptive Vector Quantized Attention — production-grade attention infrastructure for PyTorch (drop-in AVQ-Attention backend)

attention / avq-attention / efficient-attention / long-context / pytorch

axiom

A faster deterministic fully dynamic maximal matching algorithm — pure-Python reproduction of Chuzhoy, Khanna, and Song (arXiv:2605.00797v1, STOC 2026), supporting both single-level and multilevel modes.

algorithm / complexity / data-structures / deterministic-algorithm / dynamic-algorithms

beacon

Training-free Sliding Window Attention with sinks: a reproduction of 'Sliding-window beats linear attention' (Jolicoeur-Martineau et al., 2026) for decoder-only LMs.

attention / inference / llm / long-context / python

braid

Polymorphic LLM-backed ranker for recommendation systems.

hashing / information-retrieval / llm / python / pytorch

btx

Parse Bitcoin transactions, extract ECDSA and Schnorr signatures (r, s, z), detect nonce reuse, edit PSBTs, decode scripts, and broadcast via Blockstream / Mempool / blockchain.info — pure-Python secp256k1 with optional libsecp256k1 backend.

bip-174 / bip-340 / bip-341 / bitcoin / bitcoin-script

cedrus

Compile organizational authorization intent into validated, deployable Cedar policies.

authentication / authorization / cedar / iam / llm

cinestudio

Multi-agent AI film rendering platform. 27 specialised Graph agents + render dispatcher + copilot generate 30-second to 20-minute films from a single prompt.

ai-film / ai-video / cinematography / multi-agent-orchestration / nextjs

context

Partition-Prompt-Aggregate context engineering for LLMs (ACE, OKF, PPA) — Apache-2.0

ace / agentic-context-engineering / context-engineering / litellm / llm

convexfolio

Closed-form and numerical portfolio optimization

convex-optimization / docker / portfolio-optimization / python / quantitative-finance

delta-search

Delta-graph search with learned guidance and anytime planning.

graph-algorithms / observability / optimization / planning / protocol

document-intelligence

docendo: grounded RAG agent for RBI regulatory documents. Local SQLite + Qwen3 embeddings, citation-grounded answers, atomic-claim hallucination evaluation.

ai-agents / document-ai / extraction / llm / policy

driftflow

Information-Geometric Adaptive Sampling for Graph Diffusion

adaptive-sampling / dvs / generative-model / graph-diffusion / pure-python

factorise

High-performance prime factorisation in pure Python.

library / number-theory / prime-factorization / python / sdk

find

Sub-second secp256k1 scalar discovery via 512-variant range-splitting and Montgomery batch inversion. Educational/research-only; for cryptographic pedagogy and high-performance Rust study.

bitcoin / cryptography / education / hashing / rust

fleetpilot

FleetPilot — ALNS + BRKGA metaheuristics for time-windowed pickup-and-delivery VRP with multi-depot and traffic-aware routing.

alns / brkga / javascript / logistics / metaheuristic

gethired

Production-grade multi-agent CV tailoring system grounded in a master resume and verified against ATS gates.

ai-agents / anthropic / ats / career / cli

gradient-regularized-newton-boosting-trees

Pure-NumPy reproduction of Zozoulenko et al. (2026) Gradient Regularized Newton Boosting Trees. Implements Vanilla + GRN + multi-class boosting, four analytic losses, and Hilbert-space diagnostics for paper-fidelity experiments.

boosting / gbdt / gradient-boosting / machine-learning / newton-method

greedy-find

Metal-accelerated secp256k1 private-key discovery using multi-variant range-splitting on Apple Silicon. Sibling of the Rust find crate. Educational and research use only.

apple-silicon / authentication / bitcoin / cryptography / education

hone

JSONL data preparation for Soup-driven fine-tuning on Apple Silicon.

apple-silicon / data-preparation / fine-tuning / huggingface / jsonl

hurstify

RK-SAVR (Angelini & Bianchi 2025, arXiv:2509.20015v3) estimator of the Hurst parameter of log-volatility series. Pure JavaScript, zero runtime dependencies, deterministic.

finance / fractal-analysis / javascript / nextjs / statistics

intelliqx

Autonomous QA Intelligence Platform — agent-native test automation for software teams.

agents / knowledge-graph / llm / monorepo / okf

kernos

Refresh-aware hybrid continuous-discrete low-rank kernel learning for streaming, large-scale regression.

adaptive-kernels / benchmark / kernel-methods / machine-learning / nystrom

knowledge

SDK + CLI for LLM-powered OKF knowledge-bundle creation.

bundles / code-formatter / graph-algorithms / knowledge-base / llm

kvcompress

Universal plug-and-play KV cache compression for decoder-only LLMs (JoLT, Flash, INT/FP8, LowRank). Hugging Face + vLLM integration.

authentication / caching / compression / inference / kv-cache

laker

PyTorch kernel-regression library for spectrum cartography.

attention-kernel / cccp / conjugate-gradient / convex-concave-procedure / kernel-regression

magic

Strands Agents-powered web app for true agentic software engineering on any repo

agentic / ai-agents / code-review / developer-tools / llm

membrane

LLM inference server with multi-backend routing and persistence.

caching / cryptography / hashing / in-memory / inference-server

methodos

Public source code. The owner has not provided a repository description.

morel

Bandit and recommender research library with masked submodels.

bandit / graph-algorithms / python / recommender-system / research

nonce-cracker

High-speed parallel ECDSA private key recovery for secp256k1 using a single-signature affine relation attack.

bitcoin / cryptography / ecdsa / rust / scheduling

optimising-cryptocurrency-portfolios

Build, backtest, and serve Sharpe-optimised cryptocurrency portfolios using consensus Louvain clustering. Includes ARIMA / GARCH / LSTM forecasting, yfinance + ccxt data, a stateless REST API, and Docker.

consensus-clustering / correlation-network / cryptocurrency / garch / ledoit-wolf

pjepa

Persistent-JEPA: a persistent graph world model for continual developmental learning.

continual-learning / developmental-ai / geometric-deep-learning / graph-classification / graph-neural-networks

promptsheon

Git-native version control for AI agents — content-addressable prompt management, evaluation engine, maker-checker approvals — built in TypeScript (Fastify + Next.js).

ai-agents / authentication / developer-tools / graph-algorithms / javascript

reachq

Parallel reachability and shortest paths on non-sparse digraphs.

graph-algorithms / python / reachability / research / shortest-path

redax

Self-hosted PII redaction engine for use before a self-hosted LLM.

fastapi / gliner2 / llm / ner / pii

regulo

Adaptive Norm-Based Regularization for Neural Networks — pure NumPy/SciPy reproduction of Qasim & Javed (2024)

machine-learning / neural-networks / numpy / python / regularization

revex

Hybrid retrieval (vector + BM25 + graph) for TypeScript.

ai-agents / authentication / authorization / graph-algorithms / hybrid-search

sachncs

Applied AI Architect & Engineer — production-grade AI systems: agents, RAG, evaluation, and the integrations that make them stick. Selected OSS work below.

agents / ai / ai-agents / ai-engineering / applied-ai

tensift

Tensor-network Schnorr's sieving for secp256k1 (Rust).

algebra / cryptography / discrete-log / number-theory / rust

traceprojector

Pure-JS Π^l de Rham projector for combinatorial topology.

computational-geometry / de-rham-complex / discrete-operators / finite-element-method / hodge-star

trivium

Hybrid search benchmark for BM25 + dense vector + cross-encoder rerank on distractor-augmented BEIR scifact.

beir / benchmark / bm25 / cross-encoder / faiss

tsn-affinity

Continual offline reinforcement learning with training-aware sparse networks and affinity routing.

ai / continual-learning / decision-transformer / machine-learning / offline-rl

ucsa

Unified Cognitive State Architecture: a research-grade foundation model whose computation orbits a single persistent differentiable cognitive state.

cognitive-architecture / foundation-model / foundation-models / in-memory / jepa

underwrite

Event-driven nano-service platform for Indian retail lending, DPDPA/RBI compliant with Ed25519 cryptographic attestation.

authorization / cryptography / docker / fintech / graph-algorithms

xaker

Fused Exclusive Self Attention (XSA) + Kernel Ridge Regression for PyTorch Transformers.

attention / code-formatter / deep-learning / kernel-methods / kernel-ridge-regression