01Log
Archive
The full log, newest first.
OSWorld-Science: A Benchmark of Computer Use Agents for Learning and Using Scientific Software
Extends OSWorld to scientific software โ tests whether computer-use agents can learn and operate domain tools, not just navigate generic UIs. Scientific software presents a demanding test.
Show HN: Made an open-source Lego AI generator
Show HN: open-source Lego AI generator โ uses ChatGPT/Claude to emit LDraw assembly code for brick models. 134 points, 49 comments; a fun stress test of LLM codegen for a constrained DSL.
Copilot Selected models in GitHub deprecated
GitHub Copilot deprecates selected models across all experiences as of Oct 2 โ check if your team's pinned models are affected and migrate.
Devin September 30, 2026
Devin fixes message loss on connection drops โ messages sent right before a disconnect are now saved server-side. Also read-only shell preview and session preview cards.
OpenAPPA Guardrails
OpenAPPA is a deterministic guardrail layer that sits between an AI agent and its tools to enforce permission policies on data flows. It is notable for using a formal policy algebra (APPA) to make agent permissions auditable, though validation is still early.
Moria IoT Identification
Moria is a tool for identifying and extracting filesystems, kernels, and other structures embedded in IoT firmware without root privileges. It is notable for its recursive unpacking of many filesystem types and clean JSON output designed for scripting and LLM agents.
Mithril IoT Scanner
mithril is a static scanner for IoT firmware contents that detects secrets, builds SBOMs, finds CVEs, checks licenses, and assesses boot-security posture. It is notable for reporting evidence and confidence per finding and for its JSON-first design aimed at both humans and LLM agents.
Lagune Security Copilot
Lagune is a security copilot that points an AI agent at a codebase to guide developers and auditors through relevant security work without requiring an API key. It is notable for supporting 73 agents and any programming language, though it remains early-stage with low traction.
PCYBox Attack Graph
pcybox-attackgraph is a browser-based lab for building simulated infrastructure and observing attack paths, with explanations of why each step succeeds and which defensive controls would break it. It is notable as an educational and threat-modeling tool that runs entirely in the browser.
Strypt Metadata Remover
strypt is an offline Rust desktop and CLI tool that removes metadata such as GPS coordinates, serial numbers, author names, and timestamps from photos, PDFs, Office documents, audio, and video. It is notable for its memory-safe implementation and strict no-network code paths.
Synapse CE Control
Synapse CE is a governed control plane for the security assessment lifecycle, covering SCA, SAST, secrets scanning, recon, cloud posture, and blue-team agents. It is notable for combining server-side scope enforcement, tamper-evident evidence, and deterministic reporting into a single auditable workflow.
Tincan CLI
A serverless peer-to-peer voice and text chat CLI for terminals. Notable for using Iroh to create rooms without servers, VPNs, or accounts, making it useful for remote developers.
Issue Graph
A CLI that traces linked GitHub issues and pull requests to find related work before starting. Notable for helping with triage and agent workflows, backed by Vercel Labs.
Suiko Japanese Diagnostics
A Rust CLI for deterministic diagnostics of Japanese text naturalness and readability. Notable for its reproducible, local-only analysis with embedded dictionaries, useful for non-native writers.
Runyte Workspace
A modal text editor combined with terminals, files, and Git in a single terminal workspace. Notable as an ambitious TUI aiming to unify development tools in one persistent session.
Bast SSH Manager
A terminal tool for browsing SSH hosts, syncing cloud VMs, managing keys, and connecting via SFTP. Notable as a fast, fresh alternative to Termius for cloud-heavy developers.
k10s Kubernetes TUI
A clickable Kubernetes TUI with instant search, logs, exec, port-forwarding, themes, and context-aware AI. Notable as a practical, feature-rich alternative to k9s, distributed as a single Go binary.
GhostHub Terminal
A Ghostty-native terminal multiplexer for local and remote tmux, Herdr, and Zellij sessions with Git worktree management. Notable for integrating multiple multiplexers across SSH hosts in one interface.
Cadence Learning System
Cadence is a brain-inspired AI framework that learns continuously from live experience, combining perception, memory, plasticity, and action. It models a simulated human-like brain from simplified biological mechanisms rather than using feed-forward deep neural networks.
Metriplane Robotics Observability
Metriplane provides deterministic replay and verifiable incident evidence for recorded robotics workcell runs. It generates human-readable reports and regression tests from timestamped object positions and process rules, enabling engineers to rerun cases after software changes.
RL Token Pipeline
RL-Token-Pi05-open provides a pipeline for fine-tuning the ฯ0.5 VLA, training an RL Token, and deploying online reinforcement learning on the SO-101 robot arm. It includes a critical-phase classifier and step-by-step hardware, recording, training, and inference workflows.
FLUX Action Model
FLUX 3 Action is an open-weights 7B world action model from Black Forest Labs. It takes camera frames, robot state, and text instructions to predict actions and future video frames for robots, simulators, and games.
HOI-Retarget System
HOI-Retarget retargets human-object interaction clips into humanoid robot trajectories that reproduce the interaction, not just poses. It uses contact-centric trajectory optimization to ensure the robot grasps the same object locations at any scale.
XGO-Duck Robot
XGO-Duck is a 3D-printable bipedal robot duck built around an Arduino UNO Q and 15 serial servos, adapted from Pollen Robotics' Microduck. It includes mechanical models, PCB designs, a BOM, and an assembly guide for building and exploring biped motion.
Open Reality AI
Open Reality converts phone video into persistent, AI-queryable 3D scenes with measurement, path planning, and robot-training exports. It provides MCP tools for Claude, Codex, and Cursor, plus a self-hostable broker and the VGGT-SLAM library.
JevSpawn: Adaptive Agentic Inference through Compositional Action Spaces
Training-free approach to Jev-style agentic inference โ composes action spaces so agents adapt without retraining. Worth a look if you're tuning agent action policies.
EgoTools: Towards Tool-Centric Reasoning in Real-World Egocentric Videos
Benchmark for tool-centric reasoning in egocentric video โ tests whether agents can reason about tool use from first-person real-world footage.
Kiro โ Claude Sonnet 5.5 Now Available
Kiro adds Claude Sonnet 5.5 (experimental) across IDE, CLI, Crew, and Web for Pro+ in us-east-1 and eu-central-1 โ Anthropic's fastest Sonnet.
Copilot GitHub in VS Code, September 2026 releases
Copilot in VS Code September release: Agents window, workspaces and environments, chat and integrations. The Agents window is the operator-facing piece.
Hermes Agent โ Pinned inputs a
Hermes Agent ships pinned inputs โ a way to pin context inputs across sessions, aimed at long-running agent workflows.
Ask HN: Is anybody producing good code with coding agents?
Senior engineers are asking whether anyone is actually producing good code with coding agents โ 27 comments comparing Claude, AGY, Copilot, and Codex, and the through-line is exhaustion with review overhead.
VisionHOPE
MIT-licensed image classification model, no parameter count published.
Clef Flash
9.4B open image-text-to-text model under Apache 2.0, small enough to run locally while handling multimodal input.
GEV 26B Decide
A decision specialist that classifies instead of generating โ judgment calls without a full LLM pass.
Kolibri 1
Open-weight German-English MoE built for sovereign, regulated EU deployments rather than raw scale.
JEV 27B VL
A mid-size vision-language model for builders who need image understanding inside an open stack, with no API dependency.
KaliBench: A Fine-Grained Benchmark for Cybersecurity Tool Use on Kali Linux with Runtime-Free Verifiable Rewards
Benchmark for agents driving Kali Linux security tools, with verifiable rewards that don't require running the tools โ lets you evaluate offensive-security agents without sandbox risk.
An RL View of OPD: Least Square Policy Distillation for Sample-Efficient LLM Reasoning
Least-square policy distillation for sample-efficient LLM reasoning โ a cheaper route to reasoning models if you're fine-tuning agent planners.
Measuring the Microtask Eligibility Gap: When Is an Off-the-Shelf SLM Enough for an Agent Harness?
Benchmarks when an off-the-shelf SLM can handle agent-harness microtasks โ a direct answer to where you can swap frontier models for cheap local ones.
Svelte Framework
Svelte is a compiler that converts declarative components into efficient JavaScript that updates the DOM surgically. It is notable for its compiler-based approach to UI development, offering a novel alternative to runtime frameworks.