Tools – Tokenstrom

Tools Oct 24, 2026GitHub (neue Projekte)6 min

Survival LLM: offline AI with an emergency library and cited sources

Survival LLM is a free desktop app for Mac and Windows that pairs a local AI model with roughly 1.5 million emergency and how-to articles. It answers questions about first aid, power outages or water purification without internet, and shows the sources for every answer.

Survival LLM is a free, open-source desktop app (MIT license) that combines an AI assistant and a survival library on your computer. After setup, it works without internet. You can ask, for example, how to keep food safe during a power outage, how to treat a burn or how to purify water. The app searches a local library, answers from what it finds and shows the passage the answer comes from. There are versions for …

The app runs entirely offline on your own computer, your questions never leave it; code is MIT-licensed.
The full library has about 1.5 million articles (8.8 GB download); the Essentials version has 500,000 (1.3 GB).
Depending on your RAM, the app picks a model: from Granite 4.2 8B (16–31 GB) up to Nemotron 3.5 Super (Mac with 128 GB or more, 80 GB download).

Tools Oct 28, 2026r/StableDiffusion5 min

Anima Ref Adapter: Generate a Character from One Reference Image in ComfyUI

A new adapter for the anime image model Anima picks up a character's look, and partly their style, from a single reference image. Pose, background and composition still come from your prompt. It works for original characters too, with no lengthy training or descriptions.

Under the name "Anima Reference Adapter," the developer "ElectronGirl" has released a ComfyUI extension that teaches the anime image model Anima to work with a character reference. You supply a single reference image and write a prompt that describes only the scene. The adapter keeps *who* is drawn (hair, eyes, clothing, accessories) and some of *how* it is drawn. Pose, background and composition come from the …

The adapter by "ElectronGirl" is a rank-32 LoRA with 45.9 million parameters for the Anima model (Base v1.0) and runs in ComfyUI.
One reference image is enough: hair, eyes, clothing and accessories carry over, while pose, background and composition follow the prompt.
It was trained only as a pilot, on 300 characters and 2,000 steps, using exclusively images generated by Anima itself.

GitHub open-sources Replica: AI prototyping with your own design system

GitHub Design has released Replica as an open-source project under the MIT license. The tool lets designers, product managers and developers build interactive prototypes by describing them in plain language in the GitHub Copilot app, using their product's existing components.

On October 28, 2026, GitHub released Replica as an open-source project. The tool comes from GitHub Design, the company's design team. It's meant to help turn ideas into interactive prototypes quickly, meaning clickable mockups of a software interface. The code is under the MIT license, which allows free use and modification. Replica is not an AI model with downloadable weights, but a tool that uses AI through the …

Replica is an AI-powered prototyping tool from GitHub Design and is available as an open-source project under the MIT license.
Prototypes are created in natural language in the GitHub Copilot app.
Replica uses the team's existing components, patterns and documentation, which designers set as guidelines.

Tools Oct 28, 2026ITmedia AI+ (JP)3 min

Don't launch it in Downloads: Tampered AGENTS.md hijacks coding agents

Security firm Rebora warns of attacks where a malicious AGENTS.md in your Downloads folder tricks coding agents like Codex or Claude Code into harmful actions. In its tests, Codex followed the instructions more than 90% of the time, Claude Code about 50%. Rebora advises never starting an agent in the Downloads folder.

The Israeli security firm Rebora warns against launching AI coding agents in your Downloads folder. According to a blog post from October 22 (local time), the company found a malicious file named "AGENTS.md" on a customer's machine. This is a Markdown text file containing instructions for AI agents, similar to "CLAUDE.md" in Claude Code. ITmedia AI+ reported on the findings.

The Israeli security firm Rebora found a malicious AGENTS.md in a customer's Downloads folder.
Websites can sometimes drop files into the Downloads folder without the user doing anything; according to Rebora, attackers exploit this.
If launched in such a folder, the agent can rewrite its own configuration, install malware or send its login token to third parties.

Tools Oct 26, 2026GitHub Trending4 min

NVIDIA's SkillSpector: an open-source security scanner for AI agent skills

NVIDIA has released SkillSpector, an open-source tool that checks skills for AI agents such as Claude Code or Codex for malicious code, prompt injection and data exfiltration before you install them. NVIDIA says more than a quarter of the skills it examined contain vulnerabilities, and just over 5% look malicious.

SkillSpector is a security scanner from NVIDIA for agent skills. These are add-on packages (usually a SKILL.md with instructions and sometimes scripts) that give AI coding agents like Claude Code, Codex CLI or Gemini CLI new abilities. According to the README, such skills run on "implicit trust" with little vetting. The tool is meant to answer one question: "Is this skill safe to install?" It's open source …

SkillSpector is an Apache-2.0-licensed Python scanner (3.12+) from NVIDIA for agent skills, for example for Claude Code, Codex CLI and Gemini CLI.
In a subset of 31,132 analyzed skills, 26.1% had vulnerabilities and 5.2% showed likely malicious intent, according to the README.
The scanner checks 71 vulnerability patterns in 17 categories, including prompt injection, data exfiltration, supply chain and MCP tool poisoning.

Tools Oct 25, 2026Bing News (es-ES)5 min

Microsoft's Agensh Lets 1,024 AI Agents Work Together Without a Boss

Microsoft Research has open-sourced Agensh, a system in which up to 1,024 AI agents work on one code repository with no central coordinator. In tests on hard programming tasks, the success rate rose with the number of agents, though with diminishing returns.

Microsoft Research has open-sourced Agensh. In tests, the system coordinated up to 1,024 AI agents working on the same code repository at the same time, with no lead agent handing out tasks. A *harness* is the runtime environment around a language model: it supplies tools, memory and context management. Agensh is a harness built for many agents at once. Its core idea is to drop the orchestrator (the coordinator). …

Agensh is a multi-agent harness from Microsoft Research, released under the MIT license, that works without a central orchestrator.
It was tested on the five hardest ProgramBench tasks (FFmpeg, GROMACS, pandoc, PHP-src, ctags) with runs of up to six hours.
With GPT-5.6-sol, the mean test pass rate rose from 19.31% (1 agent) to 28.78% (128 agents).