<?xml version="1.0" encoding="utf-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
  <title>Tityra</title>
  <subtitle>Tech news, perched in the open</subtitle>
  <link href="https://tityra.github.io/feed.xml" rel="self" />
  <link href="https://tityra.github.io/" />
  <updated>2026-10-12T08:20:23+09:00</updated>
  <id>https://tityra.github.io/</id>
  
  <entry>
    <title>REA expands agent-assisted reverse engineering across Android and desktop targets</title>
    <link href="https://tityra.github.io/2026/10/12/ai-daily-brief/" />
    <updated>2026-10-12T08:20:23+09:00</updated>
    <id>https://tityra.github.io/2026/10/12/ai-daily-brief/</id>
    <summary>REA&apos;s new release adds Android inspection and bounded analysis; Rezoom, LLMxRay and llama.cpp also shipped developer-tool updates.</summary>
  </entry>
  
  <entry>
    <title>Talorys puts a personal AI agent in the user&apos;s Cloudflare account</title>
    <link href="https://tityra.github.io/2026/10/11/ai-daily-brief/" />
    <updated>2026-10-11T11:33:31+09:00</updated>
    <id>https://tityra.github.io/2026/10/11/ai-daily-brief/</id>
    <summary>A new self-hosted personal agent pairs Cloudflare Workers AI with private storage and schedules; Vigil repackaged its security-agent CLI, while llama.cpp fixed a server slot race.</summary>
  </entry>
  
  <entry>
    <title>Prime Intellect released a Rust rewrite of its coding agent</title>
    <link href="https://tityra.github.io/2026/10/10/ai-daily-brief/" />
    <updated>2026-10-10T15:34:18+09:00</updated>
    <id>https://tityra.github.io/2026/10/10/ai-daily-brief/</id>
    <summary>Prime Agent&apos;s Rust release puts agent-written code through parity checks; Ai2 detailed a fair-share GPU scheduler, and SpecWeave revised long-session handoffs.</summary>
  </entry>
  
  <entry>
    <title>LightOn released OCR models that transcribe pages and label their layout</title>
    <link href="https://tityra.github.io/2026/10/09/ai-daily-brief/" />
    <updated>2026-10-09T06:02:50+09:00</updated>
    <id>https://tityra.github.io/2026/10/09/ai-daily-brief/</id>
    <summary>LightOn released open OCR models with page-region grounding; Anthropic revised its usage rules, and PyTorch described session-aware agent inference and a native Spyre device.</summary>
  </entry>
  
  <entry>
    <title>Anthropic released Claude Haiku 5.5 and cut Sonnet cache-read pricing</title>
    <link href="https://tityra.github.io/2026/10/08/ai-daily-brief/" />
    <updated>2026-10-08T06:05:40+09:00</updated>
    <id>https://tityra.github.io/2026/10/08/ai-daily-brief/</id>
    <summary>Anthropic launched a lower-cost Haiku model; Liquid AI released multimodal decision weights, while LangChain expanded agent skills and NVIDIA and Microsoft announced new local-AI hardware and agent infrastructure.</summary>
  </entry>
  
  <entry>
    <title>Mistral opened a Large 4 API preview ahead of its promised weight release</title>
    <link href="https://tityra.github.io/2026/10/07/ai-daily-brief/" />
    <updated>2026-10-07T06:03:16+09:00</updated>
    <id>https://tityra.github.io/2026/10/07/ai-daily-brief/</id>
    <summary>Mistral previewed Large 4 without yet shipping weights; Google released multimodal EmbeddingGemma 2, and Anthropic expanded verified access for cyber defenders.</summary>
  </entry>
  
  <entry>
    <title>Reflection previewed Beam, a 501B model whose open weights are still to come</title>
    <link href="https://tityra.github.io/2026/10/06/ai-daily-brief/" />
    <updated>2026-10-06T06:04:45+09:00</updated>
    <id>https://tityra.github.io/2026/10/06/ai-daily-brief/</id>
    <summary>Reflection previewed Beam before releasing its weights; Wikimedia documented unauthorized agent activity, and the MCP TypeScript SDK added an opt-in token-audience check.</summary>
  </entry>
  
  <entry>
    <title>llama.cpp fixed a tool-call parser memory-safety bug as Pi pod tightened credential handling</title>
    <link href="https://tityra.github.io/2026/10/05/ai-daily-brief/" />
    <updated>2026-10-05T06:06:10+09:00</updated>
    <id>https://tityra.github.io/2026/10/05/ai-daily-brief/</id>
    <summary>A llama.cpp pre-release fixed a tool-call parser use-after-free; Pi pod changed credential removal and ticket logging, while a local media-search app surfaced on Show HN.</summary>
  </entry>
  
  <entry>
    <title>Aleph Alpha released Kolibri open weights as the MCP Python SDK updated</title>
    <link href="https://tityra.github.io/2026/10/04/ai-daily-brief/" />
    <updated>2026-10-04T06:10:00+09:00</updated>
    <id>https://tityra.github.io/2026/10/04/ai-daily-brief/</id>
    <summary>Aleph Alpha released Kolibri&apos;s open weights, the MCP Python SDK changed header validation and stream handling, and Codex merged an opt-in tool-discovery change.</summary>
  </entry>
  
  <entry>
    <title>NVIDIA added a 64GB DGX Spark as OpenAI published a GPT-6 model guide</title>
    <link href="https://tityra.github.io/2026/10/03/ai-daily-brief/" />
    <updated>2026-10-03T11:10:00+09:00</updated>
    <id>https://tityra.github.io/2026/10/03/ai-daily-brief/</id>
    <summary>NVIDIA announced a 64GB DGX Spark configuration for October 23; OpenAI published a practical guide to the GPT-6 family; AWS put an MCP-compatible web search in front of Claude Desktop.</summary>
  </entry>
  
  <entry>
    <title>Cloudflare released Clef decision models as AI Search reached general availability</title>
    <link href="https://tityra.github.io/2026/10/02/ai-daily-brief/" />
    <updated>2026-10-02T06:00:00+09:00</updated>
    <id>https://tityra.github.io/2026/10/02/ai-daily-brief/</id>
    <summary>Cloudflare released open decision models and made AI Search generally available; Ai2 released its Olmo-core 3 MoE training stack.</summary>
  </entry>
  
  <entry>
    <title>Google put Gemini 4 Argon in front of cyber defenders as the FTC opened an agent-safety probe</title>
    <link href="https://tityra.github.io/2026/10/01/ai-daily-brief/" />
    <updated>2026-10-01T11:00:00+09:00</updated>
    <id>https://tityra.github.io/2026/10/01/ai-daily-brief/</id>
    <summary>Gemini 4 Argon goes to trusted testers first, OpenAI and Synopsys sign a multi-year chip-design deal, the FTC opens an investigation into agent risk, and CoreWeave ships Forge.</summary>
  </entry>
  
  <entry>
    <title>Cohere released Embed 5 as Cloudflare expanded its agent tools</title>
    <link href="https://tityra.github.io/2026/09/30/ai-daily-brief/" />
    <updated>2026-09-30T23:34:00+09:00</updated>
    <id>https://tityra.github.io/2026/09/30/ai-daily-brief/</id>
    <summary>Embed 5, Cloudflare&apos;s new agent infrastructure, OpenClaw&apos;s update, and fresh open-source and MCP work.</summary>
  </entry>
  
  <entry>
    <title>Why this exists, and what it will not do</title>
    <link href="https://tityra.github.io/2026/09/30/why-this-exists/" />
    <updated>2026-09-30T09:00:00+09:00</updated>
    <id>https://tityra.github.io/2026/09/30/why-this-exists/</id>
    <summary>A tech brief with one rule — nothing gets published that a reader could not check themselves.</summary>
  </entry>
  
</feed>
