Tityra
Tityra
Tech news, perched in the open. Every claim carries its source.
REA expands agent-assisted reverse engineering across Android and desktop targets
REA's new release adds Android inspection and bounded analysis; Rezoom, LLMxRay and llama.cpp also shipped developer-tool updates.
-
Talorys puts a personal AI agent in the user's Cloudflare account
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.
-
Prime Intellect released a Rust rewrite of its coding agent
Prime Agent's Rust release puts agent-written code through parity checks; Ai2 detailed a fair-share GPU scheduler, and SpecWeave revised long-session handoffs.
-
LightOn released OCR models that transcribe pages and label their layout
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.
-
Anthropic released Claude Haiku 5.5 and cut Sonnet cache-read pricing
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.
-
Mistral opened a Large 4 API preview ahead of its promised weight release
Mistral previewed Large 4 without yet shipping weights; Google released multimodal EmbeddingGemma 2, and Anthropic expanded verified access for cyber defenders.
-
Reflection previewed Beam, a 501B model whose open weights are still to come
Reflection previewed Beam before releasing its weights; Wikimedia documented unauthorized agent activity, and the MCP TypeScript SDK added an opt-in token-audience check.
-
llama.cpp fixed a tool-call parser memory-safety bug as Pi pod tightened credential handling
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.
-
Aleph Alpha released Kolibri open weights as the MCP Python SDK updated
Aleph Alpha released Kolibri's open weights, the MCP Python SDK changed header validation and stream handling, and Codex merged an opt-in tool-discovery change.
-
NVIDIA added a 64GB DGX Spark as OpenAI published a GPT-6 model guide
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.
-
Cloudflare released Clef decision models as AI Search reached general availability
Cloudflare released open decision models and made AI Search generally available; Ai2 released its Olmo-core 3 MoE training stack.
-
Google put Gemini 4 Argon in front of cyber defenders as the FTC opened an agent-safety probe
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.
-
Cohere released Embed 5 as Cloudflare expanded its agent tools
Embed 5, Cloudflare's new agent infrastructure, OpenClaw's update, and fresh open-source and MCP work.
-
Why this exists, and what it will not do
A tech brief with one rule — nothing gets published that a reader could not check themselves.