Category / T-CAT / SOFTWARE
Software
Software / T-2026-5633
llama.cpp b10375 ships a quiet fix that says a lot about open-source AI
llama.cpp b10375 tightens Qwen function parsing. The small fix shows how open-source AI actually matures.
Sourceggerganov/llama.cpp b10375
Software / T-2026-5599
llama.cpp b10362: HIP's missing CUB exposes the cost of backend sprawl
llama.cpp b10362 skips tests on HIP. The real news is the growing cost of backend sprawl.
Sourceggerganov/llama.cpp b10362
Software / T-2026-3243
OpenAI's Python SDK drops httpx, and the ecosystem should feel it
OpenAI's openai-python v3.0.0 makes HTTPX2 the default HTTP client, dropping automatic httpx installation. A breaking change with ecosystem-wide implications.
Sourceopenai/openai-python v3.0.0
Software / T-2026-5572
llama.cpp b10356 drops ROCm 7.2.1 for 7.14, and the AMD inference story gets sharper
llama.cpp b10356 moves to ROCm 7.14, AMD's first TheRock-based release. What that means for local AI builders.
Sourceggerganov/llama.cpp b10356
Software / T-2026-6056
Paperclip turns AI agents into a company org chart with budgets and governance
Paperclip is open-source software that turns teams of AI agents into a company with org charts, budgets, and governance.
Sourcepaperclipai/paperclip
Software / T-2026-5507
llama.cpp b10333: A one-line fix that shows the real cost of AI fragmentation
llama.cpp b10333 patches a missing Q5_0 dispatch for SpaceMiT. The trivial fix reveals the fragmentation tax on open-source AI.
Sourceggerganov/llama.cpp b10333
Software / T-2026-2249
Harvey's open-source LAB benchmark puts legal agents on the stand
Harvey open-sources LAB, a legal agent benchmark with all-pass rubric scoring. The hard part is the rubrics.
Sourceharveyai/harvey-labs
Software / T-2026-6758
Semantica bets that AI accountability is a graph problem, not a model problem
Semantica's open-source Context Graph layer gives agents auditable decisions and causal reasoning, no LLM required.
Sourcesemantica-agi/semantica
Software / T-2026-5479
llama.cpp b10326 fixes TTS timing honesty, and that is the real story
llama.cpp b10326 adjusts TTS timing to count vocoder work. Small fix, big signal about local AI honesty.
Sourceggerganov/llama.cpp b10326
Software / T-2026-2237
Prime Agent treats context as code: the RLM pattern behind PrimeIntellect's new open-source agent
PrimeIntellect's open-source Prime Agent makes context a variable and memory a codebase.
SourcePrimeIntellect-ai/prime-agent
Software / T-2026-4252
Cloudflare Computer: the agent filesystem that makes state boring again
Cloudflare's Computer preview gives agents a SQLite-backed virtual filesystem, with three execution backends.
Sourcecloudflare/computer
Software / T-2026-5233
LoopX puts a durable state kernel between your agents and their drift
LoopX adds durable goals, quota, and evidence to Codex and Claude Code loops. The control plane stays local and human-gated.
Sourcehuangruiteng/loopx