Category / T-CAT / SOFTWARE
Software
Software / T-2026-5442
Crynta's Terax: The 7MB Terminal That Rethinks AI Dev Tools
Crynta's Terax is a 7MB terminal-first AI dev workspace built on Tauri 2 and Rust. It runs local models, offers agentic workflows, and ships no telemetry.
Sourcecrynta/terax-ai
Software / T-2026-6025
llama.cpp b9870: The StepFun trim that fixes long reasoning loops
llama.cpp b9870 patches a whitespace issue in the StepFun parser that caused infinite reasoning loops. The fix shows how the open-source AI stack is maturing.
Sourceggerganov/llama.cpp b9870
Software / T-2026-4550
Anthropic's SDK adds agent memory header, hinting at a product launch
Anthropic's Python SDK v0.116.0 adds a beta header for agent memory, signaling a product launch on July 22. A look at what it means for AI agents and builders.
Sourceanthropics/anthropic-sdk-python v0.116.0
Software / T-2026-8414
Xiaomi's MiMo Code open-sources a memory system that beats Claude Code on long tasks
Xiaomi open-sources MiMo Code V0.1.0, a terminal AI coding assistant that beats Claude Code on long-horizon tasks with a cross-session memory system.
SourceXiaomi's new open source, agentic AI coding harness MiMo Code beats ...
Software / T-2026-5908
llama.cpp b9837 adds a reasoning-preserve flag, and that matters more than it sounds
llama.cpp b9837 ships a `--reasoning-preserve` flag for chat templates. The feature is small. The problem it solves is not.
Sourceggerganov/llama.cpp b9837
Software / T-2026-7289
Strix turns LLM agents into autonomous pentesters
Strix is an open-source platform that uses autonomous AI agents to dynamically find and validate software vulnerabilities, offering a developer-first CLI and CI/CD integration.
Sourceusestrix/strix
Software / T-2026-9579
OpenSpec proposes a spec layer for AI coding, but the hard part is still the model
Fission-AI's OpenSpec adds a spec layer to AI coding workflows. The idea is sound, but execution depends on models it cannot control.
SourceFission-AI/OpenSpec
Software / T-2026-8437
AWS's Agent Toolkit gives coding agents an official backstage pass
AWS releases an official toolkit bundling MCP servers, curated skills, and IAM guardrails for AI coding agents. The move signals a platform play for the agent middleware layer.
Sourceaws/agent-toolkit-for-aws
Software / T-2026-5871
llama.cpp b9821: The Quiet Infrastructure Milestone
llama.cpp b9821 ships 27 platform builds across CPU, GPU, and specialized backends, marking a shift from experimental project to production-grade AI runtime.
Sourceggerganov/llama.cpp b9821
Software / T-2026-6381
Alibaba's Page Agent runs GUI automation from a single script tag
Alibaba open-sources Page Agent, an in-page JavaScript GUI agent that manipulates the DOM directly with natural language, no screenshots or browser extensions needed.
Sourcealibaba/page-agent
Software / T-2026-5811
llama.cpp b9803: The quiet infrastructure of local AI
llama.cpp b9803 ships with a minor OpenCL fix and 27 platform binaries. The release tells a bigger story about where local AI is headed.
Sourceggerganov/llama.cpp b9803
Software / T-2026-0648
Stably AI's Orca turns the agent IDE into a fleet command center
Stably AI's open-source Orca lets developers run five coding agents in parallel, each in its own worktree, and merge the best result. A sign of how agentic coding is maturing.
Sourcestablyai/orca