Category / T-CAT / SOFTWARE
Software
Software / T-2026-4488
Anthropic SDK v0.118.0: Managed Agents and the streaming delta pattern
Anthropic's SDK update adds managed agent effort controls, session events, and a new streaming pattern that hints at a platform play.
Sourceanthropics/anthropic-sdk-python v0.118.0
Software / T-2026-4518
Anthropic SDK patch adds a new refusal category: what it tells us about model safety
Anthropic's SDK v0.117.1 adds a new refusal category. The small patch signals a bigger shift in how frontier labs classify and audit model refusals.
Sourceanthropics/anthropic-sdk-python v0.117.1
Software / T-2026-9055
Moonshine Voice beats Whisper on latency with 245M parameters, not 1.5B
Moonshine Voice is an open-source speech toolkit that claims 5x faster streaming transcription than Whisper, using 245M parameters vs 1.5B, challenging OpenAI's dominance for…
Sourcemoonshine-ai/moonshine
Software / T-2026-5478
G0DM0D3: A red-teaming toolkit that treats jailbreak as engineering
Elder Plinius releases G0DM0D3, an open-source multi-model chat interface with 60 providers, 33 perturbation techniques, and a philosophy of 'cognition without control'
Sourceelder-plinius/G0DM0D3
Software / T-2026-7040
MoonshotAI's Kimi CLI: The terminal agent that skips the hype
MoonshotAI's kimi-cli is an open-source terminal agent that combines shell command mode, VS Code integration, and MCP support. It is a pragmatic take on AI-assisted development.
SourceMoonshotAI/kimi-cli
Software / T-2026-2969
Code Review Graph: A local-first intelligence graph that cuts AI token waste 82x
A new open-source tool uses Tree-sitter to build a persistent code graph, cutting AI coding tool token consumption by a median 82x on review tasks.
Sourcetirth8205/code-review-graph
Software / T-2026-4519
Anthropic SDK adds 'dreaming' and MCP Tunnels
Anthropic's Python SDK v0.117.0 adds support for 'dreaming' and MCP Tunnels, signaling shifts in model self-correction and agent tooling.
Sourceanthropics/anthropic-sdk-python v0.117.0
Software / T-2026-7744
Open Interpreter rewrites itself in Rust to cut agent costs
Open Interpreter rewrites in Rust to optimize coding agents for low-cost models, emulating harnesses from Claude Code to Qwen Code.
Sourceopeninterpreter/openinterpreter
Software / T-2026-2561
llama.cpp b10012 ships 27 platform builds, a quiet snapshot of AI fragmentation
llama.cpp b10012 ships 27 platform-specific binaries. The release is a quiet snapshot of AI inference fragmentation across hardware backends.
Sourceggerganov/llama.cpp b10012
Software / T-2026-3616
The 18-Chapter AI Compendium That Got Friends Into DeepMind and OpenAI
An open-source textbook covering maths, CS, and AI from the ground up, with an MCP server for AI assistants.
SourceHenryNdubuaku/maths-cs-ai-compendium
Software / T-2026-5339
Nutlope/hallmark: a design skill that refuses to look AI-generated
A design skill for Claude Code, Cursor, and Codex that actively rejects the visual fingerprints of LLM training data.
SourceNutlope/hallmark
Software / T-2026-6372
Destructive Command Guard: An open-source safety belt for AI coding agents
The Destructive Command Guard (dcg) blocks catastrophic git and shell commands from AI coding agents, offering sub-millisecond latency and 50+ security packs.
SourceDicklesworthstone/destructive_command_guard