diagram-design
github.comrepository
27 editorial diagram types for Claude Code as a self-contained skill — no Mermaid, no shadows, no build step, just HTML and SVG that looks drawn by a person.
The Daily Desk
A useful mess, carefully indexed.
A working catalogue of tools, research, and references selected for practical use, not shelf appeal.
On the desk now
Current file
The first carefully selected file will appear here.
The useful inventory
Catalogue
GLM-5.3
z.aitool
Z.ai's next model reuses the GLM-5.2 base and puts every gain into post-training — weights held back roughly two weeks this time, after training produced cybersecurity capabilities the team didn't plan for.
Bullet
codewithbullet.comtool
Claims 30-60% faster than Claude Code and Codex on the same tasks — the coding-agent speed race gets its own leaderboard.
Gemini 3.7 Flash
blog.googletool
A second Flash release just three weeks after 3.6 — same base model, algorithmic improvements only, and double-digit gains on coding and agent benchmarks.
deepseek-harness
github.comrepository
DeepSeek's MIT-licensed agent harness: everything — model adapter, tool registry, session log, sandbox, even the agent loop — is a plugin. 95k+ stars in its first two days.
BetterClaw
www.betterclaw.iotool
Deploy an AI agent in 60 seconds for $0 — a no-code agent builder riding the same wave OpenClaw opened up earlier in the year.
bb
getbb.apptool
An IDE that builds itself — pitched as proof that the editor and the agent building it can be the same evolving codebase.
Tines 3B
www.tines.comtool
A governance layer for the AI workflows, apps and agents employees are already building — isolated execution, protected credentials, a full audit trail.
Continuum
roland.leth.rotool
Remembers what you know about the people you manage — a memory layer built for 1:1s and performance context, not general note-taking.
Grok Bot
x.aitool
xAI's always-on agent gives each task its own persistent cloud computer and its own logins, and keeps working after you close the laptop. Built under Cursor's "Sand" codename.
Seed 2.1 Turbo
seed.bytedance.comtool
ByteDance's low-latency tier of Seed 2.1, on Volcano Engine — pitched as beating Claude Opus 4.6 at up to 80% lower cost for high-frequency agent workloads.
awesome-mcp
github.comrepository
A daily-updated, curated list of MCP servers, clients and frameworks — useful precisely because it doesn't go stale like a normal awesome-list.
Kiro Crew
kiro.devtool
AWS open-sources an orchestrator that runs a crew of coding agents unattended — persistent memory, approval gates, and a live dashboard for watching them work. Grew out of an internal Amazon tool already used by 39,000+ engineers.
ngrok AI Gateway
ngrok.comtool
One private gateway in front of every model an org uses, hosted or local — ngrok extends its tunnel infrastructure into AI traffic.
Terminal Candy
terminalcandy.comtool
A native macOS terminal built to be skinned and themed properly, for anyone tired of living in a default black rectangle.
AgentMicro
agentmicro.cctool
Live Codex task status in the macOS menu bar — the kind of small utility that only exists once agents run long enough to need a status light.
DeepSeek-V4-Flash-0731
huggingface.cotool
DeepSeek retrains its flash tier of V4 for coding, agents and tool use — 284B total parameters, 13B active, and a 1M-token context window.
Port22
tryport22.comtool
Claude Code, Codex and other coding agents, driven from a phone — for the moment a build breaks while you're away from a keyboard.
Prefactor
prefactor.techtool
Evaluates every step an agent takes, live, instead of just scoring the final output — built for catching failures mid-run.
Qwen3.7 Flash
help.aliyun.comtool
Alibaba's cheap, fast vision-language tier of Qwen3.7 — $0.03/M input tokens, a 991K context window, and no accompanying technical report.
Kimi K3
huggingface.cotool
Moonshot open-sources a 2.8-trillion-parameter, 1M-context model — billed as the largest open-weight model released so far, with only 104B parameters active per token.
Claude Opus 5
www.anthropic.comtool
Anthropic's flagship rolls out across providers — close to Fable 5's frontier intelligence at half the price, and the new state of the art on coding and knowledge-work evals.
Gemini 3.6 Flash family
blog.googletool
Google ships three models in one day — Gemini 3.6 Flash, 3.5 Flash-Lite, and a cybersecurity-tuned 3.5 Flash Cyber — with Gemini 3.5 Pro still nowhere in sight.
Backdrop
www.backdropagent.comtool
AI coworkers that run projects and day-to-day operations rather than just answering questions about them.
Browse the drawers