← Home
2026-11-04 · 中文版
Only AI tools I have used, seriously evaluated, or that are popular enough to be worth knowing. Usage and rating are scored separately: usage is how much I actually use a tool; rating is my recommendation — for tools I haven’t used, the rating is based on reputation and public material, not hands-on testing. ★★★★★ highly recommended, ★★★★☆ worth using/trying, ★★★☆☆ worth a look, ★★☆☆☆ evaluated, not adopted. At most 5 full entries per category; overflow goes in an “Others” line with names only. Continuously updated; v3 has 21 entries, plus 3 more listed under “Others”. Skills live on their own page: AI Skills.
1. Code Understanding & Graphs
| Tool |
Usage |
Rating |
Comment |
| Understand-Anything |
Not used yet — popular, want to try |
★★★☆☆ |
90k-star code knowledge-graph plugin, Tree-sitter + LLM graph building; best for getting oriented in a large unfamiliar codebase before reading code; not yet tested, in the queue. |
| Graphify (inside thinktank) |
Regular use |
★★★★☆ |
Curated code graph built into thinktank: 1,029 nodes / 1,701 edges / 61 communities; read the graph before a review to see what a change touches. |
2. Multi-Model & Multi-Agent Orchestration
| Tool |
Usage |
Rating |
Comment |
| muse-thinktank |
Daily use |
★★★★★ |
One question to N models in parallel, dual judges cross-validate, synth merges a verdict; used in weekly reviews — disagreement is signal. |
| model-fuse |
In development |
★★★☆☆ |
My opencode plugin (private repo): one question to many models, merge the best; direction for D10/D11 settled, not replacing with Copilot Fusion. |
| GitHub Copilot Fusion |
Evaluated, not adopted |
★★☆☆☆ |
D10 adversarial review is roughly Critique, D11 verdict differs from Cascade quality gates — not equivalent; worth knowing, that’s it. |
| LangGraph |
Not used yet — popular, want to try |
★★★★☆ |
Graph orchestration framework; state-machine agent flows are a production-grade pattern; Meta’s Thin Orchestrator paper describes the same idea — worth reading side by side. |
| AutoGen |
Not used yet — popular, want to try |
★★★☆☆ |
Microsoft’s multi-agent framework, the classic conversational agent orchestration; conceptually comparable to thinktank’s jury review, but a different job. |
3. Memory & Knowledge
| Tool |
Usage |
Rating |
Comment |
| open-memex |
Daily use |
★★★★★ |
Local-first memory layer: Markdown as source of truth, SQLite FTS5 search, MCP + CLI; npm 0.8.2. |
| Mem0 |
Not used yet — popular, want to try |
★★★☆☆ |
Memory layer for agents, marketed as pluggable long-term memory; same space as open-memex, useful as a design reference. |
4. AI Coding Assistants
| Tool |
Usage |
Rating |
Comment |
| Claude Code |
Daily use |
★★★★★ |
My primary coding agent; native host for plugins like Understand-Anything. |
| GitHub Copilot CLI |
Regular use |
★★★★☆ |
Company-approved AI tool; WorkBuddy’s main execution backend, called through AgentAdapter. |
| OpenCode |
Regular use |
★★★★☆ |
Open-source AI coding CLI; WorkBuddy’s personal hand, interchangeable with Copilot CLI. |
| Cursor |
Not used yet — popular, want to try |
★★★★☆ |
The most popular AI editor, completion + agent mode in one; never used it, but too popular not to know — worth trying side by side when there’s time. |
| Codex |
Not used yet — popular, want to try |
★★★★☆ |
OpenAI’s coding agent, CLI + cloud in parallel; same space as Claude Code, try before comparing. |
Others: Gemini CLI, Cline, Aider (names only, no rating).
5. Agent Frameworks & Runtimes
| Tool |
Usage |
Rating |
Comment |
| DeepSeek Harness (dsh) |
Evaluated (borrowed ideas) |
★★★☆☆ |
v0.2.2-alpha.2, plugin-based micro-kernel agent runtime; high borrowing value (skill catalog lazy loading, flow skills, “model-visible <=> logged”), not adopted directly. |
| LangChain |
Not used yet — popular, want to try |
★★★★☆ |
The oldest agent framework, the biggest ecosystem; new projects rarely start from it, but reading it is reading the industry consensus. |
| CrewAI |
Not used yet — popular, want to try |
★★★☆☆ |
Role-based multi-agent orchestration (Role + Task + Crew), easier to start than AutoGen; a reference for small-team multi-agent prototypes. |
| OpenAI Agents SDK |
Not used yet — popular, want to try |
★★★☆☆ |
OpenAI’s official lightweight agent SDK; worth a look if you want to stay close to the OpenAI ecosystem. |
| Pi |
Not used yet — personal project |
★★★★☆ |
117k-star thinnest agent harness: the kernel only connects model/tools/persistence, deliberately without permissions, Plan mode, or sub-agents; many FDEs use it directly as their harness; can’t enter company production, kept for personal projects to learn harness design. |
6. Investing
| Tool |
Usage |
Rating |
Comment |
| moomoo OpenD |
Regular use |
★★★★☆ |
The only real API for quotes + paper trading; 1024-bit RSA encrypted handshake, laptop listening on 0.0.0.0 over Tailscale; live-trading unlock expires ~Dec 1. |
| tt/stock_context.py |
Regular use |
★★★★☆ |
Homegrown (inside thinktank): pull technicals before writing any investment analysis; every number needs a source; a draft with framework but no data is half-baked. |
Admission rule: I decide what enters and leaves this list; a tool gets in after I’ve used it, seriously evaluated it, or because it’s popular enough to be worth knowing; entries are hands-on notes, not vendor copy; homegrown tools need a repo link to enter, third-party AIs don’t enter. Send me a link if you want me to try something.