AI UNDERDOGSDAILY PICK
AI UNDERDOGS
喂 AI 你的数据,让它写代码
Give AI your data, then let it write the code
approximatelabs/sketch
大多数 AI 写代码工具
懂语法但不懂你的数据
Sketch 不一样——你把数据扔给它
它帮你写代码
Most AI code tools know syntax but not
your actual data. Sketch flips that — point
it at your data, it writes the code
它到底解决了什么问题
The real problem it solves
你用过就知道
让 AI 写 pandas 代码
它猜你有哪些列、猜数据长什么样
十次有八次跑不通
Sketch 的思路很简单:既然 AI 不了解你的数据
那就把数据告诉它
We've all been there — you ask an
AI to write pandas code and it hallucinates
your column names. Sketch's fix is dead simple
just give the AI your data context
★ SIGNAL 1
不复制数据,但真懂数据
Understands your data without copying it
你指向一个 pandas DataFrame
它自动推导出 schema
数据分布、采样结果
打包成上下文喂给大模型
不用你手动写描述、不用导出 CSV
这个设计是真的有产品脑
Point Sketch at a pandas DataFrame and it
auto-derives schema, distributions, and samples — then feeds
that context to the LLM. No manual description
needed. That's a product brain right there
★ SIGNAL 2
2300 星,全跑本地
2.3k stars, runs fully local
2300 个 star 了
但作者最近专门加了纯本地模式
零外部 API 调用
做数据工具的都知道,数据隐私是命门
能花功夫把 LLM 推理拉到本地
说明作者是认真的
2.3k stars and counting, but the author recently
pushed a pure local mode — zero external
API calls. For a data tool, that's not
optional. It shows the dev takes data privacy
seriously
没有作者原话
但看项目更新频率就知道
这不是那种扔上去就不管的玩具
两千多星、持续迭代
坚持本地推理——这个节奏,值得留意
No quotes from the author, but the commit
history tells you everything — 2.3k stars, active
development, local-first philosophy. This isn't a weekend throwaway
AI UNDERDOGS
approximatelabs/sketch
approximatelabs/sketch
approximatelabs/sketch
关注 · 每天发现更多 AI 神作
github.com/approximatelabs/sketch