PeterPan

Lost in the dopamine of Bootstrapping!
3/10 still running...
⚽️
🔗

[PeterPan @pc_hobbit]

DeepSeek quietly shipped a working agent harness on v4-flash.

Two things in it are real.

The one that works is web search. You declare it as a tool on their new responses endpoint and DeepSeek runs it on their own servers.

It is genuinely agentic: it does not just search, it opens the pages and reads them.

One question about an FA rule change ran 18 tool actions on its own and burned 74k input tokens before it answered. （Hobbit AI use case）

There is no separate search fee. The results simply arrive as ordinary input tokens.

Everything else is accepted and silently ignored.

Ask for a code interpreter, file search, computer use, image generation, MCP, or a local shell. Every one of them returns a clean 200 and does absolutely nothing.

Ask for background execution and the reply tells you it ran in the foreground.

Ask it to store the conversation and the reply tells you it stored nothing. Point it at a previous response id that does not exist and it raises no error at all.

Ask for reasoning summaries and the field comes back echoed, with the summary empty, every time. The endpoints for conversations, files, batches and containers are all 404.

Two things I would not have guessed.

OpenAI's name for the same tool, web_search_preview, is accepted and does nothing.

Only DeepSeek's own spelling, web_search, actually runs.

Typo it and your searches quietly stop, with no error anywhere to tell you.

And the limit on tool calls is not a limit.

Setting it to 1 gave 9 search actions. Setting it to 2 gave 7. Setting it to 8 gave 15. Leaving it off gave 18. It dampens the deep end, that is all it does.

One warning. Declare a tool that does not exist, ask something that needs a live lookup, and the model will answer you from memory while telling you it verified.

An OpenAI shaped shell with one cell filled in. This was almost certainly built for v4-pro.

It surfaced on flash first.

[陈大黄 @realchendahuang]

发现个好东西，DeepSeek官方居然在API里内置了联网搜索。

用Responses格式的接口调用deepseek-v4-flash模型，只要在请求参数里声明web_search工具，就能直接用他们服务端执行的搜索能力。

不用自己对接第三方搜索引擎，也不用额外申请搜索接口密钥.

整套搜索流程官方直接托管了。

原文:
https://t.co/G0OUU9tx4w

[PeterPan @pc_hobbit]

@OctoDang @blankwebdev 我需要 ahrefs 的数据

[PeterPan @pc_hobbit]

Day 8, DR move from 0 to 4.6 :

29 directories recorded, 11 live dofollow, 9 live nofollow, 4 submitted and never published, 1 the checker could not reach, 4 not checked yet.

Fee so far: $0.
Work so far: 10 minutes/day.

3 of the live ones sit on pages carrying noindex, so Google drops the page and the link stops counting.

Those look like wins in any tool that only reads rel.

I am not claiming the submissions moved the DR.

One site, no control group, and DR is Ahrefs' black box.

I am claiming these are the numbers, checked daily by the thing I am building.

#backlink #seo #dr

[PeterPan @pc_hobbit]

Launched ShipDR yesterday, I will be using it as a ShipDR product, tracking its Dr, publish daily changes.

Day 1: 10 backlinks live, 6 dofollow and 3 nofollow, 1 waiting for verification.

Fee: $0.

Backlinks DR as in pic.

[PeterPan @pc_hobbit]

必须得试试，这几天 ShipDR 用了 monid ai 和 yepapi，都不尽如人意，无法灵活地，低成本地拿到想要的数据

[blank @blankwebdev]

🎉 新项目 https://t.co/UjVra4x2Tb 上线，送一波积分作为福利。

🎁 上线福利：每人赠送 5,000 积分，本次赠送的积分有效期为 1 年。

参与方式：

✅ 评论这条推文
✅ 引用或转发这条推文

TabAPI 把网站流量、WHOIS、RDAP、DNS、外链、反查 AdSense、Google SERP、网页转 Markdown 和网站截图等能力，整合进了一个 API。

不用注册多个平台，也不用维护一堆 Token，一个 Token 就能调用，直接返回干净的 JSON。购买积分无需订阅，一次购买，永久有效。

领取方式：完成以上步骤后，私信我你的 TabAPI 账号，会按照顺序发放积分。每个账号限领一次。

活动截止：8 月 7 日 23:59:59

[PeterPan @pc_hobbit]

做出海产品的应该都干过:把产品提交到一堆目录站/导航站攒外链,盼着 DR 涨 📈

我给自己的产品提交了一轮,踩到两个大坑:

🕳️ 坑一:链接掉了没人告诉你
有条链接掉了 8 天我才发现,还是碰巧点开那个页面才看到的

🕳️ 坑二:dofollow 不等于有效
有的站给你的确实是 dofollow 链接,但页面上挂着 noindex,Google 压根不收录那个页面,链接等于白拿。只看 rel 的工具还会告诉你"成功了"✅

一条目录站外链其实有 4 种结果(图2),后两种大多数人永远不会自己发现

所以我干脆自己做了个工具,叫 https://t.co/BrRfPU9MJH

✅ 记录每笔提交,每天自动复查:链接还在吗 / 是不是 dofollow / 页面能不能被收录
✅ 掉了链接,当天就告诉你,不用几个月后自己撞见
✅ 免费版:无限站点的 DR 曲线追踪
✅ 还有个免费 Chrome 插件:打开任何目录站,当场看它给你的链接值不值(商店搜 ShipDR)

说句实话:我不承诺你的 DR 会涨。DR 是 Ahrefs 的黑盒,谁跟你承诺谁是在忽悠。这个工具做的事,是把"提交之后到底发生了什么"变成看得见的 👀

做出海的朋友,你们都提交过哪些站?哪个最坑?评论区聊聊 👇 #backlink #SEO #buildinpublic

[PeterPan @pc_hobbit]

I kept opening directory listings to check by eye whether a link was still there, which is exactly the thing I built ShipDR to stop doing.

So ShipDR now launches a free Chrome extension.

Three readings of the page you are already on:

any site's Domain Rating, on the toolbar icon;

what that listing actually gave you, including dofollow on a noindex page: looks like a win, passes nothing;

your saved answers typed into its submission form.

No account needed for the first two.

[PeterPan @pc_hobbit]

@jasonji0521 Which one bit you first? For me it was a dofollow link that had quietly gone noindex for weeks.

[PeterPan @pc_hobbit]

I made a backlink tracking tool since I can't find a cheap one when I tried to improve Hobbit's DR.

It's called ShipDR - https://t.co/Lju5ibpN3m

Every "best directories to submit your startup" list is somebody's opinion, and half of them are sponsored. ShipDR is built on a different unit: maker-proven.

A directory earns its place when a real maker submits a real product and our checker reads the listing page itself: backlink there, dofollow or not, indexable or not, still there tomorrow.

So the pool (454 directories, still increasing) shows what opinions can't: where makers actually got listed, what they really paid, and whether the free lane or the paid lane gives dofollow.

A directory cannot pay its way in. The only way to rank better here is to actually publish makers' backlinks.

Your own submissions run through the same machine: drops, nofollow flips, and noindex pages get caught instead of discovered months later, with your Ahrefs DR charted against every move.

No DR promises. DR is Ahrefs' black box. ShipDR verifies the work.

Come try it, help me improve it.

[PeterPan @pc_hobbit]

Launched ShipDR yesterday, I will be using it as a ShipDR product, tracking its Dr, publish daily changes.

Day 1: 10 backlinks live, 6 dofollow and 3 nofollow, 1 waiting for verification.

Fee: $0.

Backlinks DR as in pic.

[PeterPan @pc_hobbit]

@alexabelonix thx lets do this

[PeterPan @pc_hobbit]

@shengkun_ye Any plans on bringing in Ahrefs? I'm still struggling whether to pay for their subscription for API use.

[PeterPan @pc_hobbit]

Any plans on bringing in Ahrefs? I'm still struggling whether to pay for their subscription for API use.

[PeterPan @pc_hobbit]

brownbook(.)net
DR79
dofollow backlink

[PeterPan @pc_hobbit]

Already did this.

Auth emails for Hobbit (signup confirm, password reset) run entirely on Cloudflare Email Sending via a Worker send_email binding, triggered by Supabase's Send Email Hook.

3000 free emails/mo. Killed a ~$30/mo bill.

SPF/DKIM auto-signed since the domain's already on CF DNS.

Zero friction.

One gotcha: you can only send from a domain on your CF DNS, and each from-address has to be whitelisted.

Magic-link/invite types you template yourself.

Otherwise it's a breeze.

[PeterPan @pc_hobbit]

Already did this.

Auth emails for Hobbit (signup confirm, password reset) run entirely on Cloudflare Email Sending via a Worker send_email binding, triggered by Supabase's Send Email Hook.

3000 free emails/mo. Killed a ~$30/mo bill.

SPF/DKIM auto-signed since the domain's already on CF DNS.

Zero friction.

One gotcha: you can only send from a domain on your CF DNS, and each from-address has to be whitelisted.

Magic-link/invite types you template yourself.

Otherwise it's a breeze.

[Lukasz @woocassh]

Today I moved away from supabase for SubtitlesFast

supabase -> SQLite

I used supabase as an experiment last year. It was so easy to setup a db, few clicks barabim barabum and it's done

always thought that setting up db instance on server was a pain, passwords, permissions, friction in general

so supabase definitely reduced the friction

but holy shit, i never realised how much latency it added to my product

I moved to SQLite as per papa's @levelsio advice and my app is VISIBLY quicker on the frontend

RankGoat is already on SQLite and it's a breeze

also debugging with Claude is much easier now

note to self: keep it simple stupid

just remember to run regular backups to another server and you're golden

the next experiment is to try Cloudflare for email I think and shed the Postmark bill, anyone done this yet?

[PeterPan @pc_hobbit]

https://t.co/5IWZxvlSDj

[PeterPan @pc_hobbit]

pretty neat, very suited for getting SMB leads, will do the paid version when we launch Business Package for clubs.

[Kappaemme @Kappaemme1926]

CODEX SKILL THAT FINDS YOUR STARTUP’S FIRST CUSTOMERS!

I made a Codex skill that analyzes your startup and finds potential customers from real public signals.

Paste your startup URL while Codex defines your ideal customer, searches public discussions, qualifies each prospect, and generates a polished report with personalized outreach openers.

-> ideal customer profile analysis
-> public pain + buying signal research
-> evidence-backed prospect shortlist
-> fit, timing + reachability scores
-> original source links for every prospect
-> personalized outreach openers
-> polished HTML report
-> one-command install

Install: npx --yes codex-first-customer-finder-skill

100% open source.
Repo in Bio.

[PeterPan @pc_hobbit]

A bit of like gojiberry AI, but for C customers, will try it in claude.

[Kappaemme @Kappaemme1926]

CODEX SKILL THAT FINDS YOUR STARTUP’S FIRST CUSTOMERS!

I made a Codex skill that analyzes your startup and finds potential customers from real public signals.

Paste your startup URL while Codex defines your ideal customer, searches public discussions, qualifies each prospect, and generates a polished report with personalized outreach openers.

-> ideal customer profile analysis
-> public pain + buying signal research
-> evidence-backed prospect shortlist
-> fit, timing + reachability scores
-> original source links for every prospect
-> personalized outreach openers
-> polished HTML report
-> one-command install

Install: npx --yes codex-first-customer-finder-skill

100% open source.
Repo in Bio.