I was at a coffee shop in Williamsburg, trying to get an agent to pull competitor pricing from a handful of websites, when I realized the agent wasn't failing because it was dumb. It was failing because the web is a mess -- JavaScript-rendered pages, paywalls, nested navigation -- and most scraping tools weren't built for the way AI builders actually work.

That's when I switched to Firecrawl.

Firecrawl

https://firecrawl.link/UN5jbla · web scraping and crawling API for AI builders

What it is Firecrawl is a web scraping and crawling API that turns any website into clean, structured data your AI can actually use -- no browser automation, no brittle CSS selectors, no babysitting.

Who it's for Founders and developers building AI products who need reliable web data without dedicating weeks to scraping infrastructure. If you're shipping fast and want your agents to have internet access that actually works, this is the tool.

What I use it for

• Feeding agent workflows with live web data (competitor monitoring, research pipelines, content aggregation)

• Turning documentation sites into structured context for RAG systems

• Scraping product pages without wrestling with JavaScript rendering

• One-shot crawls of entire sites to build training datasets or knowledge bases

What I like

• Handles JavaScript-heavy pages that kill most other scrapers

• Output comes back as clean markdown -- agents can read it without preprocessing

• Crawl mode maps and extracts entire sites with a single API call

• Built specifically for AI workflows -- the output format is designed for LLMs, not humans parsing CSVs

What to watch out for

• Pricing scales with usage -- large crawl jobs can run up credits fast, batch wisely

• Some paywalled or login-gated content still requires manual workarounds

Bottom line If you're building anything with AI agents that needs to read the web, Firecrawl is the cleanest integration I've found. It's the difference between an agent that actually knows what's happening and one that's guessing.

Try it here → Firecrawl

FAQ

Q: What can you do with Firecrawl?

A: Firecrawl lets you scrape any URL or crawl entire websites and get back clean, structured data -- usually markdown or JSON. Use it to feed AI agents real-time web content, build research pipelines, monitor competitors, or extract information from documentation sites and product pages.

Q: Is Firecrawl free?

A: Firecrawl has a free tier that covers basic scraping and crawling, with paid plans for higher volume and more advanced features. For light AI prototyping it's free to start; for production agents running at scale you'll want a paid plan.

Q: Does Firecrawl work with AI agents?

A: Yes -- it's purpose-built for AI workflows. It returns clean markdown output that language models can process directly and integrates with LangChain, LlamaIndex, and custom agent frameworks. It's one of the few scraping tools that actually treats AI builders as the primary customer.

That was the last time I Googled anything for research.

What Perplexity actually is

Perplexity is a search engine that answers questions instead of listing websites. You type what you want to know. It reads the internet in real time, synthesizes the answer, and shows you exactly where it got each piece of information -- inline citations, clickable sources.

No ten blue links. No sponsored results. Just the answer, with receipts.

Why it's different from ChatGPT

ChatGPT is trained on data with a knowledge cutoff. Ask it what happened last week and it either doesn't know or makes something up. Perplexity is connected to the live internet. Every answer is pulled from current sources.

This matters if you're researching tools, markets, or competitors. It matters a lot.

ChatGPT is better for: writing, brainstorming, working through ideas, code

Perplexity is better for: research, fact-checking, staying current, finding sources you can actually cite

I use both. They're not substitutes.

How I actually use it

I use Perplexity Pro, which unlocks access to multiple models (including Claude and GPT-4) and lets you upload files for research.

Daily research use cases for me:

Competitive research

When I want to understand what a tool does, how it's positioned, what people are saying about it -- Perplexity gives me a synthesized answer with sources in under a minute. That's work that used to take 20 minutes of tab-hopping.

Fact-checking before publishing

Before anything goes out in this newsletter, I run claims through Perplexity. I want the source, not the vibe.

Quick briefings

Before a call with someone new, I run a Perplexity search on their company. Three minutes, informed.

Finding tools and comparisons

Asking Perplexity "what are the best tools for X in 2025" gets you a ranked, sourced list. Way faster than reading comparison blogs that are 18 months out of date.

What I wish I'd known earlier

Perplexity has Spaces -- essentially shared research workspaces where you can build a context library and collaborate with a team. I started using this to keep research threads organized rather than starting from scratch each time.

Also: the Focus feature lets you narrow searches to specific sources -- Reddit, academic papers, news, YouTube. The Reddit filter alone is valuable for understanding what real users think about something, unfiltered.

Who it's for

If you do any kind of research -- for content, for business decisions, for competitive intelligence -- Perplexity Pro is worth it. The free tier is useful but the Pro model access and unlimited searches make it a different tool.

If you're already paying for ChatGPT Plus and barely using it for research, Perplexity is the better allocation of that same budget.

This is one of the tools I'd keep if I had to cut everything else down to five.

Try it here → Perplexity

Next week: same problem. Different project. You type into the chat box and... nothing. It doesn't know you. The three hours are gone.

I've been using AI daily for about six months. In that time, I've generated something like 19.5 million tokens of conversations -- debugging sessions, product decisions, writing drafts, architectural choices, photo workflow decisions, business ideas. The kind of accumulated context that would normally live in a second brain.

Except AI doesn't have a second brain. It has amnesia.

I tried everything. Memory files dumped into system prompts. Elaborate markdown logs I paste at the start of sessions. Every memory feature that ships with Claude and ChatGPT. None of it worked the way I wanted -- either too lossy (it summarized away the exact thing I needed) or too manual (I had to remember to use it in the first place).

Then I found MemPalace.

What MemPalace Actually Is

MemPalace is a free, local, open-source AI memory system. It stores your AI conversations verbatim -- every word -- in ChromaDB on your own machine. No cloud. No subscription. No API costs for processing. No deciding what's worth keeping.

The metaphor is in the name: your memory is organized like a palace, not a flat search index.

Conversations live in wings (broad domains like "work" or "personal projects"), rooms (specific projects or topics), closets (sub-topics), and drawers (individual conversation chunks). Searching it feels less like querying a database and more like walking through a building where you already know the layout.

Why Verbatim Beats Summaries

The dominant approach to AI memory is to summarize. You had a long conversation, so the system condenses it to a few bullet points and stores those instead.

This sounds smart. It's not.

When I'm debugging something six months later, "fixed auth issue" doesn't help me. The actual conversation does -- the wrong turns, the exact error messages, the moment we found the fix. Summaries throw away the thing that makes memory useful.

MemPalace keeps everything. And it's benchmarked:

• Score: 96.6% R@5 on LongMemEval -- the leading AI memory benchmark

• Questions: 500-question evaluation suite

• API calls required: Zero

• Against: LLM-summary approaches that cost money and lose fidelity

One honest caveat I appreciate: the team is transparent that the AAAK compression mode (experimental) currently regresses performance versus raw mode. The 96.6% is raw mode only. They say so plainly in the docs. Stick with raw mode for now -- it's the one that actually works at that level.

The Part That Changes Everything: MCP

Here's where it clicked for me.

MemPalace ships with an MCP server. If you use Claude, ChatGPT, or Cursor, you add it once to your MCP configuration and that's it. You never type a search command again.

Now when I open a new Claude session, it automatically calls mempalace_search in the background when relevant context might exist. Last week I mentioned a project name -- it surfaced a conversation from three months ago where I'd made a key architecture decision. I didn't ask it to look. It just looked.

That's the behavior I wanted all along. Memory that works like memory -- not a search bar I have to remember to use.

The Cost

Running this locally costs about $10 per year in storage. That's it.

The LLM-summary approaches MemPalace outperforms? They run continuous API calls to process and compress your conversations. Rough math: ~$507/year at current rates. For a system that scores lower on the benchmark.

Ten dollars. Per year. Everything verbatim. Fully private. No cloud.

How to Get Started

MemPalace is on GitHub at github.com/milla-jovovich/mempalace. The setup:

• Install: Python + ChromaDB, clone the repo

• Configure MCP: Add the MCP server config to Claude Desktop or your preferred client

• Structure your palace: Set up wings and rooms to match how you actually work

• Start chatting: Conversations build the palace automatically from there

If you're a daily AI user -- and if you're reading this, you are -- this is the most important infrastructure you're not running yet.

Your accumulated context isn't just useful. It's a compounding advantage. Every debugging session, every decision thread, every creative iteration -- it all stacks. Stop losing it at the end of every tab.

Get MemPalace

This is a free post. MemPalace is open-source. No affiliation -- just something I'm actually using.

Buffer · social media scheduling tool for solo founders and creators

I used to dread social media.

Not because I hate it. Because I'm bad at the discipline of it. I'd write a great thread in my head in the shower, forget to post it, then scramble to throw something half-baked up on a Tuesday because it had been 11 days and I felt guilty.

Now I batch-write 10--12 posts at a coffee shop in the East Village once a week, queue them in Buffer, and don't touch social again until the following week. Done.

What it is

Buffer is a social media scheduling tool that lets you write posts in advance and automatically publish them across Twitter, LinkedIn, and Instagram on a schedule you set.

Who it's for

Solo founders and content creators who want consistent social presence without a social media team -- or without having to think about it every single day.

What I use it for

• Sunday batch sessions: 10--12 posts written in one 90-minute sitting, queued for the week

• Cross-posting the same idea to Twitter, LinkedIn, and Instagram with slight format tweaks

• Clipping strong paragraphs from my newsletter and rescheduling them as standalone posts

• Sharing reactions and observations (2--3 sentences + a link) without stopping to manually post mid-focus

What I like

• The free plan is actually free -- no credit card, no bait-and-switch

• One dashboard for multiple platforms keeps everything from fragmenting across tabs

• The browser extension lets me queue something mid-browsing without breaking flow

• At $6/month for Essentials, it's the cheapest line item in my stack with the highest daily impact

What to watch out for

• Buffer schedules. It doesn't write. If you never batch, nothing posts.

• Analytics are basic -- fine for solo ops, not what you'd use to run a serious social strategy

Bottom line

Buffer removed the need for social media discipline by giving me one weekly system instead of daily decisions. If you're a solo builder who wants to show up online without it eating your calendar: try Buffer here.

Try Buffer →

Write once a week. Post every day. No social media manager needed.

buffer.com

Affiliate disclosure: This is an affiliate link. I only recommend tools I actually use.

FAQ

Q: What is Buffer used for?

A: Buffer is used to schedule and automatically publish social media posts across platforms like Twitter/X, LinkedIn, and Instagram. You write your posts in advance, set a posting schedule, and Buffer handles the publishing -- so your accounts stay active even when you're focused on other work.

Q: Is Buffer free?

A: Yes. Buffer has a real free plan -- no credit card required -- that covers up to three social channels with a limited post queue. Paid plans start at $6/month if you need more channels or queue capacity.

Q: How is Buffer different from Hootsuite?

A: Buffer is simpler and cheaper. Hootsuite targets larger teams and agencies with advanced reporting and workflow features. Buffer is built for individuals and small teams who need reliable scheduling without enterprise complexity. For solo founders, Buffer is the right tool.

Ranked AI · AI-powered SEO tool for solo founders and creators

I wrote 40 Substack posts last year and watched most of them disappear. Good content, wrong keywords, no structure search engines could work with. I'd sit down at a coffee shop in Williamsburg convinced I was just bad at SEO. Turns out I was just skipping the research.

Ranked AI is what I use now to not skip it.

What it is

Ranked AI is an AI-powered SEO tool that generates keyword-optimized content briefs, maps competitive gaps, and helps you structure content that ranks -- without needing a dedicated SEO team.

Who it's for

Solo founders and content creators who want their writing to show up in search without hiring an SEO agency or becoming an SEO nerd.

What I use it for

• Pre-writing research: put a topic in, get back what to cover, in what order, and what questions to answer

• Keyword clustering for my content calendar -- grouping topics by what can realistically rank together

• Quick site audits before I launch anything to catch obvious structural gaps

What I like

• Cuts my pre-writing research from 2 hours to 20 minutes

• The brief actually shows what competitors are ranking for and where the gaps are

• Clean interface -- no feature bloat getting in the way of the work

What to watch out for

• It's a research and structure tool, not a writing tool -- you still have to write

• Smaller keyword database than legacy tools like Ahrefs, so very niche industries may see thinner data

Bottom line

Ranked AI made SEO something I actually do now instead of skip because it's too complicated. If you're publishing content and want it to rank, start here.

Affiliate disclosure: This is an affiliate link. I only recommend tools I actually use.

FAQ

Q: What does Ranked AI do?

A: Ranked AI helps you create SEO-optimized content by generating keyword research, content briefs, and site structure recommendations powered by AI. It tells you what to write, how to structure it, and what questions to answer to rank in search.

Q: Is Ranked AI worth it for solo founders?

A: Yes -- especially if you're publishing content regularly. It replaces hours of manual keyword research and gives you a clear brief before you start writing. For a one-person operation, it functions like having a lightweight SEO consultant on call.

Q: How is Ranked AI different from Surfer SEO or Ahrefs?

A: Ranked AI is simpler and more AI-native than both. Surfer SEO focuses on optimizing content you've already written. Ahrefs is a full-stack backlink and keyword suite -- powerful but overkill for solo operators. Ranked AI is built to help you create SEO-ready content from scratch, fast.

Chatbase · AI chatbot builder for founders and creators

I used to sit down at a coffee shop in the West Village to do actual work and end up spending 45 minutes answering the same three support questions instead. Same questions every week. Different people, identical answers.

Chatbase is what I built to stop that from happening.

What it is

Chatbase is an AI chatbot builder that lets you train a custom assistant on your own content -- docs, website pages, PDFs, FAQs -- and embed it anywhere in about 20 minutes.

Who it's for

Founders and creators who spend real time answering pre-sale questions, onboarding the same new users, or fielding DMs they've already answered a dozen times.

What I use it for

• A 24/7 FAQ bot trained on my product docs -- lives on my landing page and handles most pre-sale questions before anyone emails me

• An onboarding assistant for new subscribers so I'm not in my inbox every Monday morning

• Testing whether a product idea has enough repeatable FAQ surface area to be worth building out

What I like

• Setup takes 20 minutes if your docs are already written -- paste a URL, done

• Handles multilingual questions without any extra configuration

• Free tier is real and actually usable, not just a teaser

What to watch out for

• Output quality is only as good as what you feed it -- a thin knowledge base gives thin answers

• Complex edge-case questions still need a human fallback

Bottom line

Chatbase turned my FAQ inbox into a solved problem. If you're fielding the same support questions more than a few times a week, try Chatbase here -- you'll wonder why you waited.

Affiliate disclosure: This is an affiliate link. I only recommend tools I actually use.

FAQ

Q: What is Chatbase used for?

A: Chatbase is used to build AI-powered chatbots trained on your own content -- websites, PDFs, help docs, or FAQs. Founders and creators use it to automate customer support, pre-sale Q&A, and product onboarding without writing any code.

Q: Is Chatbase free?

A: Chatbase has a free plan that covers basic usage. Paid plans unlock more message volume, multiple chatbots, and advanced integrations.

Q: Can Chatbase be trained on my own content?

A: Yes -- that's the core feature. You can train Chatbase on your website URL, uploaded PDFs, text files, or manually written Q&A pairs. It learns from whatever you give it and uses that to answer questions accurately in context.

I'm writing this from a coffee shop, which means I have approximately 45 minutes before my laptop battery dies and my oat milk latte goes cold. So let's make it count.

Substack dropped something this week that I've been waiting for: a built-in recording studio. Not a third-party integration. Not "export your Zoom and upload it." An actual record-edit-publish workflow that lives inside the platform you're already using.

Here's the number that made me sit up: creators who use audio or video on Substack have grown their revenue 50% faster than those who don't.

Fifty. Percent. Faster.

If you've been putting video off because the setup felt like too much -- gear, editing software, separate upload flows, scheduling live streams around your audience's timezone -- that excuse just got a lot harder to use.

What the Substack Recording Studio Actually Does

Pre-recorded content

Record when you want, publish when you're ready. No more scheduling live streams. No more "I'll do video when my setup is better." Record offline, review it, publish it.

Solo or with guests

Bring in up to two guests for a conversation. Useful for interviews, collab content, or just having someone to riff with so you're not staring at your own face for 20 minutes.

Screen sharing

Both host and guests can share their screens -- great for walkthroughs, demos, or reacting to something live. (Audio capture for shared content is coming soon, FYI.)

Auto-generated clips and thumbnails

When you finish recording, the platform automatically creates short clips and a custom thumbnail. That's the part that used to eat hours. Now it's just... done.

Publication branding

Embed your logo or wordmark directly into the video frame, positioned wherever you want it. Consistent, professional, no extra editing step.

Distributes everywhere automatically

Mobile app, web, Substack TV. You record once, it lands everywhere your subscribers already are.

Why This Is Bigger Than It Looks

Most tools that promise to simplify video still require you to leave your main platform, manage a separate workflow, and manually cross-post everywhere. Substack just collapsed that entire stack.

If your Substack is already your hub -- for writing, for your paid community, for your content calendar -- adding video now costs you almost nothing operationally.

And if that 50% revenue stat holds even directionally? That's not a marginal edge. That's the difference between a newsletter that pays for your coffee and one that pays for your rent.

How to Get Your First Recording Live

Step 1 -- Open the Substack editor on desktop and look for the new recording option. (Currently desktop only.)

Step 2 -- Do a solo recording first. No guests, no screen share. Just you, talking for 5--10 minutes about something your audience already asks you about.

Step 3 -- Let Substack generate your clips and thumbnail. Don't touch it. Don't tweak it. Just see what it produces before you decide it's not good enough.

Step 4 -- Publish it to your free list. This is a discovery play first, not a monetization play. Get reps before you put it behind a paywall.

Step 5 -- Watch the engagement data for 48 hours, then decide if video becomes a regular format for you.

The gear question: you don't need anything new. Laptop camera and built-in mic are enough to start. Get your reps in first, upgrade later.

The time question: a 5-minute recording is a legitimate piece of content. You are not committing to a podcast.

The perfectionism question: your first recording will be imperfect. So will your second. Ship it anyway.

Paid subscribers get the full content calendar template I use to plan video drops alongside newsletter posts -- including which topics convert better as video vs. written and how to batch both in the same work session. If you're not yet paid, today's a good day to fix that.

Wispr Flow · AI voice-to-text dictation tool

I'm at a coffee shop in Carroll Gardens, and I'm not typing this. I'm talking. Wispr Flow is running in the background, and everything I say is appearing in this document -- punctuated, formatted, in my voice. Total time talking: about 12 minutes.

What it is

Wispr Flow is an AI-powered voice dictation …

Read more