Building Pactify: Updates & Learnings
I'm building Pactify in public—sharing what I'm learning, what's confusing, and what's not working. Not marketing, just honest progress updates.
Try What We've Built So Far
v1.1.1 - This Week's Bug Fixes & Small Wins
- →Spent a day figuring out why images weren't exporting. Turns out each AI platform wraps image blobs differently in their DOM. Wrote custom adaptors for Gemini and ChatGPT — your DALL-E creations should finally survive the export.
- →Claude's interface update broke our button positioning (it started overlapping their 'Share' menu). The joy of building on moving platforms. Adjusted our CSS injection to play nice with their new layout.
- →The Sidepanel started feeling heavy on my 500+ message conversations. Profiling showed too many unnecessary re-renders. Switched to virtualized lists — scrolling should be buttery smooth now.
- →Felt the navigation header was stealing too much vertical space from the actual content. Compacted the design while making click targets larger. Still iterating on how to maximize reading area in such a narrow window.
Pactify v1.1.0: Finally, a Real "Second Brain Portal"
I built the v1.0 Popup following standard tutorials. Big mistake. In daily use, I found myself hating it—it covered my content and felt like an interruption, not a helper.
The deeper issue I faced: My "Second Brain" (Notion) and my "AI Brain" (ChatGPT/Claude) were completely siloed. I was a human copy-paste bridge, wasting hours moving context back and forth.
So I made a scary decision: stop all marketing, pause new features, and rewrite everything for the Chrome SidePanel API. v1.1.0 isn't just an update; it's the workflow I actually wanted.
What's New
- →Architecture Shift: Scrapped the React Popup implementation entirely. Moved to Chrome SidePanel API + Manifest V3. It was a headache to handle the different content security policies, but the persistent workspace is worth it.
- →The "Lost Conversation" Fix: I kept forgetting which AI I used for specific tasks. Built a unified local index that aggregates ChatGPT, Claude, and Gemini history into one search bar.
- →Context Injection: Solving the "Alt-Tab" fatigue. You can now select Notion pages and inject them into your current AI chat context without leaving the tab (Experimental).
- →Waitlist Reality: Notion's API limits are stricter than I realized. Had to prioritize stability for paid users over massive free growth.
What I Learned
- ✓Context switching is the enemy. If a tool makes you switch windows, it costs more cognitive load than it saves.
- ✓Rewriting is terrifying. We went "dark" for 3 weeks to fix the architecture. Usage dropped, but the foundation is finally solid.
- ✓Sustainability vs. Growth. Notion's API quotas forced a hard choice: restrict deep integration to paid users or crash the service. I chose stability over vanity user counts.
Notion Auto Sync Beta: The Bridge Between AI and Your Second Brain
I exported an AI conversation to Google Docs last month. Perfect formatting, clean structure.
Then I closed the tab and forgot it existed. Three weeks later, I repeated the same conversation.
The export worked. The workflow didn't. That's when I realized: I'd been solving the wrong problem.
AI conversations need to live where your thinking happens—not in export folders. That's why I built Notion Auto Sync.
What's New
- →Notion Auto Sync (Beta): Automatically sync ChatGPT, Claude, and Gemini conversations to your Notion workspace
- →Full Conversation Content: Properly formatted with tables, code blocks, and LaTeX formulas preserved
- →Smart Metadata: Auto-includes conversation title, date, AI platform, and direct link back to original
- →Multi-Platform Support: Works seamlessly across ChatGPT, Claude, and Gemini
- →Auto-Generated Tags: Coming soon—AI-powered conversation summarization and tagging
- →Chrome Sidepanel: Quick access to Pactify features directly in your browser (in development)
What I Learned
- ✓Export problem ≠ workflow problem. The real gap is "my conversations are disconnected from my knowledge system"
- ✓Different people organize differently: by project, by date, by topic. One-size-fits-all integration doesn't work
- ✓My workflow isn't your workflow. Need real user scenarios before building Phase 2 features
- ✓The bridge isn't just about moving text—it's about actual workflow integration between AI and your "thinking place"
- ✓Building Phase 1 (one-way sync) proves the concept. But the bigger vision is two-way interaction: Notion → AI and AI → Notion
- ✓This bet might be wrong. Maybe people don't want deep integrations. That's why I'm building in public—to find out
One-Click Notion Export: Connect Your AI Conversations to Your Second Brain
Remember spending 5-10 minutes copy-pasting ChatGPT conversations into Notion? Losing formatting, manually adding metadata, fixing broken tables?
After hearing this frustration from dozens of researchers and consultants, I built what should have existed from day one: one-click Notion export.
Now your AI conversations land directly in your Notion workspace—formatted, organized, and ready to use.
No more workflow friction. No more context switching. Just click, export, done.
What's New
- →One-Click Notion Export: Export ChatGPT, Claude, and Gemini conversations directly to Notion with a single click
- →Multi-Platform Support: Works seamlessly across all three major AI platforms with consistent quality
- →Format Preservation: Tables, code blocks, LaTeX formulas, and formatting maintained perfectly
- →Smart Metadata: Auto-includes conversation date, platform, model info, and token count
- →Instant Speed: Complete export in under 5 seconds—no waiting, no processing delays
- →Flexible Organization: Choose target database, add custom tags, customize export location
What I Learned
- ✓Export friction isn't a minor inconvenience—it's a workflow killer that makes people abandon valuable AI conversations
- ✓The real problem isn't "saving AI chats" but "integrating them into existing workflows where actual work happens"
- ✓High-frequency users (5+ conversations/day) lose 30-60 min/week on manual export—time that should go to actual thinking
- ✓Different users need different organization: some by project, some by date, some by topic. One-size-fits-all doesn't work
- ✓Quality matters more than speed: a 5-second export that preserves formatting beats a 1-second export that loses context
Beyond Export: Building Workflow Connections
After talking with dozens of researchers, I realized something: just exporting AI conversations isn't enough.
You use Zotero, Obsidian, Overleaf, Google Docs. Your workflow is fragmented by design.
What people actually need is better connections between the tools they already use.
This led us to completely rethink what Pactify should be—and what we built next.
What's New
- →Connect to Google Docs: AI conversations land directly in your Google Drive where your drafts live
- →Connect to Obsidian (Markdown): Drop conversations into your vault, link to existing notes, build knowledge graphs
- →LaTeX Copy on All Platforms: ChatGPT, Gemini, Claude—all standardized with one-click copy
- →Light/Dark Mode Toggle: Switch seamlessly without reloading, perfect for late-night work
- →Freemium Upgrade: 30 conversions/month (6x), 1MB per file (10x), all templates available
What I Learned
- ✓Format problem ≠ workflow problem. The real question is "where does this go in my system?"
- ✓You shouldn't need to hit limits on day 3 just to test if something fits your workflow
- ✓I don't have all the answers—metadata preservation, bi-directional links, Zotero integration? Need your feedback
- ✓Quality feedback beats vanity metrics. Would rather grow slowly with people who actually need this
- ✓Building in public means admitting uncertainty and asking for help, not showing off what's "done"
LaTeX Support, Multi-Platform, and Quality Improvements
After two weeks of 14-hour days, I shipped what users asked for.
LaTeX copying, Claude and Gemini support, Markdown exports, better formatting—real problems from real researchers.
But here's what keeps me up: did I build what people actually need, or what I think they need?
Am I solving core problems, or just enabling workarounds?
What's New
- →LaTeX Copy Button: Click "Copy LaTeX" on ChatGPT formulas, paste directly into your editor
- →Claude & Gemini Support: Custom handling for each platform's quirks to maintain export quality
- →Markdown Export: Academic-standard formatting with proper LaTeX delimiters and citation formats
- →LaTeX Formula Fixes: ChatGPT's non-standard delimiters ([...] vs $$) now render correctly in Word
- →Styling Improvements: Tables, headers, lists spacing and alignment match academic standards
- →Testing Obsession: 20 academic scenarios × 3 platforms × 100+ test cycles = 97%+ accuracy
What I Learned
- ✓Every change breaks something else—fixing LaTeX broke table spacing, improving Claude changed ChatGPT citations
- ✓Multi-platform support might be enabling workarounds rather than solving core problems
- ✓"Ready to use" means different things to different people—need more clarity on what quality bar matters
- ✓Infrastructure is invisible when it works. You only notice when it breaks.
- ✓The goal isn't just "export"—it's "ready to use without manual cleanup"
Why I Built This (And Why I'm Not Sure What to Build Next)
After 3 years building AI products that didn't work, I finally admitted something:
I don't know what people actually need.
So I'm building Pactify in public—starting with my own problem (2-4 hours/week wasted organizing AI conversations).
But I'm openly asking: is this even a real problem worth solving?
What's New
- →Built minimum viable export: ChatGPT conversations → professional Word documents
- →Academic document standards: LaTeX formulas, tables, proper formatting
- →Enterprise-grade security: Temporary storage, auto-deletion, no third-party exposure
- →Chrome Extension: One-click export directly from ChatGPT interface
- →Started with my own pain point: AI saves 30 minutes, but organizing takes 90 minutes
What I Learned
- ✓Closed-door development never works—spent 3 years building things nobody used
- ✓Everyone's AI use cases are completely different: research, creative, legal, technical
- ✓The gap isn't AI capability—it's the bridge from "AI generated" to "actually usable"
- ✓Most export tools just save as plain text. When there are formulas, tables, code—everything breaks
- ✓Need to hear real user scenarios before building more features, not assume I know what they want
Help Me Build Better
I'm not trying to sell you something—I'm trying to understand if this direction makes sense. Your real-world scenarios and feedback matter more than anything else.