If you use multiple AI tools daily, you probably know this exact pain:
You spend 20 minutes brainstorming a brilliant database schema in Claude Web. Then you switch over to Cursor or Copilot to actually write the code... and the AI has Alzheimer's. It has no idea what you just discussed.
You end up manually copying and pasting context back and forth. It completely breaks the flow.
By day, I juggle 10+ hardware engineering projects, and I rely heavily on AI to survive my workload. The context fragmentation was driving me insane. So, over the last few weekends, I built a local memory layer to fix it.
I open-sourced it tonight. Meet Solvoke Synap 🧠.
The Solution in Action
Instead of a brittle two-way sync, I built a background engine that passively collects all my chats into one unified, searchable dashboard.
How I Built It (The Architecture)
I wanted something completely automated that didn't rely on unofficial, fragile APIs. Here is the stack:
-
The Web Catcher (Chrome Extension): Instead of dirty DOM scraping, the extension uses the
chrome.webRequestAPI to passively intercept chat API responses locally as I type in ChatGPT or Claude. - The IDE Catcher (VSCode/Cursor Extension): A local watcher that hooks into the session files of Cursor and Copilot.
- The Brain (Next.js + PostgreSQL): Everything funnels into a self-hosted Next.js app backed by a Postgres database, featuring Full-Text Search (FTS).
Zero Telemetry & Local First
Because AI conversations contain highly sensitive IP and code snippets, security was my top priority.
Data flows directly from the extensions to your self-hosted database. No middlemen, no cloud relays, no analytics. You own 100% of the data. To prove it, the entire data layer is open-sourced under AGPL-3.0.
What's Next? (Solvoke Corp)
Synap is just the infrastructure. Currently, I'm working on adding an AI-powered context fusion feature. The goal is to select multiple chats from the dashboard and push a cohesive summary directly into your IDE.
Ultimately, this is the data foundation for my broader vision: Solvoke Corp — a visual, multi-agent workspace for solopreneurs to turn unified context into working software.
Try it out!
It's still an early release (v0.1.2), but it has effectively eliminated my "copy-pasting contexts" routine.
Would love to hear your thoughts on the architecture, or if anyone else has solved this workflow problem differently!
🔗 GitHub Repo: https://github.com/Solvoke/Solvoke-Synap
🔗 Live Demo: https://synapdemo.solvoke.com/
(If you like where this is heading, dropping a ⭐️ on GitHub would mean the world to a solo dev!)
Top comments (2)
One surprising insight we've seen is that context fragmentation often stems from not standardizing prompt engineering across tools. In our experience with enterprise teams, aligning prompts with a common structure allows smoother transitions between AI systems like Claude, ChatGPT, and Cursor. This approach helps maintain coherence and cuts down on the "re-explaining" that's so time-consuming. The key is to focus on building a reusable prompt library tailored to your use cases, which can radically improve workflow efficiency. - Ali Muwwakkil (ali-muwwakkil on LinkedIn)
Hey, thanks for the insightful comment! I completely agree that having a standardized prompt library is a great foundational practice, especially for enterprise teams to align their baseline instructions.
However, as a solo dev, I found that the most valuable context isn't just the initial prompt—it's the organic, multi-turn brainstorming history. For example, dynamically evolving a complex DB schema over 20 turns in Claude is almost impossible to capture in a static prompt library.
That's exactly why I built Solvoke Synap. Instead of adding the friction of manually managing a prompt library, I wanted a background engine that automatically captures that messy, dynamic history so I can instantly search and pull it into Cursor.
But you're right—combining a solid prompt structure with an automated memory layer like Synap would definitely be the ultimate workflow combo! Thanks for sharing your experience!😄