SynaBun was not built the traditional way. There was no whiteboard planning session, no sprint tickets, no architecture review committee. Every feature, every line of code, every architectural decision was made in conversation with Claude.
This blog exists to document that process. Not as a polished marketing narrative, but as a raw record of the research, experiments, dead ends, and breakthroughs that shaped SynaBun into what it is today.
How SynaBun Gets Built
The development process is straightforward. A problem comes up. We talk through it with Claude. Claude researches approaches, writes the code, tests it, and we iterate until it works. The entire conversation is the development process.
Some sessions are quick fixes. Others are deep multi-hour explorations into embedding models, browser automation patterns, or hook system architecture. The interesting ones are the sessions where the initial approach fails and we have to rethink everything from scratch.
What You Will Find Here
Every post on this blog comes directly from our development sessions. When we solve a hard problem, discover something unexpected about how AI memory works, or make a significant architecture decision, we write about it. Expect to find:
- Architecture deep dives on how SynaBun's systems work under the hood, from SQLite vector storage to the hook lifecycle
- Research findings on embedding models, semantic search tuning, and memory retrieval patterns
- Browser automation insights from building a headed Chrome system that AI agents can actually use
- Development stories about the bugs that took hours to find and the features that surprised us
- Workflow patterns for getting the most out of AI-assisted development with persistent memory
Why Publish This
Building software with AI is still new territory. Most teams working this way are figuring it out privately. We think the patterns we are discovering, both the ones that work and the ones that do not, are worth sharing.
SynaBun is open source. The code is on GitHub. But code alone does not explain why we chose local embeddings over cloud APIs, why we built a headed browser instead of using headless mode, or why the hook system exists at all. That context lives in the conversations, and this blog is where we bring it to the surface.
New posts go up as we build. No schedule, no filler content. Just real research from real development sessions. If you are building with AI or thinking about persistent memory for your own tools, this is the place to follow along.