Every AI session starts from scratch.

I use Claude Code for architectural planning, Gemini CLI for high-volume file processing, and Cursor for the final implementation. Each tool is world-class, but they are all essentially stateless islands.

When I switch from Claude to Gemini, I have to re-explain the task. I have to copy-paste the latest decisions. I have to manually update the todo list. This "context tax" was slowing me down and wasting tokens.

The Pain Points

1. Context Loss: Switching tools meant re-uploading the world.
2. Invisible Costs: No easy way to see how much I'd spent across tools in a single session.
3. Task Drift: Keeping todo.md updated was a manual chore that I often skipped.
4. Hallucination Loops: AI agents would occasionally get stuck in a "flatline" loop, repeatedly failing the same command while I wasn't looking.

The Solution: synlynk

I built synlynk as a "Context Switchboard." It's not another AI tool; it's the connective tissue between them.

By wrapping my existing tools (e.g., synlynk exec claude), I ensure that a single, unified project snapshot is always available. The AI tool is instructed to read .synlynk/context.md at the start of every session.

It keeps the memory alive, the costs visible, and the tasks in sync.

stdlib-only

I made a deliberate choice to keep synlynk a single-file Python CLI with zero dependencies. No pip install, no environment hell. Just a simple curl | bash and you're ready to build.

Stay tuned for more updates as we move toward the v0.2.0 release.