OpenLobster - For Those Frustrated With OpenClaw Architecture
OpenLobster - For Those Frustrated With OpenClaw Architecture
When the dominant open-source desktop agent does not fit your mental model, you build your own. OpenLobster is what happens when someone chooses Go over Python and a graph database over flat files for agent memory.
Why Go
Most AI agent projects default to Python because the ML ecosystem lives there. But a desktop agent is not a training pipeline. It is a long-running application that manages concurrent operations - watching the screen, processing user input, executing actions, updating memory - simultaneously.
Go handles concurrency natively. Goroutines for parallel operations. Channels for communication between components. No GIL. No dependency hell. The binary deploys as a single file without requiring a Python environment. For a desktop application that needs to be fast and reliable, Go is a pragmatic choice.
Why Neo4j for Memory
Flat-file memory - markdown files, JSON blobs, SQLite tables - works for simple recall. But agent memory has relationships. A contact is connected to projects. Projects connect to tools. Tools connect to workflows. These relationships matter when the agent needs to reason about context.
Neo4j stores these relationships as first-class citizens. Querying "what tools does this person's project use" is a single traversal, not a multi-table join. For an agent that needs to understand how pieces of your work life connect, a graph database makes the relationships explicit and queryable.
Claude Code as the Build Tool
Building an agent with Claude Code is recursive - using an AI agent to build an AI agent. But it works remarkably well. Claude Code understands Go idioms. It generates correct Neo4j queries. It handles the boilerplate of desktop integration while you focus on the architecture decisions that matter.
The result is a functional alternative built in weeks instead of months.
Fazm is an open source macOS AI agent. Open source on GitHub.