Built GoGogot as a self-hosted AI agent with a small, easy-to-understand, extensible core.
The goal was to avoid heavy frameworks and keep the codebase simple enough that one person can read it, modify it, and add tools without digging through layers of abstractions.
It runs as a single Go binary, supports file editing, shell commands, web access, memory, and scheduled tasks, and can work with multiple model providers. Works with MiniMax 2.5 with ~0.01$ session cost.
The main idea is: small core, explicit behavior, easy to extend.
datagobes•1h ago
Love the “small agent, explicit tools” approach. Most agent frameworks feel like Rails in 2008. Great DX until you have to debug a production edge case. Curious how you’re handling JSON/tool-call validation with MiniMax 2.5; did you roll your own schema layer or just retry with a system prompt nudge?
aspasskiy•1h ago
The goal was to avoid heavy frameworks and keep the codebase simple enough that one person can read it, modify it, and add tools without digging through layers of abstractions.
It runs as a single Go binary, supports file editing, shell commands, web access, memory, and scheduled tasks, and can work with multiple model providers. Works with MiniMax 2.5 with ~0.01$ session cost.
The main idea is: small core, explicit behavior, easy to extend.