Upstream GBrain Study
Upstream GBrain Study
This section studies upstream gbrain as an architecture donor for Gormes.
It is not a porting instruction to copy GBrain wholesale. The goal is to
extract the useful ideas, name the failure modes, and apply them to a better
Go-native gormes-agent.
Study Snapshot
- Upstream studied:
/home/xel/git/sages-openclaw/workspace-mineru/gbrain - Upstream commit:
11abb24ddd2209f8622870c2e48dc9ef050ad749 - Gormes repo studied:
/home/xel/git/sages-openclaw/workspace-mineru/gormes-agent - Gormes commit:
8c173263eb7e13b7acd4c0e2145ede19e7a0a3f2 - Date: 2026-04-24
Documents
- Architecture maps the GBrain runtime, data model, search stack, skills layer, and Minions job system.
- Good and Bad lists the design moves worth stealing and the traps Gormes should avoid.
- Gormes Takeaways translates the study into concrete Gormes architecture decisions.
One-Line Read
GBrain’s best idea is not “Postgres brain.” It is the combination of contract-first operations, a brain-first agent loop, fat procedural skills, and a durable job ledger for deterministic work. Gormes should keep the single Go binary and typed tool contracts, then borrow those ideas in a smaller, auditable, SQLite-first shape.