Frontmatter
| id | 9725 |
| title | Switch Memory Core embedding provider to Gemini and optimize RAG migration |
| state | Closed |
| labels | enhancementai |
| assignees | tobiu |
| createdAt | Apr 5, 2026, 11:42 PM |
| updatedAt | Apr 5, 2026, 11:52 PM |
| githubUrl | https://github.com/neomjs/neo/issues/9725 |
| author | tobiu |
| commentsCount | 1 |
| parentIssue | null |
| subIssues | [] |
| subIssuesCompleted | 0 |
| subIssuesTotal | 0 |
| blockedBy | [] |
| blocking | [] |
| closedAt | Apr 5, 2026, 11:52 PM |
Switch Memory Core embedding provider to Gemini and optimize RAG migration
Closedenhancementai
tobiu assigned to @tobiu on Apr 5, 2026, 11:52 PM
tobiu
Apr 5, 2026, 11:52 PM
Migration strategy successfully refactored to perform instantaneous 1:1 hardware vector transfer from Chroma legacy payload explicitly to SQLite Native tables, bypassing the expensive TextEmbeddingService.
tobiu closed this issue on Apr 5, 2026, 11:52 PM
The user reported that the local Ollama LLM is continually bottlenecking the iMac hardware during heavy Memory Core vector processing (e.g. 4-hour migrations and post-turn summaries). To stabilize the OS architecture, we will offload all vector embedding to the cloud API.
Objectives:
ai/mcp/server/memory-core/config.mjsto setneoEmbeddingProvidertogeminiby default.syncMemoryChromaToNeo.mjsto detect if the source and target embedding providers match. If they do (both Gemini), bypass the entire TextEmbeddingService loop and fetch Chroma'sembeddingsnatively, performing a 1:1 instantaneous hardware dump of the vectors into SQLite.