DSH Memgas
quqxui/dsh-memgas · productivity
DeepSeek Harness Long-Term Memory Plugin: Four Channels Search Integration + MemGAS. Memory construction and retrieval methods are from ICLR 2026 paper.
★ 7GitHub stars
1Forks
2026-09-08Last updated
TypeScriptLanguage
MITLicense
Key features
- DeepSeek Harness (dsh) long-term memory plugin.
- At least half of the seats in the final outcome will be left to both verbs and dense passages, with enhanced access only filling the remaining seats and rearranging.
- Association - Grouping by similarity when new memory is in the library, building parallels to truly relevant historical memories
- The summary and the re-use dsh model you've already got, doesn't need an additional API key or send memories to any third party service.
- Lets angent inter-session remember the engagement, decision-making and pedals of this project and keep it organized, updated and forgotten as it is used.
Requirements
- Local vector models are not validated by real models.
- The ONNX loading path is measured only on the injection of a fake run, and no real weight has been run by mean poping or dimension processing.
Install command
dsh plugin --profile web add github:quqxui/dsh-memgas