Zleap-AI/SAG
A local-first knowledge-base app built on the SAG retrieval architecture, shipping a DeepSeek Harness plugin: ingest documents and web sources, then search, trace citations back to the original chunk, inspect the event-entity graph, and answer with cited outputs. The zleap-sag / dsh-sag integration exposes the same knowledge to DSH over REST, OpenAI-compatible chat or MCP.
SAG is a knowledge-base application built on the SAG (SQL-Retrieval Augmented Generation with query-time dynamic hyperedges) architecture — an original retrieval design that replaces both dense RAG and GraphRAG with event-entity indexing and query-time dynamic hyperedges. You upload a document once; SAG parses it, splits it into chunks, embeds it, extracts events and entities, and keeps every retrieval result connected to the original text. Capacities include file and web ingestion with background processing, cross-source or source-scoped search in Fast (vector / multi_es_fast) and Precise (multi) modes, source tracing back to the exact chunk, an inspectable event-entity knowledge graph, multi-turn cited agent chat, and integration through self-hosted REST/OpenAPI, OpenAI-compatible chat, MCP and the zleap-sag Python package. The @zleap-ai/dsh-sag plugin exposes that knowledge to a DSH agent so it can search SAG and cite the original text; the README also documents using SAG as an OpenAI-compatible model.
Install
dsh plugin --profile web add @zleap-ai/dsh-sag@0.2.0Copied verbatim from the README's 'DeepSeek Harness (dsh) plugin' quick start, which precedes it with 'After dsh-sag 0.2.0 is published to npm' and follows it with dsh plugin --profile web exec dsh-sag doctor and dsh --profile web. The README's plugin usage guide documents connection options and supported capabilities.
Compatibility
The plugin is the DSH integration for the SAG knowledge-base app; the README notes npm publication of 0.2.0 as the precondition for the plugin install (it now resolves on the registry). SAG itself is local-first and single-user, starting on SQLite and LanceDB with no external database, with a path to PostgreSQL/pgvector.
Details
- Repo: Zleap-AI/SAG
- Category: Memory
- Stars: 2568
- Version: GitHub repo Zleap-AI/SAG; npm package @zleap-ai/dsh-sag at 0.2.0 (registry check 2026-10-10; registry repository field resolves back to this repo). MIT license. SAG core v1.8.11; Python 3.11+; Node 20+.
Alternatives
863683348/dsh-plugin-focus · akslcw/dsh-negative-ledger · askdkc/kiokuko-dsh