dsh-WrenAI
DiscoveredA self-hosted semantic layer for DeepSeek Harness — natural-language data questions with metric calibers riding on every result; single-SELECT guard, self-heal SQL. Implements the differentiated core
dsh-WrenAI — a self-hosted semantic layer for DeepSeek Harness
Ask data questions in natural language; get numbers with their metric calibers attached. The semantic model (dimensions/measures/metrics + 口径) rides between the question and the SQL — that's the line between governed analytics and raw text-to-SQL. 自然语言问数,数字带着指标口径一起回来。语义模型(维度/度量/指标+口径) 站在问题与 SQL 之间——这就是受治理的数据问数与裸 text-to-SQL 的分界线。
Implements the differentiated core of WrenAI (18.3k★ — GenBI / governed text-to-SQL) natively for DSH: the self-hosted semantic layer. Clean-room, zero upstream code. Not affiliated. Family: the 12-repo quant stack — dsh-TradingAgents · dsh-OpenBB · dsh-CCXT · dsh-quantstats · dsh-freqtrade …
Why (NL2SQL plugins exist — what's missing is the semantic layer)
| raw NL2SQL plugins | dsh-WrenAI | |
|---|---|---|
| Model sees | table schemas | semantic catalog — metrics with calibers, dimensions with business meaning |
| Trust | hope | 口径随行: every result cites the metric definitions it used |
| SQL safety | varies | single-SELECT guard + self-heal retry, deterministic execution (SQLite stdlib) |
| Self-hosted | — | fully local (semantic/models.yaml + SQLite), no cloud service |
Quick start
git clone https://github.com/vicfei/dsh-WrenAI && cd dsh-wrenai
python3 scripts/ask.py --init <klines.json> # load data + build semantic views
python3 scripts/ask.py --ask "最近30天的平均收盘价和年化波动率是多少?"
Real smoke (binance BTC/USDT 365d, 2026-11-17): avg close 75,870 · vol 36.7%
— with the vol metric's caliber formula (sqrt(252×mean(ret²)), 30d rolling,
non-centered simplification stated) attached to the answer. The first generated
SQL had a syntax error; the engine self-healed on the error feedback.
Status & roadmap
v0 semantic engine smoke-tested on live data. ROADMAP: M1 joins + metric versioning + caliber drift alarms; M2 MDL-style model files + multi-DB.
License & disclaimer
MIT; concept credit to WrenAI (multi-licensed upstream, core Apache-2.0) — zero upstream code used. Not investment advice.
Comments
Loading…
Similar plugins
by chenkezhen480
Semantic long-term memory for DeepSeek Harness: embedding-based retrieval over a persistent cross-session memory store, with model-facing tools, proactive per-question recall, automatic conversation summarization, and auto-selected embedding provider (apiKey present → API, otherwise local).
★ 0
Apache-2.0
TypeScript
Aug 17, 2026
dsh plugin --profile web add dsh-plugin-semantic-memoryby huchunlinnk
Recursive Self-Improvement engine for DeepSeek Harness — DSH maintains DSH: bounded perceive→integrate→verify→parity→repair→propose loop enforcing 1:1 upstream feature parity
★ 4
MIT
JavaScript
Aug 18, 2026
dsh plugin --profile web add dsh-desk-rsiby jasen215
DeepSeek Harness (DSH) plugin for self-improving AI agents: continual learning, persistent memory, cross-session knowledge, review-and-refine workflows, and automatic rollback.
★ 17
↓ 372/wk
MIT
TypeScript
Oct 14, 2026
dsh plugin --profile agent add dsh-continual-harnessby snaiheihei
Layered global and workspace memory for DeepSeek Harness, with a web panel to browse entries and tune distillation and consolidation, an index kept in context with records loaded on demand, and a per-
★ 0
MIT
TypeScript
Nov 2, 2026
dsh plugin --profile web add dsh-snail-memoryby tluoluo
Persistent memory for DeepSeek Harness: per-conversation notes, selective injection, project memory, semantic Vault search + automatic retrieval.
★ 4
↓ 168/wk
MIT
TypeScript
Aug 16, 2026
dsh plugin --profile web add dsh-persistby goodddGrades
Long-term memory and self-evolving skills for the DeepSeek Harness · DeepSeek Harness 的长期记忆与自我演进能力 —— Plain Markdown on disk, no vector store
★ 3
MIT
TypeScript
Sep 21, 2026
dsh plugin --profile web add @goodddgrades/dsh-behuman