dsh-council

精选清单收录维护状态: 活跃

a1exsun/dsh-council

DeepSeek Harness 多模型评议会：让多个模型独立作答、匿名互评，并生成可检查的最终决策。

前往 GitHub

$ dsh plugin add dsh-council

安装

dsh 没有统一的安装命令——把该插件 README 中的配置行加入你的 profile/patch 配置，然后重启即可。

了解安装方式

156

星标

Fork

TypeScript

语言

MIT

许可证

2026-08-30

创建于

2026-09-08

README

English · 简体中文

DSH Council

Independent answers. Anonymous reviews. One reasoned decision.

Multi-model deliberation inside DeepSeek Harness.

Installation · How it works · Configuration · Contributing

https://github.com/user-attachments/assets/e81dfa36-5d93-4efb-8c67-015a5e4d2179

How it works

01 · Answer

2–8 models respond independently in parallel.

02 · Review

1–8 reviewers compare anonymous answers, identify gaps, and rank them.

03 · Decide

One arbiter weighs the answers and reviews to produce a final recommendation.

Your models. Use configured DSH providers and credentials. Choose models separately for each role, with reuse across roles. English and Chinese follow DSH's language setting.

Inspectable results. Confidence notes, consensus, disagreements, blind spots, model identities, and average rankings. Original answers and reviews remain in DSH's child sessions.

Installation

1. Install DSH

Install DeepSeek Harness (dsh) by following the official installation guide.

2. Install DSH Council

Node.js	pnpm	DeepSeek Harness
`^22.19.0 \|\| >=24.0.0`	`11.7.0`	At least two configured, authenticated models

Install the prebuilt package from npm:

npx --yes @deepseek-ai/dsh@latest plugin --profile web add @a1exsun/dsh-council@0.1.0
npx --yes @deepseek-ai/dsh@latest web

Restart DSH Web if it is already running.

No source checkout or compilation is required. See Contributing for development setup.

Usage

Enter this command in a DSH Web conversation:

/council

Choose your panel

Select answerers, reviewers, and an arbiter. Selections apply to one run.

Ask your question

Each participant starts with fresh context. Include the information it needs in your question.

Example question

Compare PostgreSQL leasing with a managed message queue for three workers. Explain how each recovers when a worker crashes after an external side effect but before acknowledging the job. Recommend one design and state its assumptions.

Cost

4–17 model participants per run. Each may make multiple requests and use Web tools.

Privacy

Your question and intermediate answers go to selected providers. See Security.

Accuracy

Model agreement does not guarantee correctness. Review the evidence and confidence notes.

Need more AI models?

Configuration

Defaults and overrides

The defaults work without additional configuration. To adjust limits, set config on the dsh-council entry in your DSH profile:

- id: dsh-council
  config:
    answerMaxTokens: 16384
    reviewMaxTokens: 16384
    arbiterMaxTokens: 16384
    childTimeoutMs: 300000
    runTimeoutMs: 900000
    subagentProvider: spawn

Token limits apply per model request. Timeouts are in milliseconds; runTimeoutMs must be at least childTimeoutMs.

Contributing

Development setup · Tests · Bug reports →

Acknowledgements

Thanks to Andrej Karpathy for llm-council, which inspired this project.

_{Review dimensions also draw on OpenRouter Fusion.}