Back to Details

x-deep-research

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Deep Research

深度调研

Conduct citation-backed research using OpenAI's deep research models.
使用OpenAI的深度研究模型开展带引用的调研。

Agent Workflow

Agent工作流

bash
cd ~/brain/git/personal/agent-instructions
0. Ask model → 1. Submit → 2. Poll status → 3. Download → 4. Post-process
bash
cd ~/brain/git/personal/agent-instructions
0. 询问模型 → 1. 提交任务 → 2. 轮询状态 → 3. 下载结果 → 4. 后处理

0. Ask User: Model Selection (REQUIRED)

0. 询问用户：模型选择（必填）

Before submitting, ask the user which model to use:
ModelQualityCostUse When
o3-deep-researchHigher$1-3Important research, external-facing, deep analysis
o4-mini-deep-researchGood$0.20-0.60Quick lookups, internal use, cost-sensitive
Present this choice and wait for user response before proceeding.
提交任务前，需询问用户使用哪个模型：
模型质量成本适用场景
o3-deep-research更高$1-3重要调研、对外输出、深度分析
o4-mini-deep-research良好$0.20-0.60快速查询、内部使用、成本敏感
向用户展示该选项，等待用户回复后再继续。

1. Submit

1. 提交任务

bash
poetry run python scripts/deep_research.py submit \
  --template company \
  --topic "Company Name" \
  --model o3-deep-research \
  --output ~/brain/obsidian/Timatron/Raw\ Transcripts\ \&\ Research/research/
Templates: 
company
, 
person
, 
product
, 
custom
(use 
--query
for custom) Models: 
o3-deep-research
(default), 
o4-mini-deep-research
bash
poetry run python scripts/deep_research.py submit \
  --template company \
  --topic "Company Name" \
  --model o3-deep-research \
  --output ~/brain/obsidian/Timatron/Raw\ Transcripts\ \&\ Research/research/
模板选项：
company
、
person
、
product
、
custom
（使用
--query
参数自定义） 模型选项：
o3-deep-research
（默认）、
o4-mini-deep-research

2. Poll Status

2. 轮询状态

bash
poetry run python scripts/deep_research.py status <response_id>
Research takes 5-30 minutes. Poll every few minutes until 
completed
.
bash
poetry run python scripts/deep_research.py status <response_id>
调研耗时5-30分钟。每隔几分钟轮询一次，直到状态变为
completed
。

3. Download

3. 下载结果

bash
poetry run python scripts/deep_research.py download <response_id> \
  --output ~/brain/obsidian/Timatron/Raw\ Transcripts\ \&\ Research/research/
Report usage stats to user (shown after download):
  • Model, duration, token counts, cost
bash
poetry run python scripts/deep_research.py download <response_id> \
  --output ~/brain/obsidian/Timatron/Raw\ Transcripts\ \&\ Research/research/
向用户报告使用统计数据（下载完成后显示）：
  • 模型、耗时、token数量、成本

4. Post-Process Citations (REQUIRED)

4. 引用后处理（必填）

After downloading, edit the report to convert parenthetical citations to inline links:
Before: 
The company raised $100M ([source.com](url)).
After: 
The company [raised $100M](url).
Also consolidate duplicate citations — one link per fact is sufficient.
下载完成后，编辑报告将括号内的引用转换为内联链接：
转换前： 
The company raised $100M ([source.com](url)).
转换后： 
The company [raised $100M](url).
同时合并重复引用——每个事实只需一个链接即可。

API Key Setup

API密钥配置

Edit 
~/.config/openai/profiles.json
:
json
{
  "default": "personal",
  "profiles": {
    "personal": {
      "api_key": "sk-proj-YOUR-KEY-HERE"
    }
  }
}
Use 
--profile <name>
to select a profile. Falls back to 
OPENAI_API_KEY
env var.
编辑
~/.config/openai/profiles.json
文件：
json
{
  "default": "personal",
  "profiles": {
    "personal": {
      "api_key": "sk-proj-YOUR-KEY-HERE"
    }
  }
}
使用
--profile <name>
参数选择配置文件。若未配置，将 fallback 到
OPENAI_API_KEY
环境变量。

Pricing Reference

定价参考

ModelInput/M tokensOutput/M tokens
o3-deep-research$10.00$40.00
o4-mini-deep-research$2.00$8.00
Typical query uses 50-100K tokens. Exact cost reported after download.
模型输入/百万tokens输出/百万tokens
o3-deep-research$10.00$40.00
o4-mini-deep-research$2.00$8.00
典型查询会使用50-100K tokens。具体成本将在下载完成后显示。

Prompt Templates

提示词模板

Located in 
prompts/
:
  • base.md
    — Common instructions (word limits, citation style, format rules)
  • company.md
    — Company research structure
  • person.md
    — Person research structure
  • product.md
    — Product research structure
模板文件位于
prompts/
目录下：
  • base.md
    — 通用指令（字数限制、引用格式、排版规则）
  • company.md
    — 公司调研结构模板
  • person.md
    — 人物调研结构模板
  • product.md
    — 产品调研结构模板

Troubleshooting

故障排除