Notion2API

Notion AI → OpenAI-Compatible API

🌐 English | 中文

Notion2API reverse-engineers the Notion AI web interface and exposes it as a standard /v1/chat/completions endpoint, making it directly usable with Cherry Studio, Zotero, and any other OpenAI-compatible client.

---

Features

• OpenAI Compatible — Standard /v1/chat/completions endpoint, streaming (SSE) and non-streaming
• Three Operation Modes — Lite / Standard / Heavy to fit different use cases
• 9 AI Models — Claude Sonnet/Opus, GPT-5.x, Gemini, Kimi
• Thinking Panel — Reasoning process display for all models
• Search Panel — Web search queries and source links
• Multi-Account Pool — Round-Robin load balancing with cooldown failover
• Built-in Web UI — Minimalist design, ambient animations, dark mode
• Docker Ready — One-command deployment

---

Mode Comparison

Feature	Lite	Standard	Heavy
Memory	❌ None	✅ Client-managed	✅ Server-managed
Database	❌	❌	✅ SQLite
Thinking Panel	❌	✅	✅
Search Panel	❌	✅	✅
Rate Limit	30/min	25/min	20/min
Use Case	Simple Q&A	Short–mid conversations	Long-term conversations

Recommended: standard — full context, no database required.
Switch by setting APP_MODE in .env.

---

Quick Start

1. Get Notion Credentials

Choose the method that fits your situation:

Method A — F12 (if you already have a Notion web session)

1. Open https://www.notion.so/ai and log in
2. Press F12 → Application tab → Storage → Cookies → https://www.notion.so
3. Find token_v2 and copy its Value
4. Switch to the Console tab, paste and run scripts/extract_notion_info.js
5. The script outputs all required fields — paste the result into accounts.json

Method B — Browser-Assisted Login (no existing web session needed)

python login.py

This launches a temporary Chrome/Edge window, waits for you to sign in to Notion, then automatically extracts all credentials and writes them to accounts.json and .env.

python login.py --check          # verify a saved profile
python login.py --list           # list all saved profiles
python login.py --manual         # paste token_v2 manually if Chrome is unavailable
python login.py --profile work   # save under a named profile

Both methods write to accounts.json. Multiple accounts are supported — add more entries to the array to enable load balancing.

⚠️ accounts.json and .env contain credentials. Both are git-ignored — keep them private.

---

2. Configure .env

cp .env.example .env

At minimum, set:

APP_MODE=standard   # lite / standard / heavy

If using Heavy mode, also add:

SILICONFLOW_API_KEY=your_key_here

Heavy mode uses SiliconFlow's LLM to compress long conversations. Register free at https://siliconflow.cn.

---

3. Start the Service

Docker (Recommended)

docker-compose build --no-cache && docker-compose up -d

accounts.json is mounted as a volume — update accounts without rebuilding:

# After editing accounts.json:
docker-compose restart

Local Run

pip install -r requirements.txt
uvicorn app.server:app --host 0.0.0.0 --port 8000

Access the Web UI at http://localhost:8000.

---

Supported Models

Model Name	Description
claude-sonnet4.6	Best balance of speed and quality — most recommended
claude-opus4.6	Stronger reasoning, use sparingly
claude-opus4.7	Latest Claude, strongest reasoning
gpt-5.5	Latest GPT (Beta)
gpt-5.4	OpenAI model
gpt-5.2	OpenAI model
gemini-2.5flash	Native fast, no thinking delay — great for quick tasks
gemini-3.1pro	Google's strongest reasoning model
kimi-2.6	Moonshot AI (Beta)

Full list via API: GET http://localhost:8000/v1/models

---

API Usage

This project accepts any string as the API key (no format requirement).

Python Example

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:8000/v1",
    api_key="any-string"
)

response = client.chat.completions.create(
    model="claude-sonnet4.6",
    messages=[{"role": "user", "content": "Hello"}],
    stream=True
)

for chunk in response:
    print(chunk.choices[0].delta.content or "", end="")

Endpoints

Endpoint	Method	Description
/v1/chat/completions	POST	Chat completions (core)
/v1/models	GET	List available models
/health	GET	Health check (account pool status, uptime)
/	GET	Built-in Web UI

---

Web UI

Access http://localhost:8000 for the built-in Notion AI Studio interface:

• Conversation Management — Create, rename, delete, star/bookmark
• Model Selector — Grouped by provider (Anthropic / OpenAI / Google / Moonshot)
• Thinking Panel — Collapsible reasoning display with elapsed timer
• Search Panel — Collapsible web search queries and source links
• Ambient Animations — Weather effects: default / snow / rain / sunny / night
• Theme — Light / dark mode toggle
• Responsive — Mobile-friendly sidebar

Thinking and Search panels require standard or heavy mode.

---

Environment Variables

Variable	Description	Default
NOTION_ACCOUNTS	Notion credentials JSON array	Required
APP_MODE	lite / standard / heavy	heavy
API_KEY	Bearer token for client auth	(none)
DB_PATH	SQLite database path	./data/conversations.db
HOST	Bind address	0.0.0.0
PORT	Service port	8000
HOST_PORT	Docker host port	8000
ALLOWED_ORIGINS	CORS allowed origins	*
SILICONFLOW_API_KEY	Required for Heavy mode compression	(none)
DISABLE_RATE_LIMIT	Disable per-IP rate limiting	false
NOTION_CLIENT_VERSION	Override Notion client version header	23.13.20260228.0625
LOG_LEVEL	Logging level	INFO
TZ	Timezone	Asia/Shanghai

---

Docker Reference

# Start
docker-compose up -d

# View logs
docker-compose logs -f --tail=50

# Restart (e.g. after updating accounts.json)
docker-compose restart

# Update code and redeploy
git pull && docker-compose down && docker-compose build --no-cache && docker-compose up -d

# Stop
docker-compose down

Nginx Reverse Proxy (optional)

location / {
    proxy_pass http://127.0.0.1:8000;
    proxy_set_header Host $host;
    proxy_buffering off;
    proxy_cache off;
    proxy_read_timeout 300s;
}

---

FAQ

Thinking panel not showing?
Use APP_MODE=standard or heavy. Lite mode does not support Thinking or Search panels.

How do I switch modes?
Edit APP_MODE in .env, then restart: docker-compose restart

How do I add multiple accounts?
Edit accounts.json as an array — accounts are load-balanced automatically:

[
  {"token_v2": "token1", "space_id": "...", "user_id": "...", "space_view_id": "...", "user_name": "...", "user_email": "..."},
  {"token_v2": "token2", "space_id": "...", "user_id": "...", "space_view_id": "...", "user_name": "...", "user_email": "..."}
]

Getting 429 or Notion AI suspended?
Notion may throttle workspaces with unusual request patterns. Adding multiple accounts helps distribute load. Business Trial workspaces are especially prone to this.

Token expired?
Re-run python login.py or repeat the F12 steps to refresh credentials.

---

Compatibility

Due to Notion's own AI latency, expect ~3 seconds from request to first token.

Client	Status	Notes
Cherry Studio	✅ Full support	Recommended
Zotero Translation	✅ Full support	Slightly slow; sonnet model most accurate
Immersive Translate	⚠️ Not recommended	High latency
Claude Code	❌ Not supported	Uses Anthropic native API format

---

License

MIT License

---

If this project helps you, please give it a Star ⭐

Built with assistance from Claude Code.