GoTionAPI

Notion AI → OpenAI-compatible API server. Single Go binary, zero runtime dependencies.

Uses uTLS Chrome JA3 fingerprint impersonation to bypass Notion's bot detection — same technique as the Python cloudscraper approach, but compiled into a standalone binary.

v3.2.1 — Response filter (7-step), input filter (framework injection strip), bookend detection, broadened reasoning patterns.

Quick Start

# Linux/macOS
chmod +x gotionapi-linux-amd64
./gotionapi-linux-amd64

# Windows
gotionapi-windows-amd64.exe

On first run, paste your NOTION_ACCOUNTS JSON when prompted:

Input your key :
{
  "token_v2": "v03:your_token_here",
  "space_id": "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx",
  "user_id": "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx",
  "space_view_id": "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx",
  "user_name": "Your Name",
  "user_email": "you@example.com"
}

An API key is auto-generated and displayed:

✓ Account saved to accounts.json
🔑 API Key: sk-xxx...xxxx
   (saved to .apikey — use this as Bearer token)

Server starts on http://localhost:8000.

Features

API Endpoints (OpenAI-compatible)

Endpoint	Auth	Description
GET /	-	Root info page
GET /health	-	Health check (accounts, mode, status)
GET /v1/models	Bearer	List all registered models
POST /v1/chat/completions	Bearer	Chat completion (stream + non-stream)

Models (10 registered)

Friendly Name	Notion Workflow ID	Default
claude-sonnet4.6	almond-croissant-low	✅
claude-opus4.6	avocado-froyo-medium
claude-opus4.7	apricot-sorbet-high
claude-opus4.8	ambrosia-tart-high
gemini-2.5flash	vertex-gemini-2.5-flash
gemini-3.1pro	galette-medium-thinking
gpt-5.2	oatmeal-cookie
gpt-5.4	oval-kumquat-medium
gpt-5.5	opal-quince-medium
kimi-2.6	fireworks-kimi-k2.6

3 Operating Modes

Set via APP_MODE environment variable:

Mode	Description
lite	Minimal — single-turn, no history, lowest resource usage
standard	Full context in each request, no persistent storage
heavy (default)	Multi-turn conversations with SQLite persistence

Auth & Account Management

• API key: API_KEY env > .apikey file > auto-generate (if accounts exist)
• Account loading: accounts.json > NOTION_ACCOUNTS env > .env > interactive prompt
• CLI commands: apikey-reset, apikey-regenerate (via os.Args[1])
• Multi-account: load balancing across Notion accounts

Security & Anti-Detection

• uTLS fingerprinting — Chrome JA3/TLS fingerprint impersonation (bypass Cloudflare)
• Warm-up cookies — auto-collect __cf_bm, _cfuvid, notion_browser_id, device_id from notion.so→notion.com redirect chain
• Multi-domain cookie strategy — visits 5 domains for full cookie set
• Auth middleware — requireAuth() wrapper on all /v1/* endpoints

Response Filter (7 steps)

Notion AI responses often contain internal artifacts that leak into the output. GoTionAPI automatically cleans these through a 7-step pipeline:

Step	What	Example	Method
1	Notion markup	<lang primary="en-US">, <br>, attribute tails	Regex cleanup
1.5	Bookend detection	Simple greeting, no tools needed. at start + end	Regex match + strip both
2	Reasoning prefix	General knowledge question., Simple greeting in Indonesian.	10+ regex patterns
3	Content duplication	Same response appears twice (artifacted + clean)	Paragraph-level dedup
4	Trailing fragment	Reasoning meta-text in last paragraph	Pattern strip
5	Trailing line prefix	Reasoning text on last line	Line-level strip
6	Inline suffix	...😊Simple greeting in Indonesian.	Search last 80 chars

Applied to all response paths: non-stream (heavy + standard), stream chunks (heavy + standard), and record-map extraction.

Input Filter

Framework injection blocks are stripped from user messages before sending to Notion:

Block	Pattern	Purpose
memory-context	``	Hermes memory injection
hermes-memory	<hermes-memory>...</hermes-memory>	Hermes memory block
honcho-context	<honcho-context>...</honcho-context>	Honcho context block

Heavy Mode (SQLite Persistence)

Pass conversation_id in the request body to continue a conversation:

curl http://localhost:8000/v1/chat/completions \
  -H "Authorization: Bearer ***" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet4.6",
    "messages": [{"role": "user", "content": "What did I just say?"}],
    "conversation_id": "abc-123"
  }'

Conversations are stored in SQLite (DB_PATH, default ./data/conversations.db).

API Key Authentication

All /v1/* endpoints require Bearer token authentication. Public endpoints (/health, /) are unauthenticated.

# Include the API key in requests
curl http://localhost:8000/v1/models \
  -H "Authorization: Bearer sk-xxx...xxxx"

curl http://localhost:8000/v1/chat/completions \
  -H "Authorization: Bearer sk-xxx...xxxx" \
  -H "Content-Type: application/json" \
  -d '{"model": "claude-sonnet4.6", "messages": [{"role": "user", "content": "Hello!"}]}'

CLI Commands

# Regenerate API key
./gotionapi apikey-reset

# Same as above (alias)
./gotionapi apikey-regenerate

Key Loading Priority

1. API_KEY environment variable
2. .apikey file
3. Auto-generated on first run (if accounts exist)

Usage

# List models
curl http://localhost:8000/v1/models \
  -H "Authorization: Bearer sk-xxx...xxxx"

# Chat completion (non-stream)
curl http://localhost:8000/v1/chat/completions \
  -H "Authorization: Bearer sk-xxx...xxxx" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet4.6",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

# Chat completion (stream)
curl http://localhost:8000/v1/chat/completions \
  -H "Authorization: Bearer sk-xxx...xxxx" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "kimi-2.6",
    "messages": [{"role": "user", "content": "Hello!"}],
    "stream": true
  }'

# Health check (no auth required)
curl http://localhost:8000/health

Configuration

All via environment variables or .env file:

Variable	Default	Description
APP_MODE	heavy	lite, standard, or heavy
PORT	8000	Server listen port
API_KEY	auto-generated	Bearer token for auth (set to override)
NOTION_ACCOUNTS	—	JSON account config (fallback if no accounts.json)
DB_PATH	./data/conversations.db	SQLite path (heavy mode only)
NOTION2API_DEBUG	—	Set 1 for verbose logging

Build from Source

Requires Go 1.21+:

cd notion2api-go-mvp
go build -o gotionapi .

Releases

Pre-built binaries available on the Releases page:

• gotionapi-linux-amd64
• gotionapi-linux-arm64
• gotionapi-darwin-amd64
• gotionapi-darwin-arm64
• gotionapi-windows-amd64.exe

Architecture

Client (OpenAI SDK / curl)
  │
  ▼
GoTionAPI Server (Go binary)
  │  ┌──────────────────────┐
  │  │ Bearer Auth (API Key) │ ← auto-generated sk-* key
  │  │ uTLS Chrome JA3       │ ← impersonates Chrome TLS fingerprint
  │  │ HTTP/1.1               │
  │  └──────────┬─────────────┘
  ▼             ▼
Input Filter ← strips memory-context, hermes-memory, honcho-context
  │
  ▼
Notion API  ─  /api/v3/runInferenceTranscript
  │
  ▼
Response Filter (7 steps) ← markup, bookend, prefix, dedup, trailing, suffix
  │
  ▼
Client (clean OpenAI-compatible response)

Technical Details

• uTLS (github.com/refraction-networking/utls v1.6.7) for Chrome JA3 TLS fingerprint impersonation
• HTTP/1.1 only — strips h2 from ALPN because *utls.UConn breaks Go's h2 transport detection
• Fresh TLS spec per connection — UTLSIdToSpec(HelloChrome_Auto) + ApplyPreset(HelloCustom)
• Pure Go SQLite (modernc.org/sqlite) — no CGO required
• API Key auth — auto-generated on first run, stored in .apikey, configurable via API_KEY env var
• Response filter — 7-step pipeline: markup cleanup, bookend detection, reasoning prefix strip, deduplication, trailing fragment/line strip, inline suffix detection
• Input filter — strips `` and <hermes-memory> / `` blocks before forwarding to Notion
• Warm-up — visits notion.so→notion.com redirect chain to collect Cloudflare cookies (__cf_bm, _cfuvid, notion_browser_id, device_id)
• Multi-account — round-robin across registered Notion accounts

Hermes Integration

GoTionAPI can be used as a custom provider in Hermes Agent:

# ~/.hermes/config.yaml
custom_providers:
  local-notion:
    base_url: http://localhost:8000/v1
    api_key: sk-xxx...xxxx

License

MIT