grazie2api

Expose Grazie AI as standard OpenAI / Anthropic / Responses API endpoints.

Bring your own credentials. Get a local API server with 50+ models including Claude, GPT, Gemini, Grok, Llama, DeepSeek, Qwen, and more.

Features

• Three API formats: OpenAI Chat Completions, Anthropic Messages, OpenAI Responses
• 50+ models: All Grazie-supported models via dynamic profile listing
• One-click credential paste: Auto-detects JSON, key=value, or raw token formats
• Auto-refresh: JWT and tokens refresh automatically in the background
• Multi-credential pool: Round-robin, least-used, or most-quota rotation
• Tool/function calling: Full support across all three API formats
• Parameter forwarding: temperature, top_p, reasoning_effort (for thinking models)
• Quota monitoring: Real-time quota tracking per credential
• Webhook API: External services can inject credentials programmatically
• SillyTavern compatible: Handles prefill, trailing assistant messages, content-role quirks

Quick Start

pip install -r requirements.txt
cp config.example.yaml config.yaml
python main.py serve

Server starts at http://127.0.0.1:8800.

Adding Credentials

Option 1: Browser OAuth (interactive)

python main.py login

Opens a browser window for OAuth login. Credentials are saved automatically.

Option 2: Paste via Web UI

Open http://127.0.0.1:8800/credentials and paste your credential in any format:

{"jwt": "eyJ...", "refresh_token": "1//...", "license_id": "ABC123"}

or key=value:

jwt=eyJ...
refresh_token=1//...
license_id=ABC123

or just paste raw tokens — they're auto-detected.

Option 3: API / Webhook

# Paste endpoint (auto-parse any format)
curl http://127.0.0.1:8800/api/credentials/paste \
  -H "Content-Type: application/json" \
  -d '{"blob": "jwt=eyJ... refresh_token=1//... license_id=ABC123"}'

# Webhook (structured, for external activation services)
curl http://127.0.0.1:8800/api/credentials/webhook \
  -H "Content-Type: application/json" \
  -d '{"jwt": "eyJ...", "refresh_token": "1//...", "license_id": "ABC123"}'

Option 4: Config file

Add accounts in config.yaml for automatic OAuth login on startup:

accounts:
  - email: "your@email.com"
    password: "your_password"

API Usage

OpenAI Chat Completions

curl http://127.0.0.1:8800/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-4.6-sonnet",
    "messages": [{"role": "user", "content": "Hello!"}],
    "stream": true
  }'

Anthropic Messages

curl http://127.0.0.1:8800/v1/messages \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-4.6-sonnet",
    "messages": [{"role": "user", "content": "Hello!"}],
    "max_tokens": 1024
  }'

OpenAI Responses

curl http://127.0.0.1:8800/v1/responses \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.4",
    "input": "What is 2+2?"
  }'

Model List

curl http://127.0.0.1:8800/v1/models

Model Aliases

Use friendly names or full profile names:

Alias	Profile
claude-4.6-opus	anthropic-claude-4-6-opus
claude-4.6-sonnet	anthropic-claude-4-6-sonnet
gpt-5.4	openai-gpt-5-4
gemini-3.1-pro	google-gemini-3-1-pro
grok-4	xai-grok-4
deepseek-r1	deepseek-r1

Run GET /v1/models for the full list.

Configuration

Setting	Default	Description
server.host	127.0.0.1	Listen address
server.port	8800	Listen port
strategy	round_robin	Credential rotation strategy
tokens.jwt_margin_seconds	300	Refresh JWT this many seconds before expiry
credentials.cooldown_seconds	600	Cooldown period after errors
quota.refresh_interval_seconds	300	Quota check interval

See config.example.yaml for all options.

CLI

python main.py serve            # Start the API server
python main.py login            # Interactive browser OAuth login
python main.py serve --port 9000 --api-key my-secret

API Endpoints

Endpoint	Description
POST /v1/chat/completions	OpenAI Chat Completions
POST /v1/messages	Anthropic Messages
POST /v1/responses	OpenAI Responses
GET /v1/models	List available models
POST /api/credentials/paste	Add credential (auto-parse)
POST /api/credentials/webhook	Add credential (structured)
GET /api/credentials	List credentials (admin)
DELETE /api/credentials/{id}	Remove credential
GET /health	Health check
GET /credentials	Credential management UI

How It Works

1. OAuth PKCE: Uses RFC 7636 PKCE flow with localhost callback (RFC 8252) for browser-based login
2. License Discovery: Automatically finds your active subscription license
3. JWT Lifecycle: refresh_token → id_token (1h) → JWT (24h) — all auto-refreshed
4. Format Conversion: Translates between OpenAI/Anthropic/Responses formats and the upstream Grazie chat API
5. Message Sanitization: Fixes orphaned tool messages, trailing assistant prefills, and SillyTavern quirks

Supported Parameters

Parameter	Forwarded	Notes
temperature	Yes	Mutually exclusive with top_p
top_p	Yes	Only sent if no temperature
reasoning_effort	Yes	Only for thinking models (o3, o4)
tools / functions	Yes	Full tool calling support
stream	Yes	SSE streaming
top_k	No	Causes upstream 400
seed	No	Causes upstream 400
max_tokens	No	Managed by upstream
stop	No	Not supported by upstream

License

AGPL-3.0 — see LICENSE.

If you use this code in a network service, you must make the complete source code available to all users of that service under the same license.