ClawWork/.env.example at main · anchapin/ClawWork · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
# ============================================
# LiveBench Environment Variables
# ============================================
# Copy this file to .env and fill in your API keys
#
# IMPORTANT: Agent and Evaluator can use different API providers!
#

# ============================================
# AGENT MODEL API (for running the agent)
# ============================================
# This is used for the agent's main model (e.g., GLM-4.7, GPT-4, Claude)
# You can use OpenAI, SiliconFlow, or other OpenAI-compatible APIs

OPENAI_API_KEY=your-api-key-here
# OPENAI_API_BASE=https://api.openai.com/v1  # Default OpenAI
# OPENAI_API_BASE=https://api.siliconflow.com/v1  # Or SiliconFlow


# ============================================
# EVALUATION MODEL API (for scoring work)
# ============================================
# The evaluator uses GPT-4o to score agent work submissions
#
# RECOMMENDED: Use real OpenAI API for evaluation (most reliable)
# - Evaluation requires gpt-4o which may not be available on all providers
# - OpenAI's gpt-4o is reliable and has consistent quality
# - Evaluation is lower volume than agent calls (less cost)
#
# If not set, falls back to OPENAI_API_KEY and OPENAI_API_BASE above

# Option 1: Use OpenAI for evaluation (RECOMMENDED)
EVALUATION_API_KEY=your-openai-api-key-here
EVALUATION_API_BASE=https://api.openai.com/v1  # Default, can be omitted

# Option 2: Use same provider as agent
# (Just comment out EVALUATION_API_KEY and EVALUATION_API_BASE)

# Option 3: Use different model for evaluation
# EVALUATION_MODEL=gpt-4o  # Default, change if needed


# ============================================
# PRODUCTIVITY TOOLS APIs
# ============================================

# Web Search API (Required for search_web and learn_from_web tools)
# Provider options: "tavily" (default, recommended) or "jina"
WEB_SEARCH_PROVIDER=tavily
WEB_SEARCH_API_KEY=your-tavily-api-key-here

# Tavily Search API (Recommended - more structured results with answers)
# Get API key at: https://tavily.com
# TAVILY_API_KEY=your-tavily-api-key-here

# Jina AI Search API (Alternative - markdown-based results)
# Get free API key at: https://jina.ai
# JINA_API_KEY=your-jina-api-key-here

# Qwen VL OCR API (for OCR processing when model does not support multimodal)
# Get API key from Alibaba Cloud DashScope: https://dashscope.aliyuncs.com/
OCR_VLLM_API_KEY=your-dashscope-api-key-here

# E2B API (for code sandbox execution)
# Get API key at: https://e2b.dev/
E2B_API_KEY=your-e2b-api-key-here

# ============================================
# SERVICE CONFIGURATION
# ============================================

# MCP Service Port
LIVEBENCH_HTTP_PORT=8010

# ============================================
# CONFIGURATION EXAMPLES
# ============================================

# Example 1: Use OpenAI for everything (simple, reliable)
# OPENAI_API_KEY=sk-proj-xxxxx
# EVALUATION_API_KEY=sk-proj-xxxxx  # Same or different OpenAI key
# WEB_SEARCH_API_KEY=tvly-xxxxx  # Tavily for search

# Example 2: Use SiliconFlow for agent, OpenAI for evaluation (cost-effective)
# OPENAI_API_KEY=sk-ngksq...  # SiliconFlow key
# OPENAI_API_BASE=https://api.siliconflow.com/v1
# EVALUATION_API_KEY=sk-proj-xxxxx  # Real OpenAI key for evaluation
# EVALUATION_API_BASE=https://api.openai.com/v1
# WEB_SEARCH_API_KEY=tvly-xxxxx  # Tavily for search

# Example 3: Use SiliconFlow for everything (if they support gpt-4o)
# OPENAI_API_KEY=sk-ngksq...
# OPENAI_API_BASE=https://api.siliconflow.com/v1
# WEB_SEARCH_API_KEY=tvly-xxxxx  # Tavily for search
# Note: Check if SiliconFlow supports gpt-4o or set EVALUATION_MODEL to supported model