OpenCook/opencook_config.example.yaml at main · OpenDataBox/OpenCook · 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
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
# OpenCook Configuration Template
# Copy this file to opencook_config.yaml and fill in your values.
#
#   cp opencook_config.example.yaml opencook_config.yaml
#   opencook interactive --config-file opencook_config.yaml
#   opencook run "Your task here" --config-file opencook_config.yaml
#
# Or set the config path once:
#   export OPENCOOK_CONFIG_FILE=opencook_config.yaml

# ------------------------------------------------------------
# Database Configuration
# ------------------------------------------------------------
databases:
    default: sqlite  # Which database to use: postgresql | sqlite | duckdb | clickhouse

    postgresql:
        port: 5432                              # PostgreSQL listening port
        install_folder: /path/to/postgresql     # PostgreSQL binary/install directory
        data_folder: /path/to/postgresql/data   # PostgreSQL data directory (PGDATA)
        user: your_db_user                      # OS user that owns the PostgreSQL process
        bash_path: /bin/bash                    # Path to bash used for DB management commands
        cpu_num: 8                              # Number of CPUs available for parallel tasks

    clickhouse:
        port: 9000                              # ClickHouse native TCP port
        install_folder: /path/to/clickhouse     # ClickHouse binary/install directory
        data_folder: /path/to/clickhouse/data   # ClickHouse data storage directory

    sqlite:
        install_folder: /path/to/sqlite         # Working directory for SQLite database files
        cpu_num: 8                              # Number of CPUs available for parallel tasks

    duckdb:
        install_folder: /path/to/duckdb         # Working directory for DuckDB database files

# ------------------------------------------------------------
# Skills System
# ------------------------------------------------------------
skills:
    enabled: true                       # Enable/disable the skills system globally
    # Built-in skills (open-cookbook/) are always loaded automatically.
    standalone_project_roots:
        - .opencook/skills              # Per-project skill packages (relative to repo root)
    standalone_user_roots:
        - ~/.opencook/skills            # Per-user global skill packages
    # extra_standalone_paths: []        # Additional root dirs; each child dir is a skill package
    # extra_standalone_packages: []     # Additional direct skill dirs (each contains SKILL.md)
    resource_limit: 50                  # Max number of skill resources loaded into context

# ------------------------------------------------------------
# Agent Configuration
# ------------------------------------------------------------
agents:
    plan_agent:
        model: plan_agent_model         # Model definition to use; references a key under `models`
        max_steps: 10                   # Maximum number of agent steps before forced termination
        run_steps: 0                    # Steps to run automatically before pausing for review (0 = manual)
        tools:                          # Tools available to this agent
            - bash
            - sequentialthinking
            - task_done

    code_agent:
        model: code_agent_model         # Model definition to use; references a key under `models`
        max_steps: 50                   # Maximum number of agent steps before forced termination
        run_steps: 1                    # Run 1 step automatically, then pause
        tools:                          # Tools available to this agent
            - skill
            - bash
            - sequentialthinking
            - str_replace_based_edit_tool
            - task_done
            - test_subagent
            - plan_subagent

    test_agent:
        model: test_agent_model         # Model definition to use; references a key under `models`
        max_steps: 10                   # Maximum number of agent steps before forced termination
        run_steps: 1                    # Run 1 step automatically, then pause
        tools:                          # Tools available to this agent
            - sequentialthinking
            - task_done
            - database_verify
            - database_execute

# ------------------------------------------------------------
# Model Providers
# Configure at least one provider. Each entry needs an api_key.
# `provider` selects the client adapter / API format.
# Common values: anthropic | openai | deepseek | zhipu | qwen | siliconflow | groq | minimax | azure | openrouter | google | ollama
# ------------------------------------------------------------
model_providers:
    # Anthropic (Claude): https://console.anthropic.com/
    anthropic:
        api_key: YOUR_ANTHROPIC_API_KEY         # API key from Anthropic Console
        provider: anthropic                     # Use Anthropic's native API format

    # OpenAI (Responses API — OpenAI-specific, NOT for other vendors): https://platform.openai.com/api-keys
    openai:
        base_url: https://api.openai.com/v1     # API base URL
        api_key: YOUR_OPENAI_API_KEY            # API key from OpenAI platform
        provider: openai                        # Uses OpenAI Responses API (openai-specific)

    # DeepSeek: https://platform.deepseek.com/
    deepseek:
        base_url: https://api.deepseek.com      # DeepSeek API endpoint
        api_key: YOUR_DEEPSEEK_API_KEY          # API key from DeepSeek platform
        provider: deepseek                      # Uses standard Chat Completions API

    # Zhipu AI (GLM series): https://open.bigmodel.cn/
    zhipu:
        base_url: https://open.bigmodel.cn/api/paas/v4  # Zhipu API endpoint
        api_key: YOUR_ZHIPU_API_KEY             # API key from Zhipu AI platform
        provider: zhipu                         # Uses standard Chat Completions API

    # Alibaba DashScope (Qwen series): https://dashscope.aliyun.com/
    # qwen:
    #     base_url: https://dashscope.aliyuncs.com/compatible-mode/v1
    #     api_key: YOUR_DASHSCOPE_API_KEY       # API key from Alibaba Cloud DashScope
    #     provider: qwen                        # Uses standard Chat Completions API

    # SiliconFlow (open-source model aggregator): https://siliconflow.cn/
    # siliconflow:
    #     base_url: https://api.siliconflow.cn/v1
    #     api_key: YOUR_SILICONFLOW_API_KEY     # API key from SiliconFlow platform
    #     provider: siliconflow                 # Uses standard Chat Completions API

    # Groq (ultra-fast inference): https://console.groq.com/
    # groq:
    #     base_url: https://api.groq.com/openai/v1
    #     api_key: YOUR_GROQ_API_KEY            # API key from Groq Console
    #     provider: groq                        # Uses standard Chat Completions API

    # MiniMax: https://platform.minimaxi.com/
    # minimax:
    #     base_url: https://api.minimax.chat/v1 # MiniMax API endpoint
    #     api_key: YOUR_MINIMAX_API_KEY         # API key from MiniMax platform
    #     provider: minimax                     # Uses standard Chat Completions API

    # Ollama (local models, no API key required): https://ollama.com/
    # ollama:
    #     base_url: http://localhost:11434/v1   # Local Ollama server address (optional)
    #     api_key: ollama                       # Placeholder; Ollama ignores this value
    #     provider: ollama                      # Uses the native ollama Python library (not Chat Completions API)

# ------------------------------------------------------------
# Model Definitions
# Each entry maps a logical name (used in agent config above)
# to a specific provider + model.
# ------------------------------------------------------------
models:
    plan_agent_model:
        model_provider: deepseek        # Must match a key in model_providers above
        model: deepseek-chat            # Model name string accepted by the provider API
        # Other examples: claude-opus-4-6 (anthropic), gpt-4o (openai),
        #                 glm-4-plus (zhipu), qwen3-coder-plus (qwen), llama3 (ollama)
        max_tokens: 4096                # Maximum tokens in the model's response
        temperature: 0.4                # Sampling temperature (higher = more creative)
        top_p: 1                        # Top-p nucleus sampling (1 = disabled)
        top_k: 0                        # Top-k sampling (0 = disabled)
        max_retries: 10                 # Number of API retry attempts on transient errors
        parallel_tool_calls: true       # Allow the model to call multiple tools simultaneously

    code_agent_model:
        model_provider: deepseek        # Must match a key in model_providers above
        model: deepseek-chat            # Model name string accepted by the provider API
        max_tokens: 8192                # Larger budget for code generation tasks
        temperature: 0.0                # Deterministic output recommended for code generation
        top_p: 1                        # Top-p nucleus sampling (1 = disabled)
        top_k: 0                        # Top-k sampling (0 = disabled)
        max_retries: 10                 # Number of API retry attempts on transient errors
        parallel_tool_calls: true       # Allow the model to call multiple tools simultaneously

    test_agent_model:
        model_provider: deepseek        # Must match a key in model_providers above
        model: deepseek-chat            # Model name string accepted by the provider API
        max_tokens: 4096                # Maximum tokens in the model's response
        temperature: 0.0                # Deterministic output recommended for verification tasks
        top_p: 1                        # Top-p nucleus sampling (1 = disabled)
        top_k: 0                        # Top-k sampling (0 = disabled)
        max_retries: 10                 # Number of API retry attempts on transient errors
        parallel_tool_calls: true       # Allow the model to call multiple tools simultaneously