Model Configuration
N.E.K.O. uses different AI models for different tasks. Each can be individually configured.
Model roles
| Role | Default | Env var | Purpose |
|---|---|---|---|
| Conversation | qwen-max | - | Character chat (offline mode) |
| Summary | qwen-plus | NEKO_SUMMARY_MODEL | Conversation summarization |
| Correction | qwen-max | NEKO_CORRECTION_MODEL | Text correction |
| Emotion | qwen-flash | NEKO_EMOTION_MODEL | Emotion analysis for expressions |
| Vision | qwen3-vl-plus-2025-09-23 | NEKO_VISION_MODEL | Image/screenshot understanding |
| Agent | qwen3.5-plus | NEKO_AGENT_MODEL | Agent task execution |
| Router | qwen-plus | NEKO_ROUTER_MODEL | Memory routing decisions |
| Semantic | text-embedding-v4 | NEKO_SEMANTIC_MODEL | Text embeddings for memory |
| Reranker | qwen-plus | NEKO_RERANKER_MODEL | Search result reranking |
| Setting proposer | qwen-max | NEKO_SETTING_PROPOSER_MODEL | Proposing setting updates |
| Setting verifier | qwen-max | NEKO_SETTING_VERIFIER_MODEL | Verifying setting updates |
Custom model endpoints
Each model role can use a custom API endpoint. This is configured in core_config.json or via the Web UI:
json
{
"conversationModel": "custom-model-name",
"conversationModelUrl": "https://custom-api.example.com/v1",
"conversationModelApiKey": "sk-xxxxx"
}When a custom URL/key is set, it overrides the global assist API provider for that specific role.
Computer Use models
Computer Use requires two vision models:
| Role | Default | Purpose |
|---|---|---|
| Planning model | qwen3-vl-plus-2025-09-23 | Analyze screenshots and plan actions |
| Grounding model | qwen3-vl-plus-2025-09-23 | Locate UI elements for clicking |
Configure via core_config.json:
json
{
"computerUseModel": "qwen3-vl-plus-2025-09-23",
"computerUseModelUrl": "https://dashscope.aliyuncs.com/compatible-mode/v1",
"computerUseModelApiKey": "sk-xxxxx",
"computerUseGroundModel": "qwen3-vl-plus-2025-09-23",
"computerUseGroundUrl": "https://dashscope.aliyuncs.com/compatible-mode/v1",
"computerUseGroundApiKey": "sk-xxxxx"
}Thinking mode configuration
Some models support "thinking" or "extended reasoning" modes. N.E.K.O. disables these by default for faster responses. The disable format varies by provider:
| Provider | Disable format |
|---|---|
| Qwen, Step, DeepSeek | {"enable_thinking": false} |
| GLM | {"thinking": {"type": "disabled"}} |
| Gemini 2.x | {"thinking_config": {"thinking_budget": 0}} |
| Gemini 3.x | {"thinking_config": {"thinking_level": "low"}} |
This is handled automatically in config/__init__.py based on the model name.
Image rate limiting
| Setting | Default | Description |
|---|---|---|
NATIVE_IMAGE_MIN_INTERVAL | 1.5s | Minimum interval between screen captures |
IMAGE_IDLE_RATE_MULTIPLIER | 5x | Multiplier when no voice activity |