Repository navigation
Configuration en
The configuration lives in plugins/PandaGPT/config.yml (currently config_version: 3). After editing, run /pandaadmin reload: the plugin fully validates and builds the new modules first and swaps them in only when everything succeeds; on any error the running service is kept and the reason is shown in chat and the console.
💡 New options added by an upgrade are filled into your file automatically; existing values are kept.
Language · Agent · Conversation · Model · Requests · Perception · Vision · Features · Integrations · Commands · Logging · Proxy
| Key | Default | Meaning |
|---|---|---|
panda.language.default |
zh_CN |
Language for the console and unmatched clients |
panda.language.per_player |
true |
Show plugin messages in each player's client language |
See Languages.
| Key | Default | Range | Meaning |
|---|---|---|---|
panda.agent.max_steps |
8 | 1–16 | Model requests per turn |
panda.agent.max_tool_calls |
16 | 1–48 | Tool calls per turn |
panda.agent.turn_timeout_seconds |
120 | 10–600 | Total time per turn, including all model requests and tools |
panda.agent.max_tool_result_chars |
6000 | 500–32000 | Truncation length of one tool result |
panda.agent.proactive |
true |
Agent decides and words proactive tips; off uses built-in text | |
panda.agent.disabled_skills |
[] |
Disabled skills, e.g. [commands, vision]
|
| Key | Default | Meaning |
|---|---|---|
panda.trigger_word |
panda |
Trigger word at the start of chat (panda or @panda); no spaces, must not start with @
|
panda.personality |
(Chinese persona) | Persona written into the system prompt |
panda.cooldown |
5000 | Per-player cooldown (ms) |
panda.cancel_chat_message |
true |
Hide the chat message that triggered Panda |
panda.history_rounds |
5 | Full Q&A rounds kept (0–50) |
panda.max_sessions |
1000 | Sessions kept at once |
panda.session_ttl_seconds |
1800 | Idle time before a session is cleared |
panda.max_prompt_chars |
2000 | Max characters per question |
panda.context_char_budget |
16000 | Context budget in characters (not exact tokens) |
panda.max_output_tokens |
1000 | Max output tokens per step |
panda:
model_selection:
provider: 'gemini'
<provider>:
api_key: ''
api_url: '...'
model: '...'
vision: trueAll 15 providers are listed in Model Providers. Keys can also come from PANDAGPT_<PROVIDER>_API_KEY.
| Key | Default | Range | Meaning |
|---|---|---|---|
panda.requests.workers |
4 | 1–32 | AI worker threads |
panda.requests.queue_size |
32 | 1–1000 | Queue size; requests are rejected when full |
panda.requests.connect_timeout_seconds |
10 | 1–60 | Connect timeout |
panda.requests.timeout_seconds |
45 | 1–180 | Overall request time including retries |
panda.requests.max_response_bytes |
1048576 | 1 KiB–4 MiB | Response body cap |
panda.requests.retries |
1 | 0–3 | Retries only for 429/502/503/504 and connection failures |
| Key | Default | Meaning |
|---|---|---|
panda.perception.enabled |
true |
Master switch for perception, goals and place memory |
panda.perception.observe_seconds |
30 | Interval of 32×18 background geometry observation |
panda.perception.scene_ttl_seconds |
60 | Observation lifetime |
panda.perception.entity_radius |
20 | Entity query radius |
panda.perception.max_entities |
16 | Entities kept |
panda.perception.max_players |
128 | Players with memory loaded at once |
panda.perception.proactive |
true |
Produce proactive tips |
panda.perception.suggestion_seconds |
900 | Minimum gap between routine tips |
panda.perception.routine_per_half_hour |
2 | Routine tips per half hour; 0 disables |
| Key | Default | Meaning |
|---|---|---|
panda.vision.enabled |
true |
Master switch for vision captures |
panda.vision.width / height
|
160 / 90 | Capture resolution (16–320 / 16–180) |
panda.vision.vertical_fov |
70 | Vertical field of view (30–110) |
panda.vision.max_distance |
96 | Max ray distance (8–128) |
panda.vision.rays_per_tick |
256 | Global rays per tick |
panda.vision.cells_per_tick |
8192 | Global cells per tick |
panda.vision.time_budget_ms |
2 | Soft time budget per tick (0.25–5) |
panda.vision.max_concurrent_captures |
2 | Concurrent captures (1–4) |
panda.vision.timeout_ticks |
200 | Cancel after this many ticks |
panda.vision.cooldown_seconds |
10 | Per-player cooldown |
panda.vision.movement_tolerance |
0.35 | Allowed movement during capture (blocks) |
panda.vision.angle_tolerance |
3 | Allowed head turn during capture (degrees) |
panda.vision.max_debug_particles |
64 | Debug particle cap |
panda.vision.save_images |
true |
Save the latest images to captures/<UUID>/
|
panda.vision.textures.enabled |
true |
Use real textures; off uses map colours |
panda.vision.textures.auto_download |
true |
Download vanilla-<version>.zip from Mojang automatically |
panda.vision.textures.packs |
[assets/vanilla-26.3.zip] |
1–8 packs; later ones override earlier ones |
panda.vision.textures.cache_mib |
64 | Decoded texture cache (8–256 MiB) |
| Key | Default | Meaning |
|---|---|---|
panda.features.preview_max_blocks |
2048 | Ghost preview block cap |
panda.features.preview_seconds |
180 | Seconds before a preview disappears |
panda.features.guidance_seconds |
600 | Max particle guidance time |
panda.features.max_reminders |
10 | Reminders per player; 0 disables |
panda.features.map_url |
'' |
Web-map template with {world} {x} {y} {z}; must start with http(s):// |
| Key | Default | Meaning |
|---|---|---|
panda.integrations.disabled |
[] |
Disabled integration ids |
panda.integrations.placeholder_denylist |
[ip, address, password, token, email, uuid] |
Placeholders containing these fragments are not resolved |
panda.integrations.block_history_days |
7 | Days of CoreProtect history |
See Command Proposals.
| Key | Default |
|---|---|
panda.command_awareness.enabled |
true |
panda.command_awareness.allow_execution |
false |
panda.command_awareness.executor |
player (only value) |
panda.command_awareness.max_executions |
3 |
panda.command_awareness.confirmation_ttl_seconds |
60 |
panda.command_awareness.allowed_commands |
[help] |
panda.command_awareness.blocked_commands |
[op, deop, stop, restart, reload, execute, sudo, panda, pandaadmin] |
| Key | Default | Meaning |
|---|---|---|
panda.logging.enabled |
true |
File logging; by default only UUIDs and events |
panda.logging.conversations |
false |
Log raw questions and answers only when enabled |
panda.logging.max_bytes |
1048576 | Size of one log file |
panda.logging.files |
3 | Rotated files kept |
| Key | Default | Meaning |
|---|---|---|
panda.proxy.enabled |
false |
HTTP CONNECT proxy (not SOCKS) |
panda.proxy.host / port
|
localhost / 10808 |
|
panda.proxy.username / password
|
'' |
Requires the JVM flag -Djdk.http.auth.tunneling.disabledSchemes=
|
The proxy is used for both model requests and official texture downloads.
⬅️ ⚡ Command Proposals · 📚 Contents · 🌍 Languages ➡️
🐼 PandaGPT · 源码 Source · 中文首页 · English Home
🐼 PandaGPT
简体中文
- 🏠 首页
- 📦 安装与升级
- 🤖 模型服务商
- 💬 和 Panda 对话
- 🧠 Agent 与工具
- 📜 技能与提示词
- 🔔 主动式助手
- 👁️ 视觉与感知
- ✨ 创意功能
- 🔌 插件集成
- ⚡ 命令提案
- ⚙️ 配置参考
- 🌍 多语言
- 🔑 权限与管理命令
- 🚀 性能
- 🛡️ 安全与隐私
- ❓ 常见问题
- 🛠️ 开发指南
English
- 🏠 Home
- 📦 Installation
- 🤖 Model Providers
- 💬 Talking to Panda
- 🧠 Agent & Tools
- 📜 Skills & Prompts
- 🔔 Proactive Assistant
- 👁️ Vision & Perception
- ✨ Creative Features
- 🔌 Plugin Integrations
- ⚡ Command Proposals
- ⚙️ Configuration
- 🌍 Languages
- 🔑 Permissions & Admin
- 🚀 Performance
- 🛡️ Privacy & Security
- ❓ FAQ
- 🛠️ Development