Supported AI Providers / 支持的 AI 服务

lsbot 支持 16 种 AI 服务，涵盖国内外主流大模型平台及本地模型，按需切换。所有 provider 均通过 --provider 参数指定，也可在 lsbot onboard 交互式向导中选择。

lsbot supports 16 AI providers covering mainstream LLM platforms globally plus local models. Select via --provider flag or the lsbot onboard interactive wizard.

Provider List / 服务列表

#	Provider	名称	Default Model / 默认模型	Default Base URL	API Key URL
1	`deepseek`	DeepSeek (recommended / 推荐)	`deepseek-chat`	`https://api.deepseek.com/v1`	platform.deepseek.com
2	`qwen`	Qwen / 通义千问	`qwen-plus`	`https://dashscope.aliyuncs.com/compatible-mode/v1`	bailian.console.aliyun.com
3	`claude`	Claude (Anthropic)	`claude-sonnet-4-20250514`	Anthropic native API	console.anthropic.com
4	`kimi`	Kimi / Moonshot / 月之暗面	`kimi-k2.5`	`https://api.moonshot.cn/v1`	platform.moonshot.cn
5	`minimax`	MiniMax / 海螺 AI	`MiniMax-Text-01`	`https://api.minimax.chat/v1`	platform.minimaxi.com
6	`doubao`	Doubao / 豆包 (ByteDance)	`doubao-pro-32k`	`https://ark.cn-beijing.volces.com/api/v3`	console.volcengine.com/ark
7	`zhipu`	Zhipu / 智谱 GLM	`glm-4-flash`	`https://open.bigmodel.cn/api/paas/v4`	open.bigmodel.cn
8	`openai`	OpenAI (GPT)	`gpt-4o`	`https://api.openai.com/v1`	platform.openai.com
9	`gemini`	Gemini (Google)	`gemini-2.0-flash`	`https://generativelanguage.googleapis.com/v1beta/openai`	aistudio.google.com
10	`yi`	Yi / 零一万物 (Lingyiwanwu)	`yi-large`	`https://api.lingyiwanwu.com/v1`	platform.lingyiwanwu.com
11	`stepfun`	StepFun / 阶跃星辰	`step-2-16k`	`https://api.stepfun.com/v1`	platform.stepfun.com
12	`baichuan`	Baichuan / 百川智能	`Baichuan4`	`https://api.baichuan-ai.com/v1`	platform.baichuan-ai.com
13	`spark`	Spark / 讯飞星火 (iFlytek)	`generalv3.5`	`https://spark-api-open.xf-yun.com/v1`	console.xfyun.cn
14	`siliconflow`	SiliconFlow / 硅基流动 (aggregator)	`Qwen/Qwen2.5-72B-Instruct`	`https://api.siliconflow.cn/v1`	cloud.siliconflow.cn
15	`grok`	Grok (xAI)	`grok-2-latest`	`https://api.x.ai/v1`	console.x.ai
16	`ollama`	Ollama (local / 本地)	`llama3.2`	`http://localhost:11434/v1`	No API key needed / 无需密钥

Kimi thinking models / Kimi 思考模型： 默认模型已升级为 kimi-k2.5（思考模型）。lsbot 会自动在对话历史中保留 reasoning_content，确保 tool call 多轮对话正常工作。如需使用非思考模型，可指定 --model moonshot-v1-8k。

Aliases / 别名

以下别名可以替代 --provider 值：

Alias / 别名	Maps to / 对应
`anthropic`	`claude`
`moonshot`	`kimi`
`qianwen`, `tongyi`	`qwen`
`gpt`, `chatgpt`	`openai`
`glm`, `chatglm`	`zhipu`
`google`	`gemini`
`lingyiwanwu`, `wanwu`	`yi`
`bytedance`, `volcengine`	`doubao`
`iflytek`, `xunfei`	`spark`
`xai`	`grok`
`tencent`, `hungyuan`	`hunyuan`

Usage / 用法

# Interactive wizard / 交互式向导
lsbot onboard

# Command line / 命令行指定
lsbot relay --provider deepseek --api-key sk-xxx
lsbot gateway --provider openai --api-key sk-xxx --model gpt-4o

# Custom base URL / 自定义 API 地址
lsbot relay --provider siliconflow --api-key sk-xxx --base-url https://api.siliconflow.cn/v1

# Override default model / 覆盖默认模型
lsbot relay --provider qwen --api-key sk-xxx --model qwen-max

Named Providers / 命名 Provider 配置（推荐）

The recommended way to configure AI providers. Define each provider as a named entry, then reference by name.

推荐的 AI 配置方式。将每个 provider 定义为命名条目，然后按名称引用。

Config Format / 配置格式

providers:
  my-deepseek:
    provider: deepseek
    api_key: sk-xxx
    model: deepseek-chat
  my-kimi:
    provider: kimi
    api_key: ak-xxx
    model: kimi-k2.5
  my-minimax:
    provider: minimax
    api_key: sk-xxx
    base_url: https://api.minimax.chat/v1
    model: MiniMax-M2.5

relay:
  platform: wecom
  provider: my-kimi        # references named provider

ai:
  max_rounds: 100          # shared settings stay here
  mcp_servers: [...]

Resolution / 解析规则

When --provider is specified (CLI or config), lsbot resolves the name:

Exact key match — --provider my-kimi matches providers.my-kimi directly
Provider type match — --provider kimi scans entries for .provider == "kimi"
Backward compat — if no providers: map exists, falls back to ai: block

CLI flags --api-key, --base-url, --model still override individual fields on the resolved entry.

# Use named provider directly
lsbot relay --provider my-kimi

# Use provider type (finds first matching entry)
lsbot relay --provider kimi

# Override model on a named provider
lsbot relay --provider my-kimi --model moonshot-v1-8k

Benefits / 优势

Each provider carries its own api_key, base_url, model — no cross-contamination
--provider kimi won't inherit deepseek's base_url from the ai: block
Easy to switch between providers: just change relay.provider
Old ai: format still works unchanged (backward compatible)

Per-Platform AI Overrides / 按平台覆盖 AI 设置（旧格式）

lsbot 的独有功能：可以为不同消息平台或频道指定不同的 AI 服务。一个实例同时服务多个平台，每个平台使用最适合的模型。

A unique lsbot feature: assign different AI providers to different messaging platforms or channels. One instance serves multiple platforms, each using its best-fit model.

How It Works / 工作原理

在 ~/.lsbot.yaml 中配置 ai.overrides 数组。当收到消息时，lsbot 按以下优先级匹配：

platform + channel_id — 最精确，匹配特定平台的特定频道
platform only — 匹配该平台的所有消息
default — 无匹配时使用默认 ai 配置

Override matching priority when a message arrives:

platform + channel_id — most specific, matches a specific channel on a platform
platform only — matches all messages from that platform
default — falls back to the base ai config

Override Fields / 覆盖字段

每条 override 可覆盖以下字段（均为可选，未设置的字段从默认配置继承）：

Field / 字段	Description / 说明
`platform`	必填。匹配的平台名：`wecom`, `feishu`, `slack`, `telegram`, `discord`, `dingtalk`, `wechat` 等
`channel_id`	可选。匹配的频道/群组 ID，不填则匹配该平台所有消息
`provider`	覆盖 AI 服务商（如 `kimi`, `claude`, `openai`）
`api_key`	覆盖 API 密钥
`base_url`	覆盖 API 地址
`model`	覆盖模型名

Key Behavior: Provider Switch / 重要行为：切换 Provider

当 override 切换了 provider 时（如从 deepseek 切到 kimi），base_url 和 model 会自动清空，使用新 provider 的默认值。 这避免了错误地将 deepseek 的 API 地址传给 kimi。

When an override changes the provider (e.g. from deepseek to kimi), base_url and model are automatically cleared and the new provider's defaults are used. This prevents accidentally sending deepseek's API URL to kimi.

如果你需要同时覆盖 provider 和 model，必须在同一条 override 中同时指定：

overrides:
  - platform: wecom
    provider: kimi
    api_key: sk-xxx
    model: moonshot-v1-128k    # 必须显式指定，否则使用 kimi 默认的 moonshot-v1-8k

Example 1: Basic Override / 基础覆盖

默认使用 DeepSeek，企业微信使用 Kimi：

Default is DeepSeek, WeCom uses Kimi:

ai:
  provider: deepseek
  api_key: sk-deepseek-xxx
  model: deepseek-chat

  overrides:
    - platform: wecom
      provider: kimi
      api_key: sk-kimi-xxx

Result / 结果：

Message Source / 消息来源	Provider Used / 使用的 Provider	Model / 模型	Why / 原因
WeCom user sends "hi"	kimi	`moonshot-v1-8k` (kimi default)	override matches `platform: wecom`, provider changed → base_url & model auto-cleared → kimi defaults
Telegram user sends "hi"	deepseek	`deepseek-chat`	no override matches telegram → uses default
Slack user sends "hi"	deepseek	`deepseek-chat`	no override matches slack → uses default

Example 2: Override with Custom Model / 覆盖并指定模型

默认使用 DeepSeek，飞书使用 Claude Sonnet，Telegram 使用 GPT-4o-mini（节省成本）：

Default is DeepSeek, Feishu uses Claude Sonnet, Telegram uses GPT-4o-mini (cost saving):

ai:
  provider: deepseek
  api_key: sk-deepseek-xxx

  overrides:
    - platform: feishu
      provider: claude
      api_key: sk-ant-xxx
      model: claude-sonnet-4-20250514
    - platform: telegram
      provider: openai
      api_key: sk-openai-xxx
      model: gpt-4o-mini

Result / 结果：

Message Source	Provider	Model	Why
Feishu message	claude	`claude-sonnet-4-20250514`	override sets provider + model explicitly
Telegram message	openai	`gpt-4o-mini`	override sets provider + model explicitly
WeCom message	deepseek	`deepseek-chat`	no override → default

Example 3: Channel-Specific Override / 频道级覆盖

同一个 Slack 里，VIP 频道用 Claude，其他频道用 DeepSeek：

Same Slack workspace, VIP channel uses Claude, others use DeepSeek:

ai:
  provider: deepseek
  api_key: sk-deepseek-xxx

  overrides:
    - platform: slack
      channel_id: C06ABCDEF12          # VIP 频道
      provider: claude
      api_key: sk-ant-xxx
      model: claude-sonnet-4-20250514

Result / 结果：

Message Source	Provider	Model	Why
Slack #vip-channel (C06ABCDEF12)	claude	`claude-sonnet-4-20250514`	channel_id matches → most specific override wins
Slack #general (C09XYZABC34)	deepseek	`deepseek-chat`	channel_id doesn't match → no override → default
Telegram message	deepseek	`deepseek-chat`	no override → default

Example 4: Same Provider, Different Model / 同 Provider 不同模型

同一个 provider，不同平台使用不同模型。这时 provider 没有改变，所以 base_url 和 model 不会被清空：

Same provider for different platforms but with different models. Since provider doesn't change, base_url and model are inherited and only the explicitly set fields are overridden:

ai:
  provider: deepseek
  api_key: sk-deepseek-xxx
  model: deepseek-chat

  overrides:
    - platform: wecom
      model: deepseek-reasoner     # 企业微信用推理模型

Result / 结果：

Message Source	Provider	Model	Why
WeCom message	deepseek	`deepseek-reasoner`	same provider → inherits api_key & base_url, only model overridden
Telegram message	deepseek	`deepseek-chat`	no override → default

Example 5: Override Only API Key / 仅覆盖密钥

多个平台共用同一 provider，但使用不同的 API 密钥（用于账单拆分或配额隔离）：

Multiple platforms share the same provider but use different API keys (for billing or quota isolation):

ai:
  provider: openai
  api_key: sk-default-xxx
  model: gpt-4o

  overrides:
    - platform: discord
      api_key: sk-discord-team-xxx     # Discord 团队单独的 API Key
    - platform: telegram
      api_key: sk-telegram-team-xxx    # Telegram 团队单独的 API Key

Result / 结果：

Message Source	Provider	Model	API Key	Why
Discord message	openai	`gpt-4o`	`sk-discord-team-xxx`	same provider → only api_key overridden
Telegram message	openai	`gpt-4o`	`sk-telegram-team-xxx`	same provider → only api_key overridden
Slack message	openai	`gpt-4o`	`sk-default-xxx`	no override → default

Example 6: Custom Base URL (API Proxy) / 自定义 API 地址（代理）

使用第三方 API 代理或聚合平台的兼容端点：

Using a third-party API proxy or aggregator's compatible endpoint:

ai:
  provider: deepseek
  api_key: sk-deepseek-xxx

  overrides:
    - platform: wecom
      provider: openai
      api_key: sk-proxy-xxx
      base_url: https://my-proxy.example.com/v1    # 自定义代理地址
      model: gpt-4o

Result / 结果：

Message Source	Provider	Base URL	Model
WeCom message	openai	`https://my-proxy.example.com/v1`	`gpt-4o`
Other platforms	deepseek	`https://api.deepseek.com/v1` (default)	`deepseek-chat`

Note / 注意： 当 override 同时改变了 provider 和 base_url，必须显式指定 base_url，因为切换 provider 会自动清空继承的 base_url。

Example 7: Complex Multi-Platform Setup / 复杂多平台配置

一个 lsbot 实例服务 4 个平台，各用不同模型：

One lsbot instance serving 4 platforms with different models:

ai:
  provider: deepseek
  api_key: sk-deepseek-xxx
  model: deepseek-chat

  overrides:
    # 企业微信用 Kimi（中文对话优化）
    - platform: wecom
      provider: kimi
      api_key: sk-kimi-xxx

    # 飞书 VIP 群用 Claude（最强推理）
    - platform: feishu
      channel_id: oc_abc123
      provider: claude
      api_key: sk-ant-xxx
      model: claude-sonnet-4-20250514

    # 飞书其他群用 Qwen（性价比高）
    - platform: feishu
      provider: qwen
      api_key: sk-qwen-xxx
      model: qwen-plus

    # Telegram 用 Gemini（免费额度大）
    - platform: telegram
      provider: gemini
      api_key: AIza-xxx

Result / 结果：

Message Source	Provider	Model	Why
WeCom	kimi	`moonshot-v1-8k`	matches `platform: wecom`
Feishu #vip (oc_abc123)	claude	`claude-sonnet-4-20250514`	matches `platform: feishu` + `channel_id: oc_abc123` (most specific)
Feishu #general (oc_xyz789)	qwen	`qwen-plus`	matches `platform: feishu` (no channel_id match, falls to platform-only)
Telegram	gemini	`gemini-2.0-flash`	matches `platform: telegram`, model auto-set to gemini default
Discord	deepseek	`deepseek-chat`	no override → default
Slack	deepseek	`deepseek-chat`	no override → default

Override Resolution Algorithm / 覆盖解析算法

message arrives (platform="feishu", channel_id="oc_abc123")
  │
  ├─ Pass 1: scan overrides for platform="feishu" AND channel_id="oc_abc123"
  │   └─ found? → use this override ✓
  │
  ├─ Pass 2: scan overrides for platform="feishu" AND channel_id="" (empty)
  │   └─ found? → use this override ✓
  │
  └─ Pass 3: no match → use default ai config ✓

Override Field Merge Rules / 字段合并规则

if override.provider != default.provider:
    # Provider changed — clear inherited base_url and model
    # 切换了 provider — 清空继承的 base_url 和 model
    base_url = ""              → new provider's default
    model    = ""              → new provider's default

# Then apply non-empty override fields
# 然后应用 override 中非空的字段
if override.api_key != "":   api_key  = override.api_key
if override.base_url != "":  base_url = override.base_url   # explicit override
if override.model != "":     model    = override.model       # explicit override

Common Mistakes / 常见错误

Wrong: Switching provider without specifying base_url / 错误：切换 provider 但用了旧 base_url

# ❌ BAD — base_url will be cleared because provider changed,
#          but if you set base_url from default, it won't carry over
ai:
  provider: deepseek
  api_key: sk-deepseek-xxx
  base_url: https://api.deepseek.com/v1    # this will NOT carry to kimi

  overrides:
    - platform: wecom
      provider: kimi          # provider changes → base_url auto-cleared ✓
      api_key: sk-kimi-xxx    # kimi will use its own default base_url ✓
      # No need to worry — this is handled automatically

Wrong: Expecting channel override without platform / 错误：期望仅 channel_id 匹配

# ❌ BAD — channel_id without platform won't match anything
ai:
  overrides:
    - channel_id: C12345      # missing platform! this override will never match
      provider: claude

# ✅ GOOD — always include platform
ai:
  overrides:
    - platform: slack
      channel_id: C12345
      provider: claude

Ollama (Local Models / 本地模型)

Ollama 在本地运行开源大模型，无需 API 密钥，默认监听 http://localhost:11434。

Ollama runs open-source LLMs locally with no API key required, listening on http://localhost:11434 by default.

Setup / 安装

# Install Ollama / 安装
# macOS
brew install ollama

# Linux
curl -fsSL https://ollama.com/install.sh | sh

# Pull a model / 拉取模型
ollama pull llama3.2

# Start the server (if not running as a service) / 启动服务
ollama serve

# Stop the server / 停止服务
# macOS (installed via brew or .dmg): quit from menu bar, or:
launchctl stop com.ollama.ollama
# Linux (systemd):
sudo systemctl stop ollama
# Foreground process: Ctrl+C

Usage / 用法

# Default model (llama3.2) / 默认模型
lsbot relay --provider ollama

# Specify a model / 指定模型
lsbot relay --provider ollama --model mistral
lsbot relay --provider ollama --model qwen2.5:7b
lsbot relay --provider ollama --model deepseek-r1:8b

# Connect to a remote Ollama instance / 连接远程实例
lsbot relay --provider ollama --base-url http://192.168.1.100:11434/v1

Available Models / 可用模型

Run ollama list to see installed models. Popular choices:

Model	Size	Notes
`llama3.2`	3B	Default, good general-purpose / 默认，通用
`llama3.2:1b`	1B	Lightweight / 轻量
`mistral`	7B	Strong reasoning / 推理能力强
`qwen2.5:7b`	7B	Good Chinese support / 中文支持好
`deepseek-r1:8b`	8B	Code & reasoning / 代码与推理

Notes / 说明

All non-Claude providers use the OpenAI-compatible API format, making it easy to add new providers.
除 Claude 外，所有 provider 均使用 OpenAI 兼容 API 格式，便于扩展。
siliconflow is an aggregator platform that provides access to many open-source models (Qwen, DeepSeek, Llama, etc.) through a single API key.
siliconflow 是一个聚合平台，通过一个 API Key 即可访问多种开源模型（Qwen、DeepSeek、Llama 等）。
You can always override the default model with --model and the default API URL with --base-url.
可通过 --model 和 --base-url 覆盖默认模型和 API 地址。

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Supported AI Providers / 支持的 AI 服务

Provider List / 服务列表

Aliases / 别名

Usage / 用法

Named Providers / 命名 Provider 配置（推荐）

Config Format / 配置格式

Resolution / 解析规则

Benefits / 优势

Per-Platform AI Overrides / 按平台覆盖 AI 设置（旧格式）

How It Works / 工作原理

Override Fields / 覆盖字段

Key Behavior: Provider Switch / 重要行为：切换 Provider

Example 1: Basic Override / 基础覆盖

Example 2: Override with Custom Model / 覆盖并指定模型

Example 3: Channel-Specific Override / 频道级覆盖

Example 4: Same Provider, Different Model / 同 Provider 不同模型

Example 5: Override Only API Key / 仅覆盖密钥

Example 6: Custom Base URL (API Proxy) / 自定义 API 地址（代理）

Example 7: Complex Multi-Platform Setup / 复杂多平台配置

Override Resolution Algorithm / 覆盖解析算法

Override Field Merge Rules / 字段合并规则

Common Mistakes / 常见错误

Ollama (Local Models / 本地模型)

Setup / 安装

Usage / 用法

Available Models / 可用模型

Notes / 说明

FilesExpand file tree

AI-PROVIDERS.md

Latest commit

History

AI-PROVIDERS.md

File metadata and controls

Supported AI Providers / 支持的 AI 服务

Provider List / 服务列表

Aliases / 别名

Usage / 用法

Named Providers / 命名 Provider 配置（推荐）

Config Format / 配置格式

Resolution / 解析规则

Benefits / 优势

Per-Platform AI Overrides / 按平台覆盖 AI 设置（旧格式）

How It Works / 工作原理

Override Fields / 覆盖字段

Key Behavior: Provider Switch / 重要行为：切换 Provider

Example 1: Basic Override / 基础覆盖

Example 2: Override with Custom Model / 覆盖并指定模型

Example 3: Channel-Specific Override / 频道级覆盖

Example 4: Same Provider, Different Model / 同 Provider 不同模型

Example 5: Override Only API Key / 仅覆盖密钥

Example 6: Custom Base URL (API Proxy) / 自定义 API 地址（代理）

Example 7: Complex Multi-Platform Setup / 复杂多平台配置

Override Resolution Algorithm / 覆盖解析算法

Override Field Merge Rules / 字段合并规则

Common Mistakes / 常见错误

Ollama (Local Models / 本地模型)

Setup / 安装

Usage / 用法

Available Models / 可用模型

Notes / 说明