dsh-vision-router 3.0.4 → 3.1.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/presets/README.md CHANGED
@@ -1,36 +1,9 @@
1
- # presets/
1
+ # presets/ — runtime compatibility only
2
2
 
3
- 每个文件是 `$DSH_HOME/settings.yaml` 中 `llm-pi-ai` 段的一份即插即用片段:
4
- 把它合并进你自己的 `settings.yaml`(或复制整个段),填一个 `apiKeyEnv` 对应的 Key 即可。
5
- **仓库绝不内置任何第三方 Key**。
3
+ **DSH owns user provider/model registration, credentials and the model catalog.** Vision Router no longer maintains vendor-specific YAML examples, and the vision backend UI reads DSH's model list.
6
4
 
7
- > [!IMPORTANT]
8
- > DSH 当前的 Web「设置 → 模型 → 添加自定义提供方」表单不会写入图片输入能力元数据。即使端点实际是智谱 GLM-4.6V、Qwen-VL 等视觉模型,如果模型没有显式的 `input: [text, image]`(或提供方没有 `defaultInput: [text, image]`),DSH 仍会把它报告为纯文本模型,Vision Router 会为安全起见从视觉后端下拉中隐藏它。下面这些 preset 已经带好 `input` 声明;如果你是通过 Web 手动添加的自定义视觉提供方,请在 `$DSH_HOME/settings.yaml` 补上该字段。
5
+ **Do not delete `web-connection-ready/`.** It is a real private Web/Host startup bridge loaded through `cordis.patch.yml`. [PR #676](https://github.com/ysr666/dsh-vision-router/pull/676) introduced it for DSH alpha.2: old `webRuntime` and new `webStartup` own different trusted-host readiness generations. It must survive packaging and remains until exact old/new Host tests justify removal.
9
6
 
10
- - `dashscope.yaml` —— 阿里云百炼,大陆直连,新用户每系列 100 万 token/90 天(推荐首选)
11
- - `zhipu.yaml` —— 智谱 bigmodel.cn,`glm-4.6v-flash` 永久免费(provider key 用 `zhipu`,避免与 pi-ai 内置目录里的 `zai` 编码端点冲突)
12
- - `siliconflow.yaml` —— 硅基流动,¥14 赠金覆盖 Qwen2.5-VL
13
- - `openrouter.yaml` —— OpenRouter 免费模型(50 次/天,名单会轮换)
14
- - `ovh.yaml` —— OVHcloud AI Endpoints 匿名层:**无需账号、无需 Key**,匿名额度为每个 IP、每个模型每分钟 2 次;
15
- 该端点已内置为插件的默认 `httpProviders`,通常无需手动配置
7
+ [PR #89](https://github.com/ysr666/dsh-vision-router/pull/89) identified a genuine missing `image` declaration for custom DSH models. Prefer DSH model input types when available; for **verified visual models only**, older Hosts may need `llm-pi-ai` `input: [text, image]` and a reload. Verify actual image delivery with Test Vision instead of trusting metadata alone.
16
8
 
17
- 用法示例(以 dashscope 为例,合并进 `settings.yaml`):
18
-
19
- ```yaml
20
- llm-pi-ai:
21
- providers:
22
- dashscope:
23
- api: openai-completions
24
- baseURL: https://dashscope.aliyuncs.com/compatible-mode/v1
25
- apiKeyEnv: DASHSCOPE_API_KEY
26
- models:
27
- - id: qwen-vl-plus
28
- name: "百炼: Qwen-VL-Plus"
29
- contextWindow: 8000
30
- maxTokens: 8192
31
- input: [text, image]
32
- ```
33
-
34
- Key 放在 `~/.dsh/.credentials.yaml`(或用环境变量导出),然后重启 `dsh web`。
35
- 之后既可在模型选择器里手动使用,也可把 `provider: dashscope` 加进本插件的
36
- `providers` 链路。
9
+ The built-in anonymous OVH fallback lives in `src/lib/core-primitives.js`'s `DEFAULT_HTTP_PROVIDERS`, not the old `presets/ovh.yaml` example. Retired example files can be read at [their original August 2026 commit](https://github.com/ysr666/dsh-vision-router/commit/16e29d07b943d2b2e1dfb615683ab07f24cc96af). See [architecture audit](../docs/architecture/model-catalog-ownership.md).
@@ -1,18 +0,0 @@
1
- # 阿里云百炼 DashScope(大陆直连,新用户每系列 100 万 token/90 天)
2
- llm-pi-ai:
3
- providers:
4
- dashscope:
5
- api: openai-completions
6
- baseURL: https://dashscope.aliyuncs.com/compatible-mode/v1
7
- apiKeyEnv: DASHSCOPE_API_KEY
8
- models:
9
- - id: qwen-vl-plus
10
- name: "百炼: Qwen-VL-Plus"
11
- contextWindow: 8000
12
- maxTokens: 8192
13
- input: [text, image]
14
- - id: qwen2.5-vl-7b-instruct
15
- name: "百炼: Qwen2.5-VL-7B-Instruct"
16
- contextWindow: 32768
17
- maxTokens: 8192
18
- input: [text, image]
@@ -1,17 +0,0 @@
1
- # OpenRouter 免费视觉模型(50 次/天,需 Key;免费名单轮换频繁,以官方为准)
2
- llm-pi-ai:
3
- providers:
4
- openrouter:
5
- baseURL: https://openrouter.ai/api/v1
6
- apiKeyEnv: OPENROUTER_API_KEY
7
- models:
8
- - id: google/gemma-4-31b-it:free
9
- name: "OpenRouter: Gemma 4 31B (free)"
10
- contextWindow: 262144
11
- maxTokens: 32768
12
- input: [text, image]
13
- - id: google/gemma-4-26b-a4b-it:free
14
- name: "OpenRouter: Gemma 4 26B A4B (free)"
15
- contextWindow: 262144
16
- maxTokens: 32768
17
- input: [text, image]
package/presets/ovh.yaml DELETED
@@ -1,14 +0,0 @@
1
- # OVHcloud AI Endpoints 匿名层:免注册、免 Key(2 次/分钟/IP,尽力而为)
2
- # 该端点已内置为插件的默认 httpProviders,通常无需手动配置;
3
- # 若想把它加进 llm-pi-ai 路由链(例如供模型选择器使用),可用本段。
4
- llm-pi-ai:
5
- providers:
6
- ovh:
7
- api: openai-completions
8
- baseURL: https://oai.endpoints.kepler.ai.cloud.ovh.net/v1
9
- models:
10
- - id: Qwen2.5-VL-72B-Instruct
11
- name: "OVHcloud: Qwen2.5-VL-72B-Instruct (免费匿名)"
12
- contextWindow: 128000
13
- maxTokens: 8192
14
- input: [text, image]
@@ -1,18 +0,0 @@
1
- # SiliconFlow 硅基流动(大陆直连,¥14 赠金覆盖 Qwen2.5-VL)
2
- llm-pi-ai:
3
- providers:
4
- siliconflow:
5
- api: openai-completions
6
- baseURL: https://api.siliconflow.cn/v1
7
- apiKeyEnv: SILICONFLOW_API_KEY
8
- models:
9
- - id: Qwen/Qwen2.5-VL-7B-Instruct
10
- name: "硅基流动: Qwen2.5-VL-7B-Instruct"
11
- contextWindow: 32768
12
- maxTokens: 8192
13
- input: [text, image]
14
- - id: Qwen/Qwen2.5-VL-32B-Instruct
15
- name: "硅基流动: Qwen2.5-VL-32B-Instruct"
16
- contextWindow: 32768
17
- maxTokens: 8192
18
- input: [text, image]
@@ -1,18 +0,0 @@
1
- # 智谱 bigmodel.cn(大陆直连;glm-4.6v-flash 永久免费)
2
- llm-pi-ai:
3
- providers:
4
- zhipu:
5
- api: openai-completions
6
- baseURL: https://open.bigmodel.cn/api/paas/v4
7
- apiKeyEnv: ZAI_API_KEY
8
- models:
9
- - id: glm-4.6v-flash
10
- name: "智谱: GLM-4.6V-Flash (永久免费)"
11
- contextWindow: 131072
12
- maxTokens: 8192
13
- input: [text, image]
14
- - id: glm-4.5v
15
- name: "智谱: GLM-4.5V (限时免费)"
16
- contextWindow: 131072
17
- maxTokens: 8192
18
- input: [text, image]