@vionwilliams/agent-os 1.0.0-alpha.12
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/LICENSE +60 -0
- package/README.md +479 -0
- package/dist/bdi-ui/assets/index-CaE63EOo.css +1 -0
- package/dist/bdi-ui/assets/index-DNBesSJc.js +9 -0
- package/dist/bdi-ui/index.html +13 -0
- package/dist/cli.js +608745 -0
- package/dist/panel-ui/assets/index-BX-Unh58.css +2 -0
- package/dist/panel-ui/assets/index-aYo4LOni.js +58 -0
- package/dist/panel-ui/index.html +13 -0
- package/dist/vendor/ripgrep/COPYING +3 -0
- package/dist/vendor/ripgrep/arm64-darwin/rg +0 -0
- package/dist/vendor/ripgrep/arm64-linux/rg +0 -0
- package/dist/vendor/ripgrep/arm64-win32/rg.exe +0 -0
- package/dist/vendor/ripgrep/x64-darwin/rg +0 -0
- package/dist/vendor/ripgrep/x64-linux/rg +0 -0
- package/dist/vendor/ripgrep/x64-win32/rg.exe +0 -0
- package/package.json +164 -0
package/LICENSE
ADDED
|
@@ -0,0 +1,60 @@
|
|
|
1
|
+
Agent-OS License
|
|
2
|
+
================
|
|
3
|
+
|
|
4
|
+
Copyright (c) 2026 VION WILLIAMS
|
|
5
|
+
|
|
6
|
+
This repository contains two categories of source code with different
|
|
7
|
+
licensing provenance. See the NOTICE section at the bottom of this file
|
|
8
|
+
for the fork-heritage disclosure.
|
|
9
|
+
|
|
10
|
+
--------------------------------------------------------------------------
|
|
11
|
+
MIT License — applies to all original Agent-OS contributions
|
|
12
|
+
--------------------------------------------------------------------------
|
|
13
|
+
|
|
14
|
+
Permission is hereby granted, free of charge, to any person obtaining a
|
|
15
|
+
copy of this software and associated documentation files (the "Software"),
|
|
16
|
+
to deal in the Software without restriction, including without limitation
|
|
17
|
+
the rights to use, copy, modify, merge, publish, distribute, sublicense,
|
|
18
|
+
and/or sell copies of the Software, and to permit persons to whom the
|
|
19
|
+
Software is furnished to do so, subject to the following conditions:
|
|
20
|
+
|
|
21
|
+
The above copyright notice and this permission notice shall be included
|
|
22
|
+
in all copies or substantial portions of the Software.
|
|
23
|
+
|
|
24
|
+
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
25
|
+
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
26
|
+
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
|
|
27
|
+
THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
28
|
+
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
|
|
29
|
+
FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
|
|
30
|
+
DEALINGS IN THE SOFTWARE.
|
|
31
|
+
|
|
32
|
+
--------------------------------------------------------------------------
|
|
33
|
+
NOTICE — Upstream Fork Heritage
|
|
34
|
+
--------------------------------------------------------------------------
|
|
35
|
+
|
|
36
|
+
Agent-OS began as a fork of Anthropic's Claude Code CLI. Substantial
|
|
37
|
+
portions of the execution-layer infrastructure (tool protocols, REPL
|
|
38
|
+
surface, MCP integration, permission system, telemetry scaffolding, etc.)
|
|
39
|
+
are derived from upstream Claude Code source.
|
|
40
|
+
|
|
41
|
+
The Agent-OS additions — including but not limited to:
|
|
42
|
+
|
|
43
|
+
* Multi-Model Router and Cost Intelligence subsystem
|
|
44
|
+
* Knowledge Forge (FCA-based knowledge compilation engine)
|
|
45
|
+
* Super-Coordination layer (production lines, task board,
|
|
46
|
+
knowledge sync, Auto-OS evolution)
|
|
47
|
+
* BYOK hardening and the removal of subscription / OAuth paths
|
|
48
|
+
|
|
49
|
+
are original works authored for Agent-OS and are licensed under the
|
|
50
|
+
MIT terms above.
|
|
51
|
+
|
|
52
|
+
Portions of this codebase derived from Anthropic's Claude Code remain
|
|
53
|
+
subject to the license terms under which Anthropic distributes that
|
|
54
|
+
software. Users intending to redistribute Agent-OS — particularly in
|
|
55
|
+
commercial contexts — should review Anthropic's upstream license and
|
|
56
|
+
consult legal counsel before doing so.
|
|
57
|
+
|
|
58
|
+
"Claude", "Claude Code", and "Anthropic" are trademarks of Anthropic, PBC.
|
|
59
|
+
Their use in this repository is limited to factual attribution of the
|
|
60
|
+
fork heritage and does not imply endorsement.
|
package/README.md
ADDED
|
@@ -0,0 +1,479 @@
|
|
|
1
|
+
# Agent-OS
|
|
2
|
+
|
|
3
|
+
> 智能任务通用工作系统 · 多模型路由 + 意图编排 + 多智能体协调 + DataHub + 可编程面板 + LLM 知识编译器
|
|
4
|
+
|
|
5
|
+
**Status**: `1.0.0-alpha.12` · Beta 测试版 · npm 公共包可用
|
|
6
|
+
|
|
7
|
+
---
|
|
8
|
+
|
|
9
|
+
## 这是什么
|
|
10
|
+
|
|
11
|
+
Agent-OS 是一套面向真实工作的 AI 操作系统。它把**多模型路由、意图编排、多智能体协调、DataHub 数据枢纽、可编程面板系统、LLM 知识编译器**合成到同一个工作界面里,让 AI 从一次性调用走向持续运转的任务系统。
|
|
12
|
+
|
|
13
|
+
它的目标不是“多一个会对话的窗口”,而是把任务从输入到执行到沉淀完整跑通:
|
|
14
|
+
|
|
15
|
+
- 一句任务进入系统
|
|
16
|
+
- 系统持续理解与分流
|
|
17
|
+
- 调度执行与状态回写
|
|
18
|
+
- 结果可视化呈现
|
|
19
|
+
- 经验沉淀为可复用知识
|
|
20
|
+
|
|
21
|
+
---
|
|
22
|
+
|
|
23
|
+
## 核心能力
|
|
24
|
+
|
|
25
|
+
| 模块 | 说明 |
|
|
26
|
+
|---|---|
|
|
27
|
+
| 多模型路由 | 统一接入 Anthropic / DeepSeek / Qwen / MiniMax / GLM / Moonshot / Ollama 与兼容端点 |
|
|
28
|
+
| 灵枢意图编排 | 自然语言意图进入状态机与事件总线,驱动任务分流与执行路径 |
|
|
29
|
+
| BDI Runtime v0.1 | 独立长期目标控制台,管理长期意图、判断依据、目标图谱、当前承诺、行动路线与手动复盘 |
|
|
30
|
+
| 多智能体协调层 | 生产线、任务看板、同步、自动化策略协作运行 |
|
|
31
|
+
| DataHub 数据枢纽 | 上层统一只读查询入口,连接 task / event / line / panel 等多源数据 |
|
|
32
|
+
| 可编程面板系统 | 通过 Block Primitive、模板、生成器、解析器构建持续工作的操作界面 |
|
|
33
|
+
| LLM 知识编译器 | 将文档与运行结果逐步沉淀为结构化知识资产 |
|
|
34
|
+
|
|
35
|
+
---
|
|
36
|
+
|
|
37
|
+
## 快速开始
|
|
38
|
+
|
|
39
|
+
### BDI 长期目标控制台
|
|
40
|
+
|
|
41
|
+
```bash
|
|
42
|
+
bun run dev:bdi-console
|
|
43
|
+
```
|
|
44
|
+
|
|
45
|
+
默认打开本地端口:`http://127.0.0.1:3222`。BDI 控制台不嵌入任务面板;它只在需要执行时把 active intention 编译为 task-system 工作对象。
|
|
46
|
+
|
|
47
|
+
### 0. 前置环境
|
|
48
|
+
|
|
49
|
+
```bash
|
|
50
|
+
node -v
|
|
51
|
+
bun -v
|
|
52
|
+
```
|
|
53
|
+
|
|
54
|
+
要求:
|
|
55
|
+
|
|
56
|
+
- Node.js `>= 20`
|
|
57
|
+
- Bun `>= 1.3.0`
|
|
58
|
+
|
|
59
|
+
### 1. npm 安装(推荐)
|
|
60
|
+
|
|
61
|
+
```bash
|
|
62
|
+
npm install -g @vionwilliams/agent-os
|
|
63
|
+
agent-os --version
|
|
64
|
+
```
|
|
65
|
+
|
|
66
|
+
### 2. 二进制安装
|
|
67
|
+
|
|
68
|
+
```bash
|
|
69
|
+
curl -fsSL https://raw.githubusercontent.com/VIONWILLIAMS/agent-os/main/scripts/install.sh | bash
|
|
70
|
+
agent-os --version
|
|
71
|
+
```
|
|
72
|
+
|
|
73
|
+
> 注意:二进制安装依赖 GitHub Releases 的 latest 标签;若 release 未同步到最新版本,优先使用 npm 安装。
|
|
74
|
+
|
|
75
|
+
### 3. 从源码运行(开发者)
|
|
76
|
+
|
|
77
|
+
```bash
|
|
78
|
+
git clone https://github.com/VIONWILLIAMS/agent-os.git
|
|
79
|
+
cd agent-os
|
|
80
|
+
bun install
|
|
81
|
+
bash scripts/build.sh
|
|
82
|
+
bun dist/cli.js --version
|
|
83
|
+
```
|
|
84
|
+
|
|
85
|
+
---
|
|
86
|
+
|
|
87
|
+
## Provider 配置示例
|
|
88
|
+
|
|
89
|
+
你可以使用统一 Router 变量:
|
|
90
|
+
|
|
91
|
+
```bash
|
|
92
|
+
export MODEL_PROVIDER="deepseek"
|
|
93
|
+
export MODEL_BASE_URL="https://api.deepseek.com"
|
|
94
|
+
export MODEL_API_KEY="your-key"
|
|
95
|
+
export MODEL_NAME="deepseek-v4-pro"
|
|
96
|
+
```
|
|
97
|
+
|
|
98
|
+
也可以使用 provider 专属 key:
|
|
99
|
+
|
|
100
|
+
```bash
|
|
101
|
+
export MODEL_PROVIDER="qwen"
|
|
102
|
+
export MODEL_BASE_URL="https://dashscope.aliyuncs.com/compatible-mode/v1"
|
|
103
|
+
export QWEN_API_KEY="your-key"
|
|
104
|
+
export MODEL_NAME="qwen-plus"
|
|
105
|
+
```
|
|
106
|
+
|
|
107
|
+
Anthropic 直连同样支持:
|
|
108
|
+
|
|
109
|
+
```bash
|
|
110
|
+
export ANTHROPIC_API_KEY="sk-ant-..."
|
|
111
|
+
```
|
|
112
|
+
|
|
113
|
+
本地 Ollama 示例:
|
|
114
|
+
|
|
115
|
+
```bash
|
|
116
|
+
export MODEL_PROVIDER="ollama"
|
|
117
|
+
export MODEL_BASE_URL="http://localhost:11434/v1"
|
|
118
|
+
export MODEL_NAME="qwen2.5-coder:32b"
|
|
119
|
+
```
|
|
120
|
+
|
|
121
|
+
### 图片生成(MiniMax `image-01`)
|
|
122
|
+
|
|
123
|
+
Agent-OS 内置了 `ImageGenerate` 工具(保存到 `~/.agent-os/generated-images/`):
|
|
124
|
+
|
|
125
|
+
```bash
|
|
126
|
+
export MINIMAX_API_KEY="your-key"
|
|
127
|
+
# 可选:覆盖图片接口地址(默认 https://api.minimax.chat/v1)
|
|
128
|
+
export MINIMAX_IMAGE_BASE_URL="https://api.minimax.chat/v1"
|
|
129
|
+
```
|
|
130
|
+
|
|
131
|
+
在会话里直接说:
|
|
132
|
+
|
|
133
|
+
```text
|
|
134
|
+
/image --ratio 16:9 未来城市夜景海报
|
|
135
|
+
```
|
|
136
|
+
|
|
137
|
+
### MiniMax Coding Plan(`understand_image` + `web_search`)
|
|
138
|
+
|
|
139
|
+
MiniMax 的 `coding-plan-vlm` / `coding-plan-search` 在 Agent-OS 里更适合作为 MCP 工具接入,而不是普通 `/model` 模型切换。现在可以直接用内置命令挂上官方 MCP 服务:
|
|
140
|
+
|
|
141
|
+
```text
|
|
142
|
+
/minimax setup
|
|
143
|
+
```
|
|
144
|
+
|
|
145
|
+
这会在 Agent-OS 的 MCP 配置里添加官方服务:
|
|
146
|
+
|
|
147
|
+
```text
|
|
148
|
+
uvx minimax-coding-plan-mcp -y
|
|
149
|
+
```
|
|
150
|
+
|
|
151
|
+
使用前请先准备:
|
|
152
|
+
|
|
153
|
+
```bash
|
|
154
|
+
export MINIMAX_API_KEY="your-key"
|
|
155
|
+
```
|
|
156
|
+
|
|
157
|
+
接入后,Agent-OS 可以通过 MCP 工具调用:
|
|
158
|
+
|
|
159
|
+
- `understand_image`
|
|
160
|
+
- `web_search`
|
|
161
|
+
|
|
162
|
+
查看状态:
|
|
163
|
+
|
|
164
|
+
```text
|
|
165
|
+
/minimax status
|
|
166
|
+
```
|
|
167
|
+
|
|
168
|
+
### 本地多会话与 Peer 消息
|
|
169
|
+
|
|
170
|
+
Agent-OS 现在支持在同一台机器上发现本地活跃会话,并通过本地 inbox socket 进行最小跨会话消息投递。
|
|
171
|
+
|
|
172
|
+
查看当前可发现的本地会话:
|
|
173
|
+
|
|
174
|
+
```text
|
|
175
|
+
/peers
|
|
176
|
+
```
|
|
177
|
+
|
|
178
|
+
查看单个 peer 的详情:
|
|
179
|
+
|
|
180
|
+
```text
|
|
181
|
+
/peers show pid:1234
|
|
182
|
+
```
|
|
183
|
+
|
|
184
|
+
向指定 peer 发送消息:
|
|
185
|
+
|
|
186
|
+
```text
|
|
187
|
+
/peers send pid:1234 继续第 2 步并汇报结果
|
|
188
|
+
```
|
|
189
|
+
|
|
190
|
+
查看当前 daemon、Remote Control、peer 的统一运行态:
|
|
191
|
+
|
|
192
|
+
```text
|
|
193
|
+
/runtime
|
|
194
|
+
```
|
|
195
|
+
|
|
196
|
+
在工具链中,也可以先调用 `ListPeers` 查看可用目标,再用 `SendMessage` 向:
|
|
197
|
+
|
|
198
|
+
- `uds:/path/to.sock` 本地会话
|
|
199
|
+
- `bridge:session_...` 远程控制会话
|
|
200
|
+
|
|
201
|
+
发送纯文本消息。
|
|
202
|
+
|
|
203
|
+
### 隐藏保留式 Remote Skills
|
|
204
|
+
|
|
205
|
+
如果你希望保留 remote skill 能力,但暂时不把它做成默认主入口,可以启用隐藏 registry 模式:
|
|
206
|
+
|
|
207
|
+
```bash
|
|
208
|
+
export AGENT_OS_ENABLE_REMOTE_SKILLS=1
|
|
209
|
+
```
|
|
210
|
+
|
|
211
|
+
初始化远程技能注册表:
|
|
212
|
+
|
|
213
|
+
```text
|
|
214
|
+
/remote-skills init
|
|
215
|
+
```
|
|
216
|
+
|
|
217
|
+
检查状态与已注册技能:
|
|
218
|
+
|
|
219
|
+
```text
|
|
220
|
+
/remote-skills status
|
|
221
|
+
/remote-skills list
|
|
222
|
+
```
|
|
223
|
+
|
|
224
|
+
注册表与缓存默认位于:
|
|
225
|
+
|
|
226
|
+
- `~/.agent-os/remote-skills.json`
|
|
227
|
+
- `~/.agent-os/remote-skills-cache/`
|
|
228
|
+
|
|
229
|
+
配置完成后,可以通过 `SkillTool` 以 `remote/<slug>` 的形式引用远程技能。
|
|
230
|
+
|
|
231
|
+
### 智能任务系统入口
|
|
232
|
+
|
|
233
|
+
Agent-OS 现在内置了一套 task-system 工作流入口,适合直接进入不同工作模式:
|
|
234
|
+
|
|
235
|
+
```text
|
|
236
|
+
/panel
|
|
237
|
+
/today
|
|
238
|
+
/task
|
|
239
|
+
/project
|
|
240
|
+
/weekly-review
|
|
241
|
+
```
|
|
242
|
+
|
|
243
|
+
它们会分别把你带进:
|
|
244
|
+
|
|
245
|
+
- `panel`:根据一句意图需求直接创建对应的智能工作面板
|
|
246
|
+
- `today-work`:整理今天工作、生成今日重点
|
|
247
|
+
- `task-triage`:把零碎事项归位成待办
|
|
248
|
+
- `project-push`:推进项目、阶段、里程碑和执行队列
|
|
249
|
+
- `weekly-review`:生成回顾、沉淀模板与知识资产
|
|
250
|
+
|
|
251
|
+
如果你已经在某个 task-system 面板里,面板动作会继续通过内部 `task-system://...` 协议回写对象层,而不是只做静态 UI 刷新。
|
|
252
|
+
|
|
253
|
+
### Web 工作台与移动端桥接
|
|
254
|
+
|
|
255
|
+
本地 Web 工作台默认监听 `3212`:
|
|
256
|
+
|
|
257
|
+
```bash
|
|
258
|
+
bun run dev:workbench
|
|
259
|
+
```
|
|
260
|
+
|
|
261
|
+
脚本会先探测 `http://127.0.0.1:3212/api/task-system/health`。如果服务已在线,会直接输出当前工作台、待确认动作、Agent 运行数和移动端待推送数量;如果未在线,会启动 panel server。
|
|
262
|
+
|
|
263
|
+
Android 真机 USB 调试时使用:
|
|
264
|
+
|
|
265
|
+
```bash
|
|
266
|
+
bun run dev:mobile-bridge
|
|
267
|
+
```
|
|
268
|
+
|
|
269
|
+
它会建立 `adb reverse tcp:3212 tcp:3212`,让手机 App 通过 `http://127.0.0.1:3212` 访问桌面服务,并复用同一个 health 检查确认桥接是否可用。
|
|
270
|
+
|
|
271
|
+
### Workshop 教学支持模式
|
|
272
|
+
|
|
273
|
+
Workshop 是独立于 task panel 的局域网教学支持通道,用于老师在同一网络内观察并修复学员本地 Agent-OS 环境。它只保留两档:
|
|
274
|
+
|
|
275
|
+
- `observe`:观察模式,只读采集诊断信息。
|
|
276
|
+
- `repair`:修复模式,配对授权后允许受控 patch、命令和重启请求。
|
|
277
|
+
|
|
278
|
+
学生端启动 Workshop Agent:
|
|
279
|
+
|
|
280
|
+
```bash
|
|
281
|
+
agent-os workshop agent --mode observe
|
|
282
|
+
# 或允许受控修复
|
|
283
|
+
agent-os workshop agent --mode repair --port 33401
|
|
284
|
+
```
|
|
285
|
+
|
|
286
|
+
老师端启动支持面板:
|
|
287
|
+
|
|
288
|
+
```bash
|
|
289
|
+
agent-os workshop panel
|
|
290
|
+
```
|
|
291
|
+
|
|
292
|
+
也可以直接扫描局域网端口:
|
|
293
|
+
|
|
294
|
+
```bash
|
|
295
|
+
agent-os workshop scan --ports 33401-33410
|
|
296
|
+
```
|
|
297
|
+
|
|
298
|
+
配对后可采集只读诊断:
|
|
299
|
+
|
|
300
|
+
```bash
|
|
301
|
+
agent-os workshop pair http://192.168.1.23:33401 --code 123456
|
|
302
|
+
WORKSHOP_TOKEN="wks_..." agent-os workshop observe http://192.168.1.23:33401
|
|
303
|
+
```
|
|
304
|
+
|
|
305
|
+
Workshop 默认端口为学生端 `33401`、老师端面板 `33400`。配对 token 默认 2 小时有效;所有密钥会脱敏,修复操作会写入 `~/.agent-os/workshop/sessions/`,并默认阻止 `sudo`、删除目录、任意 shell 与危险 git 操作。
|
|
306
|
+
|
|
307
|
+
### Task-System DataHub 视图
|
|
308
|
+
|
|
309
|
+
任务系统相关数据现在也已经接进 DataHub,可以作为后续 dashboard / panel / tool 的统一只读查询入口。当前内置视图包括:
|
|
310
|
+
|
|
311
|
+
- `task-system-overview`
|
|
312
|
+
- `project-velocity`
|
|
313
|
+
- `review-assets`
|
|
314
|
+
- `collaboration-overview`
|
|
315
|
+
|
|
316
|
+
这些视图覆盖:
|
|
317
|
+
|
|
318
|
+
- 输入、对象、会话、推荐、回顾的总体概览
|
|
319
|
+
- 项目阶段、里程碑、开放任务与阻塞情况
|
|
320
|
+
- 模板 / 知识 / 回顾资产沉淀
|
|
321
|
+
- 协作简报与执行单状态
|
|
322
|
+
|
|
323
|
+
---
|
|
324
|
+
|
|
325
|
+
## 启动与运行
|
|
326
|
+
|
|
327
|
+
```bash
|
|
328
|
+
# 交互模式
|
|
329
|
+
agent-os
|
|
330
|
+
|
|
331
|
+
# 一次性模式
|
|
332
|
+
agent-os -p "帮我拆解今天的产品任务"
|
|
333
|
+
|
|
334
|
+
# 查看帮助
|
|
335
|
+
agent-os --help
|
|
336
|
+
```
|
|
337
|
+
|
|
338
|
+
---
|
|
339
|
+
|
|
340
|
+
## 新用户首跑验收(建议 3 分钟完成)
|
|
341
|
+
|
|
342
|
+
```bash
|
|
343
|
+
# 1) 版本确认
|
|
344
|
+
agent-os --version
|
|
345
|
+
|
|
346
|
+
# 2) 配置一组 provider(示例:DeepSeek)
|
|
347
|
+
export MODEL_PROVIDER="deepseek"
|
|
348
|
+
export MODEL_BASE_URL="https://api.deepseek.com"
|
|
349
|
+
export MODEL_API_KEY="your-key"
|
|
350
|
+
export MODEL_NAME="deepseek-v4-pro"
|
|
351
|
+
|
|
352
|
+
# 3) 打开路由调试并做最小请求
|
|
353
|
+
MODEL_DEBUG=1 agent-os -p "Reply with exactly: pong" < /dev/null
|
|
354
|
+
```
|
|
355
|
+
|
|
356
|
+
若返回 `pong` 且日志里看到 `resolved config`,说明安装与路由链路正常。
|
|
357
|
+
|
|
358
|
+
---
|
|
359
|
+
|
|
360
|
+
## Router 配置优先级
|
|
361
|
+
|
|
362
|
+
Router 配置按优先级从高到低解析:
|
|
363
|
+
|
|
364
|
+
1. `MODEL_ROUTER`(快捷格式)
|
|
365
|
+
2. `MODEL_PROVIDER` / `MODEL_BASE_URL` / `MODEL_API_KEY` / `MODEL_NAME`
|
|
366
|
+
3. 项目级 `./model-router.json`
|
|
367
|
+
4. 用户级 router 配置文件
|
|
368
|
+
|
|
369
|
+
说明:
|
|
370
|
+
|
|
371
|
+
- 命中更高优先级后,低优先级不生效。
|
|
372
|
+
- 推荐在项目里使用 `model-router.json` 固化团队默认路由。
|
|
373
|
+
- 用户级默认目录为 `~/.agent-os/`(兼容变量可覆盖,详见 `docs/ROUTER.md`)。
|
|
374
|
+
|
|
375
|
+
---
|
|
376
|
+
|
|
377
|
+
## 当前已支持的主流 Provider
|
|
378
|
+
|
|
379
|
+
| Provider | 类型 | Base URL | 默认 model |
|
|
380
|
+
|---|---|---|---|
|
|
381
|
+
| Anthropic | `anthropic` | `https://api.anthropic.com` | `claude-sonnet-4-6` |
|
|
382
|
+
| DeepSeek | `deepseek` | `https://api.deepseek.com` | `deepseek-v4-pro` |
|
|
383
|
+
| Qwen | `openai-compatible` | `https://dashscope.aliyuncs.com/compatible-mode/v1` | `qwen-plus` |
|
|
384
|
+
| Moonshot | `openai-compatible` | `https://api.moonshot.cn/v1` | `moonshot-v1-8k` |
|
|
385
|
+
| GLM | `openai-compatible` | `https://open.bigmodel.cn/api/paas/v4` | `GLM-5.1` |
|
|
386
|
+
| MiniMax | `openai-compatible` | `https://api.minimax.chat/v1` | `MiniMax-M2.7-HighSpeed` |
|
|
387
|
+
| Ollama | `ollama` | `http://localhost:11434/v1` | 由 `MODEL_NAME` 指定 |
|
|
388
|
+
|
|
389
|
+
任何 OpenAI Chat Completions 兼容 endpoint 都可以接入。
|
|
390
|
+
|
|
391
|
+
DeepSeek V4 当前模型名为 `deepseek-v4-pro` 与 `deepseek-v4-flash`。旧的
|
|
392
|
+
`deepseek-chat`、`deepseek-reasoner` 仍保留为兼容别名,但 DeepSeek 官方已标记
|
|
393
|
+
将在 2026-07-24 退役。
|
|
394
|
+
|
|
395
|
+
---
|
|
396
|
+
|
|
397
|
+
## 常见问题(安装与首跑)
|
|
398
|
+
|
|
399
|
+
### 1) `agent-os --version` 显示旧版本
|
|
400
|
+
|
|
401
|
+
先确认命令来源:
|
|
402
|
+
|
|
403
|
+
```bash
|
|
404
|
+
which agent-os
|
|
405
|
+
type agent-os
|
|
406
|
+
```
|
|
407
|
+
|
|
408
|
+
如果你本地有自定义 alias/function,会覆盖 npm 全局安装。
|
|
409
|
+
|
|
410
|
+
### 2) `[provider] API error 401 invalid api key`
|
|
411
|
+
|
|
412
|
+
优先排查环境变量污染:
|
|
413
|
+
|
|
414
|
+
```bash
|
|
415
|
+
unset MODEL_API_KEY MINIMAX_API_KEY DEEPSEEK_API_KEY
|
|
416
|
+
```
|
|
417
|
+
|
|
418
|
+
然后只保留一组 provider 配置重试,并开启 `MODEL_DEBUG=1` 观察实际命中 provider/baseUrl/model。
|
|
419
|
+
|
|
420
|
+
### 3) 不确定配置文件放在哪里
|
|
421
|
+
|
|
422
|
+
默认看:
|
|
423
|
+
|
|
424
|
+
- `~/.agent-os/model-router.json`
|
|
425
|
+
- `~/.agent-os/providers.json`
|
|
426
|
+
|
|
427
|
+
更多细节见:[`docs/ROUTER.md`](./docs/ROUTER.md)。
|
|
428
|
+
|
|
429
|
+
---
|
|
430
|
+
|
|
431
|
+
## 主要目录
|
|
432
|
+
|
|
433
|
+
```text
|
|
434
|
+
src/
|
|
435
|
+
coordination/ # 协调层 + 灵枢 + DataHub + 面板引擎
|
|
436
|
+
services/api/router/ # 多模型路由
|
|
437
|
+
panel-ui/ # 面板前端
|
|
438
|
+
packages/
|
|
439
|
+
knowledge-forge/ # LLM 知识编译器
|
|
440
|
+
```
|
|
441
|
+
|
|
442
|
+
---
|
|
443
|
+
|
|
444
|
+
## 已知限制(Beta 阶段)
|
|
445
|
+
|
|
446
|
+
- 部分高复杂度任务在不同模型之间表现差异明显,建议根据任务类型设置路由策略。
|
|
447
|
+
- 当前主仓已完成 TypeScript 债清零,`bun run typecheck` / `bun run build` / `bun test` 均可通过;后续重点转向功能深化与体验打磨。
|
|
448
|
+
- 个别历史命名的环境变量仍保持兼容(后续版本会逐步统一)。
|
|
449
|
+
|
|
450
|
+
---
|
|
451
|
+
|
|
452
|
+
## 版本路线
|
|
453
|
+
|
|
454
|
+
- `alpha.7`:产品定位与 npm 介绍升级
|
|
455
|
+
- `alpha.8`:README 全量去品牌化清稿
|
|
456
|
+
- `alpha.9`:新用户首跑闭环补齐(版本一致性 / Router 文档 / 401 排障)
|
|
457
|
+
- `alpha.10`:新增 MiniMax 图片生成工具与 `/image` 命令
|
|
458
|
+
- `alpha.11`:补齐 peer messaging 用户流、`/runtime` 统一运行态,以及隐藏保留式 remote skills
|
|
459
|
+
- `alpha.12`:CRF Runtime、BDI、Agent Platform 与本地长期自治内核收敛到统一 RuntimeRoot / ActionContract / Evidence 链路
|
|
460
|
+
- `beta.1`:知识编译器与协调层深度联动稳定化
|
|
461
|
+
- `beta.2`:发布链路与文档体系标准化
|
|
462
|
+
|
|
463
|
+
完整变更记录见 [CHANGELOG.md](./CHANGELOG.md)。
|
|
464
|
+
|
|
465
|
+
---
|
|
466
|
+
|
|
467
|
+
## License
|
|
468
|
+
|
|
469
|
+
MIT License · 详见 [LICENSE](./LICENSE)
|
|
470
|
+
|
|
471
|
+
---
|
|
472
|
+
|
|
473
|
+
## 链接
|
|
474
|
+
|
|
475
|
+
- npm: <https://www.npmjs.com/package/@vionwilliams/agent-os>
|
|
476
|
+
- Repository: <https://github.com/VIONWILLIAMS/agent-os>
|
|
477
|
+
- Releases: <https://github.com/VIONWILLIAMS/agent-os/releases>
|
|
478
|
+
- Issues: <https://github.com/VIONWILLIAMS/agent-os/issues>
|
|
479
|
+
- Changelog: [CHANGELOG.md](./CHANGELOG.md)
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
:root{color:#152033;background:#f5f7fb;font-family:Inter,ui-sans-serif,system-ui,-apple-system,BlinkMacSystemFont,Segoe UI,sans-serif;line-height:1.5}*{box-sizing:border-box}body{min-width:320px;margin:0}button,textarea,a{font:inherit}button{cursor:pointer}button:disabled{cursor:not-allowed;opacity:.48}.bdi-shell{min-height:100vh}.topbar{background:#fff;border-bottom:1px solid #dfe6f0;justify-content:space-between;align-items:center;gap:24px;min-height:88px;padding:18px clamp(20px,4vw,44px);display:flex}.eyebrow{color:#66758d;letter-spacing:0;text-transform:uppercase;margin:0 0 4px;font-size:12px;font-weight:700}h1,h2{letter-spacing:0;margin:0}h1{font-size:28px}h2{font-size:23px}.topbar-actions{align-items:center;gap:10px;display:flex}.ghost-link,.icon-button,.primary-button,.action-strip button,.runtime-trace-link,.belief-item button{color:#24324a;background:#fff;border:1px solid #d8e1ee;border-radius:8px;justify-content:center;align-items:center;gap:8px;text-decoration:none;display:inline-flex}.ghost-link{min-height:40px;padding:0 14px}.icon-button{width:40px;height:40px}.workspace{grid-template-columns:minmax(280px,360px) minmax(0,1fr);gap:0;display:grid}.sidebar{background:#fff;border-right:1px solid #dfe6f0;min-height:calc(100vh - 88px);padding:24px}.new-intent label,.section-heading{color:#28364e;justify-content:space-between;align-items:center;margin-bottom:10px;font-size:14px;font-weight:800;display:flex}.new-intent textarea{resize:vertical;color:#223049;background:#f8fafc;border:1px solid #d8e1ee;border-radius:8px;outline:none;width:100%;min-height:140px;padding:13px 14px}.new-intent textarea:focus{border-color:#4c7a92;box-shadow:0 0 0 3px #4c7a9224}.primary-button{color:#fff;background:#153044;border-color:#153044;width:100%;min-height:42px;margin-top:12px;font-weight:800}.intent-list{margin-top:30px}.section-heading strong{color:#455875;text-align:center;background:#e9eef5;border-radius:999px;min-width:28px}.intent-row{text-align:left;background:#fff;border:1px solid #dfe6f0;border-radius:8px;width:100%;margin:0 0 10px;padding:12px;display:block}.intent-row.is-active{background:#edf7f9;border-color:#4c7a92}.intent-row span,.intent-row small{display:block}.intent-row span{color:#172238;font-weight:800}.intent-row small{color:#6b7c94;margin-top:4px}.console{min-height:calc(100vh - 88px);padding:28px clamp(20px,3.4vw,44px)}.intent-summary{border-bottom:1px solid #dfe6f0;grid-template-columns:minmax(0,1fr) minmax(260px,360px);align-items:start;gap:28px;padding-bottom:24px;display:grid}.intent-summary p{color:#52637d;max-width:780px;margin:10px 0 0}.summary-metrics{grid-template-columns:repeat(3,1fr);gap:10px;display:grid}.summary-metrics div{background:#fff;border:1px solid #dfe6f0;border-radius:8px;min-height:78px;padding:12px}.summary-metrics strong,.summary-metrics span{display:block}.summary-metrics strong{color:#18243a;font-size:17px}.summary-metrics span{color:#72829a;margin-top:6px;font-size:12px}.action-strip{border-bottom:1px solid #dfe6f0;flex-wrap:wrap;gap:10px;padding:18px 0;display:flex}.action-strip button,.runtime-trace-link{min-height:40px;padding:0 14px;font-weight:800}.action-strip button:first-child{color:#fff;background:#153044;border-color:#153044}.governance-lens{border-bottom:1px solid #dfe6f0;grid-template-columns:repeat(4,minmax(0,1fr));gap:10px;padding:16px 0;display:grid}.governance-lens article{background:#fbfcfe;border:1px solid #dfe6f0;border-radius:8px;min-height:104px;padding:12px}.governance-lens span,.governance-lens strong,.governance-lens p{display:block}.governance-lens span{color:#75859c;font-size:12px;font-weight:800}.governance-lens strong{color:#18243a;margin-top:6px}.governance-lens p{color:#5d6f88;margin:6px 0 0;font-size:12px}.notice-stack{gap:8px;margin:16px 0 0;display:grid}.notice,.error,.task-session,.empty-state{border-radius:8px;padding:11px 13px}.notice{color:#24613c;background:#eefaf1;border:1px solid #b9ddc7}.error{color:#9f2525;background:#fff1f1;border:1px solid #efb8b8}.task-session{color:#385371;background:#eef3fa;border:1px solid #cad7ea}.console-grid{grid-template-columns:repeat(2,minmax(0,1fr));gap:18px;margin-top:22px;display:grid}.surface{background:#fff;border:1px solid #dfe6f0;border-radius:8px;padding:18px}.surface-wide{grid-column:1/-1}.surface-title{color:#18243a;align-items:center;gap:8px;margin-bottom:14px;font-size:15px;font-weight:900;display:flex}.belief-list,.goal-tree,.plan-flow{gap:10px;display:grid}.belief-item,.goal-node,.plan-step,.review-panel{background:#fbfcfe;border:1px solid #e3e9f2;border-radius:8px;padding:13px}.belief-item{grid-template-columns:minmax(0,1fr) auto;align-items:start;gap:14px;display:grid}.belief-item p,.goal-node p,.plan-step p,.review-panel p{color:#374963;margin:0}.belief-item span,.goal-node span{color:#75859c;margin-top:6px;font-size:12px;display:block}.belief-item button{color:#22623d;min-height:32px;padding:0 10px;font-weight:800}.inline-actions{flex-wrap:wrap;justify-content:flex-end;gap:8px;display:flex}.goal-node strong,.plan-step strong,.review-panel strong{color:#172238;display:block}.goal-node p{color:#5d6f88;margin-top:8px;font-size:13px}.plan-step{grid-template-columns:32px minmax(0,1fr) auto;align-items:start;gap:12px;display:grid}.plan-step>span{color:#385371;background:#eaf0f4;border-radius:999px;justify-content:center;align-items:center;width:28px;height:28px;font-weight:900;display:inline-flex}.plan-step p{color:#66758d;margin-top:4px}.plan-step em{color:#7a8aa0;font-size:12px;font-style:normal}.review-panel div{flex-wrap:wrap;gap:8px;margin-top:12px;display:flex}.review-panel span{color:#405a78;background:#eef3fa;border-radius:999px;padding:5px 9px;font-size:12px}.empty-state{color:#74849b;border:1px dashed #ced8e6}.blank-console{background:#fff;border:1px solid #dfe6f0;border-radius:8px;max-width:720px;padding:28px}.blank-console p:last-of-type{color:#62738c}@media (width<=920px){.workspace,.intent-summary,.governance-lens,.console-grid{grid-template-columns:1fr}.sidebar{border-bottom:1px solid #dfe6f0;border-right:0;min-height:auto}.console{min-height:auto}}@media (width<=640px){.topbar{flex-direction:column;align-items:flex-start}.summary-metrics,.belief-item,.plan-step{grid-template-columns:1fr}}
|