@an-sdk/cli 0.0.8 → 0.0.10

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.
@@ -0,0 +1,110 @@
1
+ # Getting Started
2
+
3
+ ## 1. Get an API Key
4
+
5
+ Sign up at [an.dev](https://an.dev) and get your API key from [an.dev/agents/dashboard/api](https://an.dev/agents/dashboard/api).
6
+
7
+ ## 2. Install
8
+
9
+ ```bash
10
+ npm install @an-sdk/agent zod
11
+ ```
12
+
13
+ ## 3. Define Your Agent
14
+
15
+ Create `src/agent.ts`:
16
+
17
+ ```ts
18
+ import { agent, tool } from "@an-sdk/agent"
19
+ import { z } from "zod"
20
+
21
+ export default agent({
22
+ model: "claude-sonnet-4-6",
23
+ systemPrompt: "You are a helpful coding assistant.",
24
+ tools: {
25
+ greet: tool({
26
+ description: "Greet a user by name",
27
+ inputSchema: z.object({ name: z.string() }),
28
+ execute: async ({ name }) => ({
29
+ content: [{ type: "text", text: `Hello, ${name}!` }],
30
+ }),
31
+ }),
32
+ },
33
+ })
34
+ ```
35
+
36
+ ## 4. Login & Deploy
37
+
38
+ ```bash
39
+ npx @an-sdk/cli login
40
+ # Enter your API key: an_sk_...
41
+
42
+ npx @an-sdk/cli deploy
43
+ # Bundling src/agent.ts...
44
+ # Deploying my-agent...
45
+ # https://api.an.dev/v1/chat/my-agent
46
+ ```
47
+
48
+ Your agent is live.
49
+
50
+ ## 5. Embed in a React App
51
+
52
+ ```bash
53
+ npm install @an-sdk/nextjs @an-sdk/react ai @ai-sdk/react
54
+ ```
55
+
56
+ Create a token route (keeps your API key on the server):
57
+
58
+ ```ts
59
+ // app/api/an/token/route.ts
60
+ import { createAnTokenHandler } from "@an-sdk/nextjs/server"
61
+
62
+ export const POST = createAnTokenHandler({
63
+ apiKey: process.env.AN_API_KEY!,
64
+ })
65
+ ```
66
+
67
+ Add the chat UI:
68
+
69
+ ```tsx
70
+ // app/page.tsx
71
+ "use client"
72
+
73
+ import { useChat } from "@ai-sdk/react"
74
+ import { AnAgentChat, createAnChat } from "@an-sdk/nextjs"
75
+ import "@an-sdk/react/styles.css"
76
+ import { useMemo } from "react"
77
+
78
+ export default function Chat() {
79
+ const chat = useMemo(
80
+ () => createAnChat({
81
+ agent: "your-agent-slug",
82
+ tokenUrl: "/api/an/token",
83
+ }),
84
+ [],
85
+ )
86
+
87
+ const { messages, sendMessage, status, stop, error } = useChat({ chat })
88
+
89
+ return (
90
+ <AnAgentChat
91
+ messages={messages}
92
+ onSend={(msg) =>
93
+ sendMessage({ parts: [{ type: "text", text: msg.content }] })
94
+ }
95
+ status={status}
96
+ onStop={stop}
97
+ error={error}
98
+ />
99
+ )
100
+ }
101
+ ```
102
+
103
+ ## Next Steps
104
+
105
+ - [Defining Agents](./03-defining-agents.md) — Models, tools, hooks, permissions
106
+ - [React UI](./04-react-ui.md) — Theming, slots, tool renderers
107
+ - [Next.js Integration](./05-nextjs.md) — Server-side token handler
108
+ - [Node SDK](./06-node-sdk.md) — Sandboxes, threads, tokens from server code
109
+ - [CLI Reference](./07-cli.md) — All CLI commands
110
+ - [Custom Tools](./08-custom-tools.md) — Build custom tool renderers
@@ -0,0 +1,159 @@
1
+ # Defining Agents
2
+
3
+ Agents are defined with the `agent()` and `tool()` functions from `@an-sdk/agent`. These are config-only — they return exactly what you pass in, with type inference added. The actual execution happens in the AN runtime (E2B sandbox).
4
+
5
+ ## `agent(config)`
6
+
7
+ ```ts
8
+ import { agent } from "@an-sdk/agent"
9
+
10
+ export default agent({
11
+ // Model (default: "claude-sonnet-4-6")
12
+ model: "claude-sonnet-4-6",
13
+
14
+ // System prompt
15
+ systemPrompt: "You are a PR reviewer...",
16
+
17
+ // Custom tools (see below)
18
+ tools: { /* ... */ },
19
+
20
+ // Runtime: "claude-code" (default) or "codex"
21
+ runtime: "claude-code",
22
+
23
+ // Permission mode: "default", "acceptEdits", or "bypassPermissions"
24
+ permissionMode: "default",
25
+
26
+ // Max conversation turns (default: 50)
27
+ maxTurns: 50,
28
+
29
+ // Max budget in USD
30
+ maxBudgetUsd: 5,
31
+
32
+ // Lifecycle hooks (see below)
33
+ onStart: async () => {},
34
+ onToolCall: async ({ toolName, input }) => {},
35
+ onToolResult: async ({ toolName, input, output }) => {},
36
+ onStepFinish: async ({ step }) => {},
37
+ onFinish: async ({ result }) => {},
38
+ onError: async ({ error }) => {},
39
+ })
40
+ ```
41
+
42
+ ### Defaults
43
+
44
+ | Field | Default |
45
+ |-------|---------|
46
+ | `model` | `"claude-sonnet-4-6"` |
47
+ | `runtime` | `"claude-code"` |
48
+ | `permissionMode` | `"default"` |
49
+ | `maxTurns` | `50` |
50
+ | `tools` | `{}` |
51
+
52
+ ## `tool(definition)`
53
+
54
+ ```ts
55
+ import { tool } from "@an-sdk/agent"
56
+ import { z } from "zod"
57
+
58
+ const myTool = tool({
59
+ description: "What this tool does",
60
+ inputSchema: z.object({
61
+ path: z.string(),
62
+ verbose: z.boolean().optional(),
63
+ }),
64
+ execute: async (args) => {
65
+ // args is fully typed: { path: string; verbose?: boolean }
66
+ return {
67
+ content: [{ type: "text", text: "result" }],
68
+ }
69
+ },
70
+ })
71
+ ```
72
+
73
+ The `execute` function receives args typed from the Zod schema. Return value must have `content` array with `{ type: "text", text: string }` items.
74
+
75
+ ## Hooks
76
+
77
+ ### `onToolCall`
78
+
79
+ Runs before a tool executes. Return `false` to block it.
80
+
81
+ ```ts
82
+ onToolCall: async ({ toolName, input }) => {
83
+ if (toolName === "Bash" && input.command?.includes("rm -rf")) {
84
+ return false // blocked
85
+ }
86
+ }
87
+ ```
88
+
89
+ ### `onToolResult`
90
+
91
+ Runs after a tool executes with the result.
92
+
93
+ ```ts
94
+ onToolResult: async ({ toolName, input, output }) => {
95
+ console.log(`Tool ${toolName} returned:`, output)
96
+ }
97
+ ```
98
+
99
+ ### `onFinish`
100
+
101
+ Runs when the agent completes successfully.
102
+
103
+ ```ts
104
+ onFinish: async ({ result }) => {
105
+ await fetch("https://my-api.com/webhook", {
106
+ method: "POST",
107
+ body: JSON.stringify({ done: true }),
108
+ })
109
+ }
110
+ ```
111
+
112
+ ### `onError`
113
+
114
+ Runs when the agent encounters an error.
115
+
116
+ ```ts
117
+ onError: async ({ error }) => {
118
+ console.error("Agent failed:", error)
119
+ }
120
+ ```
121
+
122
+ ## Permission Modes
123
+
124
+ | Mode | Behavior |
125
+ |------|----------|
126
+ | `"default"` | Agent asks for confirmation on risky operations |
127
+ | `"acceptEdits"` | Auto-approve file edits, still confirm other actions |
128
+ | `"bypassPermissions"` | Auto-approve everything (use with caution) |
129
+
130
+ ## Type Reference
131
+
132
+ ```ts
133
+ // Agent config (all fields required — defaults filled by agent())
134
+ interface AgentConfig {
135
+ model: string
136
+ systemPrompt: string
137
+ tools: ToolSet
138
+ runtime: "claude-code" | "codex"
139
+ permissionMode: "default" | "acceptEdits" | "bypassPermissions"
140
+ maxTurns: number
141
+ maxBudgetUsd?: number
142
+ onStart?: () => Promise<void>
143
+ onToolCall?: (payload: { toolName: string; input: any }) => Promise<boolean | void>
144
+ onToolResult?: (payload: { toolName: string; input: any; output: any }) => Promise<void>
145
+ onStepFinish?: (payload: { step: any }) => Promise<void>
146
+ onFinish?: (payload: { result: any }) => Promise<void>
147
+ onError?: (payload: { error: Error }) => Promise<void>
148
+ }
149
+
150
+ // Tool definition — generic over Zod schema
151
+ interface ToolDefinition<TInput> {
152
+ description: string
153
+ inputSchema: ZodType<TInput>
154
+ execute: (args: TInput) => Promise<{ content: { type: string; text: string }[] }>
155
+ }
156
+
157
+ // Tool set
158
+ type ToolSet = Record<string, ToolDefinition<any>>
159
+ ```
@@ -0,0 +1,186 @@
1
+ # React UI
2
+
3
+ `@an-sdk/react` provides a full chat UI for AN agents. Built on [Vercel AI SDK v5](https://sdk.vercel.ai) — uses standard `useChat()` from `@ai-sdk/react`.
4
+
5
+ ## Install
6
+
7
+ ```bash
8
+ npm install @an-sdk/react ai @ai-sdk/react
9
+ ```
10
+
11
+ ## Basic Usage
12
+
13
+ ```tsx
14
+ "use client"
15
+
16
+ import { useChat } from "@ai-sdk/react"
17
+ import { AnAgentChat, createAnChat } from "@an-sdk/react"
18
+ import "@an-sdk/react/styles.css"
19
+ import { useMemo } from "react"
20
+
21
+ export default function Chat() {
22
+ const chat = useMemo(
23
+ () => createAnChat({
24
+ agent: "your-agent-slug",
25
+ getToken: async () => "your_an_sk_token",
26
+ }),
27
+ [],
28
+ )
29
+
30
+ const { messages, sendMessage, status, stop, error } = useChat({ chat })
31
+
32
+ return (
33
+ <AnAgentChat
34
+ messages={messages}
35
+ onSend={(msg) =>
36
+ sendMessage({ parts: [{ type: "text", text: msg.content }] })
37
+ }
38
+ status={status}
39
+ onStop={stop}
40
+ error={error}
41
+ />
42
+ )
43
+ }
44
+ ```
45
+
46
+ ## `createAnChat(options)`
47
+
48
+ Creates an AI SDK `Chat` instance pointed at the AN Relay.
49
+
50
+ ```ts
51
+ createAnChat({
52
+ agent: string // Agent slug from dashboard
53
+ getToken: () => Promise<string> // Returns an_sk_ key or JWT
54
+ apiUrl?: string // Default: "https://relay.an.dev"
55
+ projectId?: string // Session persistence key
56
+ onFinish?: () => void
57
+ onError?: (error: Error) => void
58
+ })
59
+ ```
60
+
61
+ For Next.js apps, use `tokenUrl` instead of `getToken` — see [Next.js Integration](./05-nextjs.md).
62
+
63
+ ## `<AnAgentChat />` Props
64
+
65
+ | Prop | Type | Description |
66
+ |------|------|-------------|
67
+ | `messages` | `UIMessage[]` | From `useChat()` |
68
+ | `onSend` | `(msg) => void` | Send handler |
69
+ | `status` | `ChatStatus` | `"ready" \| "submitted" \| "streaming" \| "error"` |
70
+ | `onStop` | `() => void` | Stop generation |
71
+ | `error` | `Error` | Error to display |
72
+ | `theme` | `AnTheme` | Theme from playground |
73
+ | `colorMode` | `"light" \| "dark" \| "auto"` | Color mode |
74
+ | `classNames` | `Partial<AnClassNames>` | Per-element CSS overrides |
75
+ | `slots` | `Partial<AnSlots>` | Component swapping |
76
+ | `className` | `string` | Root element class |
77
+ | `style` | `CSSProperties` | Root element style |
78
+
79
+ ## Customization
80
+
81
+ Four levels, from simple to full control:
82
+
83
+ ### 1. Theme tokens
84
+
85
+ Apply a theme JSON from the [AN Playground](https://an.dev/an/playground):
86
+
87
+ ```tsx
88
+ <AnAgentChat theme={playgroundTheme} colorMode="dark" />
89
+ ```
90
+
91
+ ### 2. Class overrides
92
+
93
+ ```tsx
94
+ <AnAgentChat
95
+ classNames={{
96
+ root: "rounded-2xl border",
97
+ messageList: "px-8",
98
+ inputBar: "bg-gray-50",
99
+ userMessage: "bg-blue-100",
100
+ }}
101
+ />
102
+ ```
103
+
104
+ ### 3. Slot components
105
+
106
+ Swap sub-components:
107
+
108
+ ```tsx
109
+ <AnAgentChat
110
+ slots={{
111
+ InputBar: MyCustomInput,
112
+ UserMessage: MyUserBubble,
113
+ ToolRenderer: MyToolRenderer,
114
+ }}
115
+ />
116
+ ```
117
+
118
+ ### 4. Individual component imports
119
+
120
+ ```tsx
121
+ import { MessageList, InputBar, ToolRenderer } from "@an-sdk/react"
122
+ ```
123
+
124
+ ## CSS
125
+
126
+ Import once in your app:
127
+
128
+ ```tsx
129
+ import "@an-sdk/react/styles.css"
130
+ ```
131
+
132
+ No Tailwind peer dependency — CSS is pre-compiled. All elements have stable `an-*` class names:
133
+
134
+ ```css
135
+ .an-root { }
136
+ .an-message-list { }
137
+ .an-user-message { }
138
+ .an-assistant-message { }
139
+ .an-input-bar { }
140
+ .an-send-button { }
141
+ .an-tool-bash { }
142
+ .an-tool-edit { }
143
+ ```
144
+
145
+ ## Theme Type
146
+
147
+ ```ts
148
+ interface AnTheme {
149
+ theme: Record<string, string> // Shared: font, spacing, accent
150
+ light: Record<string, string> // Light mode CSS vars
151
+ dark: Record<string, string> // Dark mode CSS vars
152
+ }
153
+ ```
154
+
155
+ ## `applyTheme(element, theme, colorMode?)`
156
+
157
+ Manually inject CSS variables from a theme JSON onto a DOM element.
158
+
159
+ ## Components
160
+
161
+ | Component | Description |
162
+ |-----------|-------------|
163
+ | `MessageList` | Auto-scrolling message container with grouping |
164
+ | `UserMessage` | User message bubble |
165
+ | `AssistantMessage` | Assistant response with parts routing |
166
+ | `StreamingMarkdown` | Markdown renderer with syntax highlighting |
167
+ | `InputBar` | Text input with send/stop buttons |
168
+ | `MessageActions` | Copy button |
169
+ | `ToolRenderer` | Routes tool parts to the correct component |
170
+
171
+ ## Built-in Tool Renderers
172
+
173
+ | Component | Renders |
174
+ |-----------|---------|
175
+ | `BashTool` | Terminal commands with output |
176
+ | `EditTool` | File edits with diff display |
177
+ | `WriteTool` | File creation |
178
+ | `SearchTool` | Web search results |
179
+ | `TodoTool` | Task checklists |
180
+ | `PlanTool` | Step-by-step plans |
181
+ | `TaskTool` | Sub-agent tasks |
182
+ | `McpTool` | MCP protocol calls |
183
+ | `ThinkingTool` | Reasoning/thinking indicator |
184
+ | `GenericTool` | Fallback for unknown tools |
185
+
186
+ See [Custom Tools](./08-custom-tools.md) for building your own tool renderers.
@@ -0,0 +1,117 @@
1
+ # Next.js Integration
2
+
3
+ `@an-sdk/nextjs` provides a server-side token handler so your `an_sk_` API key never reaches the browser. It also re-exports everything from `@an-sdk/react` for convenience.
4
+
5
+ ## Install
6
+
7
+ ```bash
8
+ npm install @an-sdk/nextjs @an-sdk/react ai @ai-sdk/react
9
+ ```
10
+
11
+ ## Setup
12
+
13
+ ### 1. Set your API key
14
+
15
+ ```env
16
+ # .env.local
17
+ AN_API_KEY=an_sk_your_key_here
18
+ ```
19
+
20
+ Get your API key from [an.dev/agents/dashboard/api](https://an.dev/agents/dashboard/api).
21
+
22
+ ### 2. Create the token route
23
+
24
+ ```ts
25
+ // app/api/an/token/route.ts
26
+ import { createAnTokenHandler } from "@an-sdk/nextjs/server"
27
+
28
+ export const POST = createAnTokenHandler({
29
+ apiKey: process.env.AN_API_KEY!,
30
+ })
31
+ ```
32
+
33
+ ### 3. Use in your page
34
+
35
+ ```tsx
36
+ // app/page.tsx
37
+ "use client"
38
+
39
+ import { useChat } from "@ai-sdk/react"
40
+ import { AnAgentChat, createAnChat } from "@an-sdk/nextjs"
41
+ import "@an-sdk/react/styles.css"
42
+ import { useMemo } from "react"
43
+
44
+ export default function Chat() {
45
+ const chat = useMemo(
46
+ () => createAnChat({
47
+ agent: "your-agent-slug",
48
+ tokenUrl: "/api/an/token",
49
+ }),
50
+ [],
51
+ )
52
+
53
+ const { messages, sendMessage, status, stop, error } = useChat({ chat })
54
+
55
+ return (
56
+ <AnAgentChat
57
+ messages={messages}
58
+ onSend={(msg) =>
59
+ sendMessage({ parts: [{ type: "text", text: msg.content }] })
60
+ }
61
+ status={status}
62
+ onStop={stop}
63
+ error={error}
64
+ />
65
+ )
66
+ }
67
+ ```
68
+
69
+ ## How It Works
70
+
71
+ ```
72
+ Browser Your Next.js Server AN Relay
73
+ | | |
74
+ |-- POST /api/an/token --------->| |
75
+ | |-- POST /v1/tokens -------->|
76
+ | | (with an_sk_ key) |
77
+ | |<-- { token, expiresAt } ---|
78
+ |<-- { token, expiresAt } ------| |
79
+ | |
80
+ |-- POST /v1/chat/:agent ------(with short-lived JWT)------->|
81
+ |<-- streaming response -----(SSE)---------------------------|
82
+ ```
83
+
84
+ The client only receives short-lived JWTs. Your API key stays on the server.
85
+
86
+ ## API
87
+
88
+ ### `createAnTokenHandler(options)`
89
+
90
+ Returns a Next.js `POST` route handler.
91
+
92
+ ```ts
93
+ createAnTokenHandler({
94
+ apiKey: string // Your an_sk_ API key
95
+ relayUrl?: string // Default: "https://relay.an.dev"
96
+ expiresIn?: string // Default: "1h"
97
+ })
98
+ ```
99
+
100
+ ### `exchangeToken(options)`
101
+
102
+ Lower-level function for custom token exchange logic.
103
+
104
+ ```ts
105
+ import { exchangeToken } from "@an-sdk/nextjs/server"
106
+
107
+ const { token, expiresAt } = await exchangeToken({
108
+ apiKey: process.env.AN_API_KEY!,
109
+ relayUrl: "https://relay.an.dev",
110
+ expiresIn: "1h",
111
+ })
112
+ ```
113
+
114
+ ## Entry Points
115
+
116
+ - `@an-sdk/nextjs` — Re-exports everything from `@an-sdk/react` (components, types, `createAnChat`)
117
+ - `@an-sdk/nextjs/server` — Server-only: `createAnTokenHandler`, `exchangeToken`
@@ -0,0 +1,125 @@
1
+ # Node SDK
2
+
3
+ `@an-sdk/node` is a server-side client for the AN API. Use it to manage sandboxes, threads, and tokens programmatically.
4
+
5
+ ## Install
6
+
7
+ ```bash
8
+ npm install @an-sdk/node
9
+ ```
10
+
11
+ ## Quick Start
12
+
13
+ ```ts
14
+ import { AnClient } from "@an-sdk/node"
15
+
16
+ const an = new AnClient({
17
+ apiKey: process.env.AN_API_KEY!, // an_sk_...
18
+ })
19
+
20
+ // Create a sandbox for your agent
21
+ const sandbox = await an.sandboxes.create({ agent: "my-agent" })
22
+
23
+ // Create a thread
24
+ const thread = await an.threads.create({
25
+ sandboxId: sandbox.sandboxId,
26
+ name: "Review PR #42",
27
+ })
28
+
29
+ // Generate a short-lived token for browser clients
30
+ const { token, expiresAt } = await an.tokens.create({
31
+ agent: "my-agent",
32
+ expiresIn: "1h",
33
+ })
34
+ ```
35
+
36
+ ## `AnClient`
37
+
38
+ ```ts
39
+ new AnClient({
40
+ apiKey: string // Your an_sk_ API key
41
+ baseUrl?: string // Default: "https://relay.an.dev"
42
+ })
43
+ ```
44
+
45
+ ## Resources
46
+
47
+ ### `client.sandboxes`
48
+
49
+ | Method | Description |
50
+ |--------|-------------|
51
+ | `create({ agent })` | Create a new sandbox for an agent |
52
+ | `get(sandboxId)` | Get sandbox details (status, threads, agent info) |
53
+ | `delete(sandboxId)` | Delete a sandbox |
54
+
55
+ ### `client.threads`
56
+
57
+ | Method | Description |
58
+ |--------|-------------|
59
+ | `list({ sandboxId })` | List all threads in a sandbox |
60
+ | `create({ sandboxId, name? })` | Create a new thread |
61
+ | `get({ sandboxId, threadId })` | Get thread with messages |
62
+ | `delete({ sandboxId, threadId })` | Delete a thread |
63
+
64
+ ### `client.tokens`
65
+
66
+ | Method | Description |
67
+ |--------|-------------|
68
+ | `create({ agent?, userId?, expiresIn? })` | Create a short-lived JWT |
69
+
70
+ Default `expiresIn` is `"1h"`.
71
+
72
+ ## Types
73
+
74
+ ```ts
75
+ interface Sandbox {
76
+ id: string
77
+ sandboxId: string
78
+ status: string
79
+ createdAt: string
80
+ }
81
+
82
+ interface SandboxDetail {
83
+ id: string
84
+ sandboxId: string
85
+ status: string
86
+ error?: string | null
87
+ agent: { slug: string; name: string }
88
+ threads: ThreadSummary[]
89
+ createdAt: string
90
+ updatedAt: string
91
+ }
92
+
93
+ interface ThreadSummary {
94
+ id: string
95
+ name?: string | null
96
+ status: string
97
+ createdAt: string
98
+ }
99
+
100
+ interface Thread {
101
+ id: string
102
+ name?: string | null
103
+ status: string
104
+ messages?: unknown
105
+ createdAt: string
106
+ updatedAt: string
107
+ }
108
+
109
+ interface Token {
110
+ token: string
111
+ expiresAt: string
112
+ }
113
+ ```
114
+
115
+ ## Error Handling
116
+
117
+ All methods throw on non-2xx responses. The error message comes from the API response body when available.
118
+
119
+ ```ts
120
+ try {
121
+ const sandbox = await an.sandboxes.get("nonexistent")
122
+ } catch (err) {
123
+ console.error(err.message) // "Sandbox not found" or similar
124
+ }
125
+ ```