-
Notifications
You must be signed in to change notification settings - Fork 212
Expand file tree
/
Copy pathindex.ts
More file actions
204 lines (192 loc) · 6.4 KB
/
Copy pathindex.ts
File metadata and controls
204 lines (192 loc) · 6.4 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
import { Anthropic } from "@anthropic-ai/sdk"
import OpenAI from "openai"
import { isRetiredProvider, type ProviderSettings, type ModelInfo } from "@roo-code/types"
import { getRouterRemovalMessage } from "../core/config/routerRemoval"
import { ApiStream } from "./transform/stream"
import {
AnthropicHandler,
AwsBedrockHandler,
OpenRouterHandler,
PoeHandler,
VertexHandler,
AnthropicVertexHandler,
OpenAiHandler,
OpenAiCodexHandler,
LmStudioHandler,
GeminiHandler,
OpenAiNativeHandler,
DeepSeekHandler,
MoonshotHandler,
MistralHandler,
VsCodeLmHandler,
RequestyHandler,
UnboundHandler,
UmansHandler,
FakeAIHandler,
XAIHandler,
LiteLLMHandler,
QwenCodeHandler,
SambaNovaHandler,
ZAiHandler,
FireworksHandler,
VercelAiGatewayHandler,
OpencodeGoHandler,
ZooGatewayHandler,
MiniMaxHandler,
MimoHandler,
BasetenHandler,
} from "./providers"
import { NativeOllamaHandler } from "./providers/native-ollama"
export interface SingleCompletionHandler {
completePrompt(prompt: string): Promise<string>
}
export interface ApiHandlerCreateMessageMetadata {
/**
* Task ID used for tracking and provider-specific features:
* - Roo: Sent as X-Roo-Task-ID header
* - Requesty: Sent as trace_id
*/
taskId: string
/**
* Current mode slug for provider-specific tracking:
* - Requesty: Sent in extra metadata
*/
mode?: string
suppressPreviousResponseId?: boolean
/**
* Controls whether the response should be stored for 30 days in OpenAI's Responses API.
* When true (default), responses are stored and can be referenced in future requests
* using the previous_response_id for efficient conversation continuity.
* Set to false to opt out of response storage for privacy or compliance reasons.
* @default true
*/
store?: boolean
/**
* Optional array of tool definitions to pass to the model.
* For OpenAI-compatible providers, these are ChatCompletionTool definitions.
*/
tools?: OpenAI.Chat.ChatCompletionTool[]
/**
* Controls which (if any) tool is called by the model.
* Can be "none", "auto", "required", or a specific tool choice.
*/
tool_choice?: OpenAI.Chat.ChatCompletionCreateParams["tool_choice"]
/**
* Controls whether the model can return multiple tool calls in a single response.
* When true (default), parallel tool calls are enabled (OpenAI's parallel_tool_calls=true).
* When false, only one tool call is returned per response.
*/
parallelToolCalls?: boolean
/**
* Optional array of tool names that the model is allowed to call.
* When provided, all tool definitions are passed to the model (so it can reference
* historical tool calls), but only the specified tools can actually be invoked.
* This is used when switching modes to prevent model errors from missing tool
* definitions while still restricting callable tools to the current mode's permissions.
* Only applies to providers that support function calling restrictions (e.g., Gemini).
*/
allowedFunctionNames?: string[]
/**
* Abort signal for cancelling the HTTP request mid-stream.
* Passed through to AI SDK's streamText() so the underlying HTTP request is aborted
* when the user clicks stop, preventing wasted API tokens/compute on the provider side.
*/
abortSignal?: AbortSignal
}
export interface ApiHandler {
createMessage(
systemPrompt: string,
messages: Anthropic.Messages.MessageParam[],
metadata?: ApiHandlerCreateMessageMetadata,
): ApiStream
getModel(): { id: string; info: ModelInfo }
/**
* Counts tokens for content blocks
* All providers extend BaseProvider which provides a default tiktoken implementation,
* but they can override this to use their native token counting endpoints
*
* @param content The content to count tokens for
* @returns A promise resolving to the token count
*/
countTokens(content: Array<Anthropic.Messages.ContentBlockParam>): Promise<number>
}
export function buildApiHandler(configuration: ProviderSettings): ApiHandler {
const { apiProvider, ...options } = configuration
if (apiProvider === "roo") {
throw new Error(getRouterRemovalMessage())
}
if (apiProvider && isRetiredProvider(apiProvider)) {
throw new Error(
`Sorry, this provider is no longer supported. We saw very few Roo users actually using it and we need to reduce the surface area of our codebase so we can keep shipping fast and serving our community well in this space. It was a really hard decision but it lets us focus on what matters most to you. It sucks, we know.\n\nPlease select a different provider in your API profile settings.`,
)
}
switch (apiProvider) {
case "anthropic":
case "anthropic-custom":
return new AnthropicHandler(options)
case "openrouter":
return new OpenRouterHandler(options)
case "umans":
return new UmansHandler(options)
case "bedrock":
return new AwsBedrockHandler(options)
case "vertex":
return options.apiModelId?.startsWith("claude")
? new AnthropicVertexHandler(options)
: new VertexHandler(options)
case "openai":
return new OpenAiHandler(options)
case "ollama":
return new NativeOllamaHandler(options)
case "lmstudio":
return new LmStudioHandler(options)
case "gemini":
return new GeminiHandler(options)
case "openai-codex":
return new OpenAiCodexHandler(options)
case "openai-native":
return new OpenAiNativeHandler(options)
case "deepseek":
return new DeepSeekHandler(options)
case "qwen-code":
return new QwenCodeHandler(options)
case "moonshot":
return new MoonshotHandler(options)
case "vscode-lm":
return new VsCodeLmHandler(options)
case "mistral":
return new MistralHandler(options)
case "requesty":
return new RequestyHandler(options)
case "unbound":
return new UnboundHandler(options)
case "fake-ai":
return new FakeAIHandler(options)
case "xai":
return new XAIHandler(options)
case "litellm":
return new LiteLLMHandler(options)
case "sambanova":
return new SambaNovaHandler(options)
case "mimo":
return new MimoHandler(options)
case "zai":
return new ZAiHandler(options)
case "fireworks":
return new FireworksHandler(options)
case "vercel-ai-gateway":
return new VercelAiGatewayHandler(options)
case "opencode-go":
return new OpencodeGoHandler(options)
case "zoo-gateway":
return new ZooGatewayHandler(options)
case "minimax":
return new MiniMaxHandler(options)
case "baseten":
return new BasetenHandler(options)
case "poe":
return new PoeHandler(options)
default:
return new AnthropicHandler(options)
}
}