forked from anomalyco/opencode
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathllm.ts
More file actions
279 lines (263 loc) · 8.84 KB
/
llm.ts
File metadata and controls
279 lines (263 loc) · 8.84 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
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
import { Installation } from "@/installation"
import { Provider } from "@/provider/provider"
import { Log } from "@/util/log"
import {
streamText,
wrapLanguageModel,
type ModelMessage,
type StreamTextResult,
type Tool,
type ToolSet,
tool,
jsonSchema,
} from "ai"
import { mergeDeep, pipe } from "remeda"
import { ProviderTransform } from "@/provider/transform"
import { Config } from "@/config/config"
import { Instance } from "@/project/instance"
import type { Agent } from "@/agent/agent"
import type { MessageV2 } from "./message-v2"
import { Plugin } from "@/plugin"
import { SystemPrompt } from "./system"
import { Flag } from "@/flag/flag"
import { PermissionNext } from "@/permission/next"
import { Auth } from "@/auth"
export namespace LLM {
const log = Log.create({ service: "llm" })
export const OUTPUT_TOKEN_MAX = ProviderTransform.OUTPUT_TOKEN_MAX
export type StreamInput = {
user: MessageV2.User
sessionID: string
model: Provider.Model
agent: Agent.Info
system: string[]
abort: AbortSignal
messages: ModelMessage[]
small?: boolean
tools: Record<string, Tool>
retries?: number
toolChoice?: "auto" | "required" | "none"
}
export type StreamOutput = StreamTextResult<ToolSet, unknown>
export async function stream(input: StreamInput) {
const l = log
.clone()
.tag("providerID", input.model.providerID)
.tag("modelID", input.model.id)
.tag("sessionID", input.sessionID)
.tag("small", (input.small ?? false).toString())
.tag("agent", input.agent.name)
.tag("mode", input.agent.mode)
l.info("stream", {
modelID: input.model.id,
providerID: input.model.providerID,
})
const [language, cfg, provider, auth] = await Promise.all([
Provider.getLanguage(input.model),
Config.get(),
Provider.getProvider(input.model.providerID),
Auth.get(input.model.providerID),
])
const isCodex = provider.id === "openai" && auth?.type === "oauth"
const system = []
system.push(
[
// use agent prompt otherwise provider prompt
// For Codex sessions, skip SystemPrompt.provider() since it's sent via options.instructions
...(input.agent.prompt ? [input.agent.prompt] : isCodex ? [] : SystemPrompt.provider(input.model)),
// any custom prompt passed into this call
...input.system,
// any custom prompt from last user message
...(input.user.system ? [input.user.system] : []),
]
.filter((x) => x)
.join("\n"),
)
const header = system[0]
await Plugin.trigger(
"experimental.chat.system.transform",
{ sessionID: input.sessionID, model: input.model },
{ system },
)
// rejoin to maintain 2-part structure for caching if header unchanged
if (system.length > 2 && system[0] === header) {
const rest = system.slice(1)
system.length = 0
system.push(header, rest.join("\n"))
}
const variant =
!input.small && input.model.variants && input.user.variant ? input.model.variants[input.user.variant] : {}
const base = input.small
? ProviderTransform.smallOptions(input.model)
: ProviderTransform.options({
model: input.model,
sessionID: input.sessionID,
providerOptions: provider.options,
})
const options: Record<string, any> = pipe(
base,
mergeDeep(input.model.options),
mergeDeep(input.agent.options),
mergeDeep(variant),
)
if (isCodex) {
options.instructions = SystemPrompt.instructions()
}
const params = await Plugin.trigger(
"chat.params",
{
sessionID: input.sessionID,
agent: input.agent,
model: input.model,
provider,
message: input.user,
},
{
temperature: input.model.capabilities.temperature
? (input.agent.temperature ?? ProviderTransform.temperature(input.model))
: undefined,
topP: input.agent.topP ?? ProviderTransform.topP(input.model),
topK: ProviderTransform.topK(input.model),
options,
},
)
const { headers } = await Plugin.trigger(
"chat.headers",
{
sessionID: input.sessionID,
agent: input.agent,
model: input.model,
provider,
message: input.user,
},
{
headers: {},
},
)
const maxOutputTokens =
isCodex || provider.id.includes("github-copilot") ? undefined : ProviderTransform.maxOutputTokens(input.model)
const tools = await resolveTools(input)
// LiteLLM and some Anthropic proxies require the tools parameter to be present
// when message history contains tool calls, even if no tools are being used.
// Add a dummy tool that is never called to satisfy this validation.
// This is enabled for:
// 1. Providers with "litellm" in their ID or API ID (auto-detected)
// 2. Providers with explicit "litellmProxy: true" option (opt-in for custom gateways)
const isLiteLLMProxy =
provider.options?.["litellmProxy"] === true ||
input.model.providerID.toLowerCase().includes("litellm") ||
input.model.api.id.toLowerCase().includes("litellm")
if (isLiteLLMProxy && Object.keys(tools).length === 0 && hasToolCalls(input.messages)) {
tools["_noop"] = tool({
description:
"Placeholder for LiteLLM/Anthropic proxy compatibility - required when message history contains tool calls but no active tools are needed",
inputSchema: jsonSchema({ type: "object", properties: {} }),
execute: async () => ({ output: "", title: "", metadata: {} }),
})
}
return streamText({
onError(error) {
l.error("stream error", {
error,
})
},
async experimental_repairToolCall(failed) {
const lower = failed.toolCall.toolName.toLowerCase()
if (lower !== failed.toolCall.toolName && tools[lower]) {
l.info("repairing tool call", {
tool: failed.toolCall.toolName,
repaired: lower,
})
return {
...failed.toolCall,
toolName: lower,
}
}
return {
...failed.toolCall,
input: JSON.stringify({
tool: failed.toolCall.toolName,
error: failed.error.message,
}),
toolName: "invalid",
}
},
temperature: params.temperature,
topP: params.topP,
topK: params.topK,
providerOptions: ProviderTransform.providerOptions(input.model, params.options),
activeTools: Object.keys(tools).filter((x) => x !== "invalid"),
tools,
toolChoice: input.toolChoice,
maxOutputTokens,
abortSignal: input.abort,
headers: {
...(input.model.providerID.startsWith("opencode")
? {
"x-opencode-project": Instance.project.id,
"x-opencode-session": input.sessionID,
"x-opencode-request": input.user.id,
"x-opencode-client": Flag.OPENCODE_CLIENT,
}
: input.model.providerID !== "anthropic"
? {
"User-Agent": `opencode/${Installation.VERSION}`,
}
: undefined),
...input.model.headers,
...headers,
},
maxRetries: input.retries ?? 0,
messages: [
...system.map(
(x): ModelMessage => ({
role: "system",
content: x,
}),
),
...input.messages,
],
model: wrapLanguageModel({
model: language,
middleware: [
{
async transformParams(args) {
if (args.type === "stream") {
// @ts-expect-error
args.params.prompt = ProviderTransform.message(args.params.prompt, input.model, options)
}
return args.params
},
},
],
}),
experimental_telemetry: {
isEnabled: cfg.experimental?.openTelemetry,
metadata: {
userId: cfg.username ?? "unknown",
sessionId: input.sessionID,
},
},
})
}
async function resolveTools(input: Pick<StreamInput, "tools" | "agent" | "user">) {
const disabled = PermissionNext.disabled(Object.keys(input.tools), input.agent.permission)
for (const tool of Object.keys(input.tools)) {
if (input.user.tools?.[tool] === false || disabled.has(tool)) {
delete input.tools[tool]
}
}
return input.tools
}
// Check if messages contain any tool-call content
// Used to determine if a dummy tool should be added for LiteLLM proxy compatibility
export function hasToolCalls(messages: ModelMessage[]): boolean {
for (const msg of messages) {
if (!Array.isArray(msg.content)) continue
for (const part of msg.content) {
if (part.type === "tool-call" || part.type === "tool-result") return true
}
}
return false
}
}