File size: 5,746 Bytes
80d7f0c | 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 | import type {
Context as AiContext,
Api,
AssistantMessage,
AssistantMessageEvent,
AssistantMessageEventStream,
Message,
Model,
SimpleStreamOptions,
Tool,
} from "@earendil-works/pi-ai";
import type { AgentMessage, ThinkingLevel } from "../../types.ts";
import { type Context, getTelemetryContext } from "../context.ts";
import type { SettledAssistantMessage } from "../session/types.ts";
import type { AgentHarnessStreamOptions } from "../types.ts";
import { AbortRequested } from "./effect-gate.ts";
/** HTTP response metadata captured before the provider response body is consumed. */
export interface AssistantResponseMetadata {
status?: number;
headers?: Record<string, string>;
}
/** Process-local lifecycle observer for one assistant stream. */
export interface AssistantStreamObserver {
start(
message: AssistantMessage,
event: Extract<AssistantMessageEvent, { type: "start" }>,
context: Context,
): void | Promise<void>;
update(message: AssistantMessage, event: AssistantMessageEvent, context: Context): void | Promise<void>;
end(message: SettledAssistantMessage, context: Context): void | Promise<void>;
}
/** Executable inputs for one already-approved assistant provider request. */
export interface HarnessAssistantStreamConfig {
model: Model<Api>;
systemPrompt: string;
tools?: Tool[];
thinkingLevel: ThinkingLevel;
streamOptions: AgentHarnessStreamOptions;
transformContext?: (
requestContext: { messages: AgentMessage[]; systemPrompt: string },
context: Context,
) => Promise<{ messages: AgentMessage[]; systemPrompt: string }>;
toProviderMessages: (messages: AgentMessage[], context: Context) => Message[] | Promise<Message[]>;
beforePayload?: (
payload: unknown,
model: Model<Api>,
context: Context,
) => unknown | undefined | Promise<unknown | undefined>;
afterResponse?: (
message: SettledAssistantMessage,
metadata: AssistantResponseMetadata,
context: Context,
) => Promise<SettledAssistantMessage>;
request(
aiContext: AiContext,
options: SimpleStreamOptions,
context: Context,
): AssistantMessageEventStream | Promise<AssistantMessageEventStream>;
observer: AssistantStreamObserver;
}
function createRequestOptions(
config: HarnessAssistantStreamConfig,
captureMetadata: (metadata: AssistantResponseMetadata) => void,
context: Context,
): SimpleStreamOptions {
const options = config.streamOptions;
return {
transport: options.transport,
timeoutMs: options.timeoutMs,
maxRetries: options.maxRetries,
maxRetryDelayMs: options.maxRetryDelayMs,
headers: options.headers,
metadata: options.metadata,
cacheRetention: options.cacheRetention,
deferred: options.deferred,
...(config.thinkingLevel === "off" ? {} : { reasoning: config.thinkingLevel }),
signal: context.abortSignal,
telemetryContext: getTelemetryContext(context),
onPayload:
config.beforePayload === undefined
? undefined
: (payload, model) => config.beforePayload?.(payload, model, context),
onResponse: (response) => {
captureMetadata({ status: response.status, headers: response.headers });
},
};
}
function isUpdateEvent(
event: AssistantMessageEvent,
): event is Exclude<AssistantMessageEvent, { type: "start" | "done" | "error" }> {
return event.type !== "start" && event.type !== "done" && event.type !== "error";
}
export async function consumeAssistantStream(
stream: AssistantMessageEventStream,
observer: AssistantStreamObserver,
afterResponse:
| ((message: SettledAssistantMessage, context: Context) => Promise<SettledAssistantMessage>)
| undefined,
context: Context,
): Promise<SettledAssistantMessage> {
let started = false;
for await (const event of stream) {
if (event.type === "start") {
if (started) throw new Error("Assistant message stream emitted more than one start event");
started = true;
await observer.start({ ...event.partial }, event, context);
} else if (isUpdateEvent(event)) {
if (!started) throw new Error(`Assistant message stream emitted ${event.type} before start`);
await observer.update({ ...event.partial }, event, context);
} else if (event.type === "done" && !started) {
throw new Error("Assistant message stream emitted done before start");
}
}
const settled = (await stream.result()) as SettledAssistantMessage;
let finalMessage = settled;
if (afterResponse !== undefined) {
try {
finalMessage = await afterResponse(settled, context);
} catch (error) {
if (!(error instanceof AbortRequested)) throw error;
await error.cancellation;
}
}
await observer.end(finalMessage, context);
return finalMessage;
}
/** Stream one assistant response without mutating the caller's message list. */
export async function streamHarnessAssistant(
messages: AgentMessage[],
config: HarnessAssistantStreamConfig,
context: Context,
): Promise<SettledAssistantMessage> {
let requestContext = { messages: messages.slice(), systemPrompt: config.systemPrompt };
if (config.transformContext) {
requestContext = await config.transformContext(requestContext, context);
}
const providerMessages = await config.toProviderMessages(requestContext.messages, context);
const aiContext: AiContext = {
systemPrompt: requestContext.systemPrompt,
messages: providerMessages,
tools: config.tools,
};
let metadata: AssistantResponseMetadata = {};
const stream = await config.request(
aiContext,
createRequestOptions(
config,
(nextMetadata) => {
metadata = nextMetadata;
},
context,
),
context,
);
const afterResponse = config.afterResponse;
return consumeAssistantStream(
stream,
config.observer,
afterResponse === undefined
? undefined
: (message, afterContext) => afterResponse(message, metadata, afterContext),
context,
);
}
|