Files
stack/apps/api/src/bridge/matrix/matrix-streaming.service.ts
Jason Woltje 8d19ac1f4b
Some checks failed
ci/woodpecker/push/infra Pipeline was successful
ci/woodpecker/push/api Pipeline failed
fix(#377): remediate code review and security findings
- Fix sendThreadMessage room mismatch: use channelId from options instead of hardcoded controlRoomId
- Add .catch() to fire-and-forget handleRoomMessage to prevent silent error swallowing
- Wrap dispatchJob in try-catch for user-visible error reporting in handleFixCommand
- Add MATRIX_BOT_USER_ID validation in connect() to prevent infinite message loops
- Fix streamResponse error masking: wrap finally/catch side-effects in try-catch
- Replace unsafe type assertion with public getClient() in MatrixRoomService
- Add orphaned room warning in provisionRoom on DB failure
- Add provider identity to Herald error logs
- Add channelId to ThreadMessageOptions interface and all callers
- Add missing env var warnings in BridgeModule factory
- Fix JSON injection in setup-bot.sh: use jq for safe JSON construction

Fixes #377

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-15 03:00:53 -06:00

249 lines
7.6 KiB
TypeScript

import { Injectable, Logger } from "@nestjs/common";
import type { MatrixClient } from "matrix-bot-sdk";
import { MatrixService } from "./matrix.service";
/**
* Options for the streamResponse method
*/
export interface StreamResponseOptions {
/** Custom initial message (defaults to "Thinking...") */
initialMessage?: string;
/** Thread root event ID for threaded responses */
threadId?: string;
/** Whether to show token usage in the final message */
showTokenUsage?: boolean;
/** Token usage stats to display in the final message */
tokenUsage?: { prompt: number; completion: number; total: number };
}
/**
* Matrix message content for m.room.message events
*/
interface MatrixMessageContent {
msgtype: string;
body: string;
"m.new_content"?: {
msgtype: string;
body: string;
};
"m.relates_to"?: {
rel_type: string;
event_id: string;
is_falling_back?: boolean;
"m.in_reply_to"?: {
event_id: string;
};
};
}
/** Minimum interval between message edits (milliseconds) */
const EDIT_INTERVAL_MS = 500;
/** Typing indicator timeout (milliseconds) */
const TYPING_TIMEOUT_MS = 30000;
/**
* Matrix Streaming Service
*
* Provides streaming AI response capabilities for Matrix rooms using
* incremental message edits. Tokens from an LLM are buffered and the
* response message is edited at rate-limited intervals, providing a
* smooth streaming experience without excessive API calls.
*
* Key features:
* - Rate-limited edits (max every 500ms)
* - Typing indicator management during generation
* - Graceful error handling with user-visible error notices
* - Thread support for contextual responses
* - LLM-agnostic design via AsyncIterable<string> token stream
*/
@Injectable()
export class MatrixStreamingService {
private readonly logger = new Logger(MatrixStreamingService.name);
constructor(private readonly matrixService: MatrixService) {}
/**
* Edit an existing Matrix message using the m.replace relation.
*
* Sends a new event that replaces the content of an existing message.
* Includes fallback content for clients that don't support edits.
*
* @param roomId - The Matrix room ID
* @param eventId - The original event ID to replace
* @param newContent - The updated message text
*/
async editMessage(roomId: string, eventId: string, newContent: string): Promise<void> {
const client = this.getClientOrThrow();
const editContent: MatrixMessageContent = {
"m.new_content": {
msgtype: "m.text",
body: newContent,
},
"m.relates_to": {
rel_type: "m.replace",
event_id: eventId,
},
// Fallback for clients that don't support edits
msgtype: "m.text",
body: `* ${newContent}`,
};
await client.sendEvent(roomId, "m.room.message", editContent);
}
/**
* Set the typing indicator for the bot in a room.
*
* @param roomId - The Matrix room ID
* @param typing - Whether the bot is typing
*/
async setTypingIndicator(roomId: string, typing: boolean): Promise<void> {
const client = this.getClientOrThrow();
await client.setTyping(roomId, typing, typing ? TYPING_TIMEOUT_MS : undefined);
}
/**
* Send an initial message for streaming, optionally in a thread.
*
* Returns the event ID of the sent message, which can be used for
* subsequent edits via editMessage.
*
* @param roomId - The Matrix room ID
* @param content - The initial message content
* @param threadId - Optional thread root event ID
* @returns The event ID of the sent message
*/
async sendStreamingMessage(roomId: string, content: string, threadId?: string): Promise<string> {
const client = this.getClientOrThrow();
const messageContent: MatrixMessageContent = {
msgtype: "m.text",
body: content,
};
if (threadId) {
messageContent["m.relates_to"] = {
rel_type: "m.thread",
event_id: threadId,
is_falling_back: true,
"m.in_reply_to": {
event_id: threadId,
},
};
}
const eventId: string = await client.sendMessage(roomId, messageContent);
return eventId;
}
/**
* Stream an AI response to a Matrix room using incremental message edits.
*
* This is the main streaming method. It:
* 1. Sends an initial "Thinking..." message
* 2. Starts the typing indicator
* 3. Buffers incoming tokens from the async iterable
* 4. Edits the message every 500ms with accumulated text
* 5. On completion: sends a final clean edit, clears typing
* 6. On error: edits message with error notice, clears typing
*
* @param roomId - The Matrix room ID
* @param tokenStream - AsyncIterable that yields string tokens
* @param options - Optional configuration for the stream
*/
async streamResponse(
roomId: string,
tokenStream: AsyncIterable<string>,
options?: StreamResponseOptions
): Promise<void> {
// Validate connection before starting
this.getClientOrThrow();
const initialMessage = options?.initialMessage ?? "Thinking...";
const threadId = options?.threadId;
// Step 1: Send initial message
const eventId = await this.sendStreamingMessage(roomId, initialMessage, threadId);
// Step 2: Start typing indicator
await this.setTypingIndicator(roomId, true);
// Step 3: Buffer and stream tokens
let accumulatedText = "";
let lastEditTime = 0;
let hasError = false;
try {
for await (const token of tokenStream) {
accumulatedText += token;
const now = Date.now();
const elapsed = now - lastEditTime;
if (elapsed >= EDIT_INTERVAL_MS && accumulatedText.length > 0) {
await this.editMessage(roomId, eventId, accumulatedText);
lastEditTime = now;
}
}
} catch (error: unknown) {
hasError = true;
const errorMessage = error instanceof Error ? error.message : "Unknown error occurred";
this.logger.error(`Stream error in room ${roomId}: ${errorMessage}`);
// Edit message to show error
try {
const errorContent = accumulatedText
? `${accumulatedText}\n\n[Streaming error: ${errorMessage}]`
: `[Streaming error: ${errorMessage}]`;
await this.editMessage(roomId, eventId, errorContent);
} catch (editError: unknown) {
this.logger.warn(
`Failed to edit error message in ${roomId}: ${editError instanceof Error ? editError.message : "unknown"}`
);
}
} finally {
// Step 4: Clear typing indicator
try {
await this.setTypingIndicator(roomId, false);
} catch (typingError: unknown) {
this.logger.warn(
`Failed to clear typing indicator in ${roomId}: ${typingError instanceof Error ? typingError.message : "unknown"}`
);
}
}
// Step 5: Final edit with clean output (if no error)
if (!hasError) {
let finalContent = accumulatedText || "(No response generated)";
if (options?.showTokenUsage && options.tokenUsage) {
const { prompt, completion, total } = options.tokenUsage;
finalContent += `\n\n---\nTokens: ${String(total)} (prompt: ${String(prompt)}, completion: ${String(completion)})`;
}
await this.editMessage(roomId, eventId, finalContent);
}
}
/**
* Get the Matrix client from the parent MatrixService, or throw if not connected.
*/
private getClientOrThrow(): MatrixClient {
if (!this.matrixService.isConnected()) {
throw new Error("Matrix client is not connected");
}
const client = this.matrixService.getClient();
if (!client) {
throw new Error("Matrix client is not connected");
}
return client;
}
}