feat(FN-4180): compact room transcript context to reduce token usage

Adds room transcript context compaction (FN-4180) to the dashboard chat layer — widens the fetch window then compacts older messages into a summary window to reduce token overhead while keeping full context recent. Includes a new 127-line test suite covering the compaction logic, a small fix to reus

Fusion-Task-Id: FN-4180

Fusion-Task-Lineage: 9e7e3a32-f2da-4c86-871b-37891f747561
This commit is contained in:
Fusion
2026-05-12 13:22:28 -07:00
committed by gsxdsm
parent 062b5a9039
commit 8d4835998c
8 changed files with 292 additions and 37 deletions

View File

@@ -0,0 +1,127 @@
import { describe, expect, it } from "vitest";
import { buildCompactedRoomTranscript } from "../chat.js";
function makeMessage(index: number, overrides: Partial<{
id: string;
role: "user" | "assistant" | "system";
content: string;
createdAt: string;
senderAgentId: string | null;
}> = {}) {
return {
id: overrides.id ?? `msg-${index}`,
role: overrides.role ?? (index % 2 === 0 ? "user" : "assistant"),
content: overrides.content ?? `message-${index}`,
createdAt: overrides.createdAt ?? `2026-01-01T00:00:${String(index).padStart(2, "0")}.000Z`,
senderAgentId: "senderAgentId" in overrides ? (overrides.senderAgentId ?? null) : (index % 2 === 0 ? null : "agent-a"),
};
}
describe("buildCompactedRoomTranscript", () => {
it("returns all messages verbatim when the transcript fits inside the recent window", () => {
const messages = Array.from({ length: 6 }, (_, index) => makeMessage(index));
const transcript = buildCompactedRoomTranscript(messages, "msg-4");
expect(transcript).not.toContain("## Earlier room context (compacted)");
expect(transcript).toContain("message-0");
expect(transcript).toContain("message-5");
expect(transcript.match(/\[LATEST USER MESSAGE — ANSWER THIS\]/g)).toHaveLength(1);
});
it("prepends a compacted summary and keeps the last 12 messages verbatim", () => {
const messages = Array.from({ length: 30 }, (_, index) => {
const olderUserLengths = [40, 80, 120, 160, 200, 220, 60, 70, 90];
const content = index < 18 && index % 2 === 0
? `older-user-${index}-` + "u".repeat(olderUserLengths[index / 2] ?? 20)
: `message-${index}`;
return makeMessage(index, { content });
});
const latestUserMessageId = "msg-28";
const transcript = buildCompactedRoomTranscript(messages, latestUserMessageId);
expect(transcript).toContain("## Earlier room context (compacted)");
expect(transcript).toContain("- Span: 18 messages from 2026-01-01T00:00:00.000Z to 2026-01-01T00:00:17.000Z");
expect(transcript).toContain("- Participants: User, Agent agent-a");
const [summaryBlock] = transcript.split("\n\n");
const highlightLines = summaryBlock.split("\n").filter((line) => line.startsWith(" - "));
expect(highlightLines).toHaveLength(5);
const highlightTimestamps = highlightLines.map((line) => line.match(/\[(.*?)\]/)?.[1] ?? "");
expect(highlightTimestamps).toEqual([...highlightTimestamps].sort());
for (let index = 18; index < 30; index += 1) {
expect(transcript).toContain(`- [2026-01-01T00:00:${String(index).padStart(2, "0")}.000Z]`);
}
expect(transcript).toContain("(user) User: message-28 [LATEST USER MESSAGE — ANSWER THIS]");
});
it("preserves the latest marker exactly once even when the transcript must shrink", () => {
const messages = Array.from({ length: 30 }, (_, index) => makeMessage(index, {
role: index === 29 ? "user" : (index % 3 === 0 ? "assistant" : "user"),
senderAgentId: index % 3 === 0 ? "agent-a" : null,
content: `message-${index}-` + "x".repeat(1500),
}));
const transcript = buildCompactedRoomTranscript(messages, "msg-29");
expect(transcript.length).toBeLessThanOrEqual(8000);
expect(transcript.match(/\[LATEST USER MESSAGE — ANSWER THIS\]/g)).toHaveLength(1);
expect(transcript).toContain("message-29-");
});
it("drops summary highlights from the bottom when the summary exceeds its cap", () => {
const olderMessages = Array.from({ length: 18 }, (_, index) => makeMessage(index, {
role: "user",
content: `older-${index}-` + "z".repeat(500),
}));
const recentMessages = Array.from({ length: 12 }, (_, index) => makeMessage(index + 18, {
role: index === 11 ? "user" : "assistant",
senderAgentId: index === 11 ? null : `agent-${index}`,
content: `recent-${index}`,
}));
const transcript = buildCompactedRoomTranscript([...olderMessages, ...recentMessages], "msg-29");
const [summaryBlock] = transcript.split("\n\n");
const highlightLines = summaryBlock.split("\n").filter((line) => line.startsWith(" - "));
expect(summaryBlock).toContain("## Earlier room context (compacted)");
expect(summaryBlock).toContain("- Span: 18 messages");
expect(summaryBlock).toContain("- Participants: User");
expect(summaryBlock).toContain("- Highlights:");
expect(summaryBlock.length).toBeLessThanOrEqual(1500);
expect(highlightLines.length).toBeLessThan(5);
});
it("keeps the total transcript under the overall cap", () => {
const messages = Array.from({ length: 80 }, (_, index) => makeMessage(index, {
role: index === 79 ? "user" : (index % 4 === 0 ? "system" : index % 2 === 0 ? "assistant" : "user"),
senderAgentId: index % 2 === 0 && index % 4 !== 0 ? `agent-${index}` : null,
content: `message-${index}-` + "q".repeat(4000),
}));
const transcript = buildCompactedRoomTranscript(messages, "msg-79");
expect(transcript.length).toBeLessThanOrEqual(8000);
expect(transcript).toContain("message-79-");
});
it("computes unique participant labels from older messages", () => {
const older = [
makeMessage(0, { role: "user", senderAgentId: null, content: "user older" }),
makeMessage(1, { role: "assistant", senderAgentId: "agent-a", content: "agent a older" }),
makeMessage(2, { role: "system", senderAgentId: null, content: "system older" }),
makeMessage(3, { role: "assistant", senderAgentId: null, content: "assistant older" }),
makeMessage(4, { role: "assistant", senderAgentId: "agent-b", content: "agent b older" }),
];
const recent = Array.from({ length: 12 }, (_, index) => makeMessage(index + 5, {
role: index === 11 ? "user" : "assistant",
senderAgentId: index === 11 ? null : "agent-c",
content: `recent-${index}`,
}));
const transcript = buildCompactedRoomTranscript([...older, ...recent], "msg-16");
expect(transcript).toContain("- Participants: User, Agent agent-a, System, Assistant, Agent agent-b");
});
});

View File

@@ -205,7 +205,7 @@ describe("Chat orchestration — rooms (FN-3805..FN-3811 contract)", () => {
expect(mockChatStore.getRoomMessages).toHaveBeenCalledWith("room-1", { limit: expect.any(Number) });
});
it("trims older room context entries from the prompt window", async () => {
it("compacts older room context entries in the responder prompt", async () => {
mockChatStore.listRoomMembers.mockReturnValue([
{ roomId: "room-1", agentId: "agent-a", role: "member", addedAt: "2026-01-01" },
]);
@@ -213,6 +213,11 @@ describe("Chat orchestration — rooms (FN-3805..FN-3811 contract)", () => {
mockAgentStore.getAgent.mockResolvedValue({ id: "agent-a", name: "Alpha", role: "executor" });
const promptSpy = vi.fn().mockResolvedValue(undefined);
mockChatStore.addRoomMessage.mockImplementationOnce((_roomId: string, input: any) => ({
id: "history-latest",
roomId: "room-1",
...input,
}));
__setCreateResolvedAgentSession(async () => ({
session: {
prompt: promptSpy,
@@ -232,7 +237,7 @@ describe("Chat orchestration — rooms (FN-3805..FN-3811 contract)", () => {
}));
history[history.length - 1] = {
...history[history.length - 1],
id: "msg-1",
id: "history-latest",
role: "user",
senderAgentId: null,
content: "hello @Alpha",
@@ -246,8 +251,13 @@ describe("Chat orchestration — rooms (FN-3805..FN-3811 contract)", () => {
await manager.sendRoomMessage("room-1", "hello @Alpha");
const prompt = promptSpy.mock.calls[0]?.[0] as string;
expect(prompt).toContain("## Earlier room context (compacted)");
expect(prompt).toContain("- Span: 18 messages from 2026-01-01T00:00:00.000Z to 2026-01-01T00:00:17.000Z");
expect(prompt).toContain("history-item-28");
expect(prompt).not.toContain("history-item-0");
expect(prompt).toContain(" - [2026-01-01T00:00:00.000Z] User: history-item-0");
expect(prompt).not.toContain("- [2026-01-01T00:00:00.000Z] (user) User: history-item-0");
expect(prompt).toContain("- [2026-01-01T00:00:29.000Z] (user) User: hello @Alpha [LATEST USER MESSAGE — ANSWER THIS]");
expect(prompt.match(/\[LATEST USER MESSAGE — ANSWER THIS\]/g)).toHaveLength(1);
});
it("throws surfaced error when room has members but no resolvable responders", async () => {

View File

@@ -19,6 +19,7 @@ import type {
ChatAttachment,
ChatInFlightGenerationState,
ChatStore,
ChatRoomMessage,
ChatSession,
ChatSessionCreateInput,
MessageStore,
@@ -137,11 +138,126 @@ const MAX_MESSAGES_PER_IP_PER_MINUTE = 30;
/** Maximum file size for # mentions (50KB). Files larger than this are skipped. */
const MAX_REFERENCED_FILE_SIZE = 50 * 1024;
const ROOM_AMBIENT_MAX_RESPONDERS = 5;
const ROOM_THREAD_CONTEXT_MAX_MESSAGES = 16;
const ROOM_THREAD_RECENT_VERBATIM_MESSAGES = 12;
const ROOM_THREAD_COMPACTION_FETCH_LIMIT = 80;
const ROOM_THREAD_CONTEXT_MAX_CHARS = 8_000;
const ROOM_THREAD_MESSAGE_CONTENT_MAX_CHARS = 1_200;
const ROOM_THREAD_SUMMARY_MAX_CHARS = 1_500;
const IN_FLIGHT_PERSIST_DEBOUNCE_MS = 200;
type RoomTranscriptMessage = Pick<ChatRoomMessage, "id" | "role" | "content" | "createdAt" | "senderAgentId">;
function getRoomSenderLabel(message: Pick<RoomTranscriptMessage, "role" | "senderAgentId">): string {
return message.role === "user"
? "User"
: message.role === "system"
? "System"
: (message.senderAgentId ? `Agent ${message.senderAgentId}` : "Assistant");
}
function truncateWithEllipsis(content: string, maxChars: number): string {
return content.length > maxChars
? `${content.slice(0, maxChars - 1)}`
: content;
}
function formatRoomThreadLine(message: RoomTranscriptMessage, latestUserMessageId: string): string {
const marker = message.id === latestUserMessageId ? " [LATEST USER MESSAGE — ANSWER THIS]" : "";
return `- [${message.createdAt}] (${message.role}) ${getRoomSenderLabel(message)}: ${truncateWithEllipsis(message.content, ROOM_THREAD_MESSAGE_CONTENT_MAX_CHARS)}${marker}`;
}
function formatRoomThreadContext(messages: RoomTranscriptMessage[], latestUserMessageId: string): string {
return messages.map((message) => formatRoomThreadLine(message, latestUserMessageId)).join("\n");
}
function buildRoomSummaryBlock(olderMessages: RoomTranscriptMessage[]): string {
if (olderMessages.length === 0) {
return "";
}
const participants = Array.from(new Set(olderMessages.map((message) => getRoomSenderLabel(message))));
const rankedHighlights = olderMessages
.map((message, index) => ({
message,
index,
score: (message.role === "user" ? 2 : message.role === "assistant" ? 1 : 0) * 1000 + message.content.length,
}))
.sort((left, right) => right.score - left.score || left.index - right.index)
.slice(0, 5)
.sort((left, right) => left.index - right.index)
.map(({ message }) => ` - [${message.createdAt}] ${getRoomSenderLabel(message)}: ${truncateWithEllipsis(message.content, 240)}`);
const summaryLines = [
"## Earlier room context (compacted)",
`- Span: ${olderMessages.length} messages from ${olderMessages[0]?.createdAt ?? ""} to ${olderMessages.at(-1)?.createdAt ?? ""}`,
`- Participants: ${participants.join(", ")}`,
"- Highlights:",
];
const baseSummary = summaryLines.join("\n");
if (rankedHighlights.length === 0) {
return baseSummary;
}
const highlights = [...rankedHighlights];
while (`${baseSummary}\n${highlights.join("\n")}`.length > ROOM_THREAD_SUMMARY_MAX_CHARS && highlights.length > 0) {
highlights.pop();
}
return highlights.length > 0
? `${baseSummary}\n${highlights.join("\n")}`
: baseSummary;
}
export function buildCompactedRoomTranscript(
messages: RoomTranscriptMessage[],
latestUserMessageId: string,
): string {
if (messages.length === 0) {
return "";
}
const messageIndexes = new Map(messages.map((message, index) => [message.id, index]));
const latestUserMessage = messages.find((message) => message.id === latestUserMessageId);
const splitIndex = Math.max(0, messages.length - ROOM_THREAD_RECENT_VERBATIM_MESSAGES);
let olderMessages = messages.slice(0, splitIndex);
let recentMessages = messages.slice(splitIndex);
if (latestUserMessage && !recentMessages.some((message) => message.id === latestUserMessageId)) {
olderMessages = olderMessages.filter((message) => message.id !== latestUserMessageId);
recentMessages = [...recentMessages, latestUserMessage]
.sort((left, right) => (messageIndexes.get(left.id) ?? 0) - (messageIndexes.get(right.id) ?? 0));
}
const summaryLines = buildRoomSummaryBlock(olderMessages).split("\n").filter((line) => line.length > 0);
const renderTranscript = () => {
const summary = summaryLines.length > 0 ? summaryLines.join("\n") : "";
const recent = formatRoomThreadContext(recentMessages, latestUserMessageId);
if (summary && recent) {
return `${summary}\n\n${recent}`;
}
return summary || recent;
};
let transcript = renderTranscript();
while (transcript.length > ROOM_THREAD_CONTEXT_MAX_CHARS && summaryLines.at(-1)?.startsWith(" - ")) {
summaryLines.pop();
transcript = renderTranscript();
}
while (transcript.length > ROOM_THREAD_CONTEXT_MAX_CHARS && recentMessages.length > 1) {
const removableIndex = recentMessages.findIndex((message) => message.id !== latestUserMessageId);
if (removableIndex === -1) {
break;
}
recentMessages.splice(removableIndex, 1);
transcript = renderTranscript();
}
return transcript;
}
function formatAttachmentSize(size: number): string {
if (size < 1024) return `${size}B`;
if (size < 1024 * 1024) return `${(size / 1024).toFixed(1)}KB`;
@@ -1072,12 +1188,12 @@ export class ChatManager {
}
systemPrompt = `${systemPrompt}\n\n${CHAT_AGENT_MESSAGE_ROUTING_GUIDANCE}`;
const roomMessages = this.chatStore.getRoomMessages(input.roomId, { limit: ROOM_THREAD_CONTEXT_MAX_MESSAGES });
const roomMessages = this.chatStore.getRoomMessages(input.roomId, { limit: ROOM_THREAD_COMPACTION_FETCH_LIMIT });
const roomPrompt = [
`You are replying as ${input.responder.name} in room #${input.roomName}.`,
"Reply to the latest user room message in the context of this shared room thread.",
"Room transcript (oldest to newest, bounded):",
this.formatRoomThreadContext(roomMessages, input.latestUserMessageId),
this.compactRoomThreadContext(roomMessages, input.latestUserMessageId),
"Latest user message to answer:",
input.content,
].join("\n\n");
@@ -1147,35 +1263,15 @@ export class ChatManager {
}
}
private formatRoomThreadContext(
messages: Array<{ id: string; role: "user" | "assistant" | "system"; content: string; createdAt: string; senderAgentId?: string | null }>,
/**
* Preserve the newest room turns verbatim while compacting older history into
* a deterministic summary block so long-running rooms keep continuity.
*/
private compactRoomThreadContext(
messages: RoomTranscriptMessage[],
latestUserMessageId: string,
): string {
const trimmedFromTail: string[] = [];
let totalChars = 0;
for (let index = messages.length - 1; index >= 0; index -= 1) {
const message = messages[index];
const senderLabel = message.role === "user"
? "User"
: message.role === "system"
? "System"
: (message.senderAgentId ? `Agent ${message.senderAgentId}` : "Assistant");
const content = message.content.length > ROOM_THREAD_MESSAGE_CONTENT_MAX_CHARS
? `${message.content.slice(0, ROOM_THREAD_MESSAGE_CONTENT_MAX_CHARS - 1)}`
: message.content;
const marker = message.id === latestUserMessageId ? " [LATEST USER MESSAGE — ANSWER THIS]" : "";
const line = `- [${message.createdAt}] (${message.role}) ${senderLabel}: ${content}${marker}`;
if (trimmedFromTail.length > 0 && totalChars + line.length > ROOM_THREAD_CONTEXT_MAX_CHARS) {
break;
}
trimmedFromTail.push(line);
totalChars += line.length;
}
return trimmedFromTail.reverse().join("\n");
return buildCompactedRoomTranscript(messages, latestUserMessageId);
}
/**

View File

@@ -173,6 +173,10 @@ function cleanupTempDir(dir?: string): void {
sleepSync(attempt * 25);
}
if (existsSync(dir)) {
throw new Error(`failed to clean temp dir: ${dir}`);
}
}
afterAll(() => {

View File

@@ -87,19 +87,29 @@ const STUB_SETTINGS = {
const createdDirs = new Set<string>();
function sleepSync(ms: number): void {
Atomics.wait(new Int32Array(new SharedArrayBuffer(4)), 0, 0, ms);
}
function cleanupTempDir(dir?: string): void {
if (!dir) return;
createdDirs.delete(dir);
for (let attempt = 1; attempt <= 3; attempt += 1) {
for (let attempt = 1; attempt <= 5; attempt += 1) {
try {
if (!existsSync(dir)) return;
rmSync(dir, { recursive: true, force: true });
return;
if (!existsSync(dir)) return;
} catch (error) {
if (attempt === 3) {
if (attempt === 5) {
throw error;
}
}
sleepSync(attempt * 25);
}
if (existsSync(dir)) {
throw new Error(`failed to clean temp dir: ${dir}`);
}
}