feat: copilot controller (#6272)

fix CLOUD-27
This commit is contained in:
darkskygit
2024-04-10 11:58:40 +00:00
parent e6a576551a
commit 7c38a54f81
18 changed files with 729 additions and 179 deletions
@@ -0,0 +1,151 @@
import {
BadRequestException,
Controller,
Get,
InternalServerErrorException,
Param,
Query,
Req,
Sse,
} from '@nestjs/common';
import {
concatMap,
connect,
EMPTY,
from,
map,
merge,
Observable,
switchMap,
toArray,
} from 'rxjs';
import { Public } from '../../core/auth';
import { CurrentUser } from '../../core/auth/current-user';
import { CopilotProviderService } from './providers';
import { ChatSessionService } from './session';
import { CopilotCapability } from './types';
export interface ChatEvent {
data: string;
id?: string;
}
@Controller('/api/copilot')
export class CopilotController {
constructor(
private readonly chatSession: ChatSessionService,
private readonly provider: CopilotProviderService
) {}
@Public()
@Get('/chat/:sessionId')
async chat(
@CurrentUser() user: CurrentUser,
@Req() req: Request,
@Param('sessionId') sessionId: string,
@Query('message') content: string,
@Query() params: Record<string, string | string[]>
): Promise<string> {
const provider = this.provider.getProviderByCapability(
CopilotCapability.TextToText
);
if (!provider) {
throw new InternalServerErrorException('No provider available');
}
const session = await this.chatSession.get(sessionId);
if (!session) {
throw new BadRequestException('Session not found');
}
if (!content || !content.trim()) {
throw new BadRequestException('Message is empty');
}
session.push({
role: 'user',
content: decodeURIComponent(content),
createdAt: new Date(),
});
try {
delete params.message;
const content = await provider.generateText(
session.finish(params),
session.model,
{
signal: req.signal,
user: user.id,
}
);
session.push({
role: 'assistant',
content,
createdAt: new Date(),
});
await session.save();
return content;
} catch (e: any) {
throw new InternalServerErrorException(
e.message || "Couldn't generate text"
);
}
}
@Public()
@Sse('/chat/:sessionId/stream')
async chatStream(
@CurrentUser() user: CurrentUser,
@Req() req: Request,
@Param('sessionId') sessionId: string,
@Query('message') content: string,
@Query() params: Record<string, string>
): Promise<Observable<ChatEvent>> {
const provider = this.provider.getProviderByCapability(
CopilotCapability.TextToText
);
if (!provider) {
throw new InternalServerErrorException('No provider available');
}
const session = await this.chatSession.get(sessionId);
if (!session) {
throw new BadRequestException('Session not found');
}
if (!content || !content.trim()) {
throw new BadRequestException('Message is empty');
}
session.push({
role: 'user',
content: decodeURIComponent(content),
createdAt: new Date(),
});
delete params.message;
return from(
provider.generateTextStream(session.finish(params), session.model, {
signal: req.signal,
user: user.id,
})
).pipe(
connect(shared$ =>
merge(
// actual chat event stream
shared$.pipe(map(data => ({ id: sessionId, data }))),
// save the generated text to the session
shared$.pipe(
toArray(),
concatMap(values => {
session.push({
role: 'assistant',
content: values.join(''),
createdAt: new Date(),
});
return from(session.save());
}),
switchMap(() => EMPTY)
)
)
)
);
}
}
@@ -1,6 +1,8 @@
import { ServerFeature } from '../../core/config';
import { QuotaService } from '../../core/quota';
import { PermissionService } from '../../core/workspaces/permission';
import { Plugin } from '../registry';
import { CopilotController } from './controller';
import { PromptService } from './prompt';
import {
assertProvidersConfigs,
@@ -8,6 +10,7 @@ import {
OpenAIProvider,
registerCopilotProvider,
} from './providers';
import { CopilotResolver, UserCopilotResolver } from './resolver';
import { ChatSessionService } from './session';
registerCopilotProvider(OpenAIProvider);
@@ -16,10 +19,14 @@ registerCopilotProvider(OpenAIProvider);
name: 'copilot',
providers: [
PermissionService,
QuotaService,
ChatSessionService,
CopilotResolver,
UserCopilotResolver,
PromptService,
CopilotProviderService,
],
controllers: [CopilotController],
contributesTo: ServerFeature.Copilot,
if: config => {
if (config.flavor.graphql) {
@@ -38,16 +38,16 @@ export class ChatPrompt {
) {
return new ChatPrompt(
options.name,
options.action,
options.model,
options.action || undefined,
options.model || undefined,
options.messages
);
}
constructor(
public readonly name: string,
public readonly action: string | null,
public readonly model: string | null,
public readonly action: string | undefined,
public readonly model: string | undefined,
private readonly messages: PromptMessage[]
) {
this.encoder = getTokenEncoder(model);
@@ -3,14 +3,16 @@ import assert from 'node:assert';
import { ClientOptions, OpenAI } from 'openai';
import {
ChatMessage,
ChatMessageRole,
CopilotCapability,
CopilotProviderType,
CopilotTextToEmbeddingProvider,
CopilotTextToTextProvider,
PromptMessage,
} from '../types';
const DEFAULT_DIMENSIONS = 256;
export class OpenAIProvider
implements CopilotTextToTextProvider, CopilotTextToEmbeddingProvider
{
@@ -50,7 +52,7 @@ export class OpenAIProvider
return OpenAIProvider.capabilities;
}
private chatToGPTMessage(messages: ChatMessage[]) {
private chatToGPTMessage(messages: PromptMessage[]) {
// filter redundant fields
return messages.map(message => ({
role: message.role,
@@ -63,7 +65,7 @@ export class OpenAIProvider
embeddings,
model,
}: {
messages?: ChatMessage[];
messages?: PromptMessage[];
embeddings?: string[];
model: string;
}) {
@@ -106,7 +108,7 @@ export class OpenAIProvider
// ====== text to text ======
async generateText(
messages: ChatMessage[],
messages: PromptMessage[],
model: string = 'gpt-3.5-turbo',
options: {
temperature?: number;
@@ -134,8 +136,8 @@ export class OpenAIProvider
}
async *generateTextStream(
messages: ChatMessage[],
model: string,
messages: PromptMessage[],
model: string = 'gpt-3.5-turbo',
options: {
temperature?: number;
maxTokens?: number;
@@ -179,7 +181,7 @@ export class OpenAIProvider
dimensions: number;
signal?: AbortSignal;
user?: string;
} = { dimensions: 256 }
} = { dimensions: DEFAULT_DIMENSIONS }
): Promise<number[][]> {
messages = Array.isArray(messages) ? messages : [messages];
this.checkParams({ embeddings: messages, model });
@@ -187,7 +189,7 @@ export class OpenAIProvider
const result = await this.instance.embeddings.create({
model: model,
input: messages,
dimensions: options.dimensions,
dimensions: options.dimensions || DEFAULT_DIMENSIONS,
user: options.user,
});
return result.data.map(e => e.embedding);
@@ -0,0 +1,260 @@
import {
Args,
Field,
ID,
InputType,
Mutation,
ObjectType,
Parent,
registerEnumType,
ResolveField,
Resolver,
} from '@nestjs/graphql';
import { SafeIntResolver } from 'graphql-scalars';
import { CurrentUser, Public } from '../../core/auth';
import { QuotaService } from '../../core/quota';
import { UserType } from '../../core/user';
import { PermissionService } from '../../core/workspaces/permission';
import {
MutexService,
PaymentRequiredException,
TooManyRequestsException,
} from '../../fundamentals';
import { ChatSessionService, ListHistoriesOptions } from './session';
import { AvailableModels, type ChatHistory, type ChatMessage } from './types';
registerEnumType(AvailableModels, { name: 'CopilotModel' });
// ================== Input Types ==================
@InputType()
class CreateChatSessionInput {
@Field(() => String)
workspaceId!: string;
@Field(() => String)
docId!: string;
@Field(() => String, {
description: 'An mark identifying which view to use to display the session',
nullable: true,
})
action!: string | undefined;
@Field(() => String, {
description: 'The prompt name to use for the session',
})
promptName!: string;
}
@InputType()
class QueryChatHistoriesInput implements Partial<ListHistoriesOptions> {
@Field(() => Boolean, { nullable: true })
action: boolean | undefined;
@Field(() => Number, { nullable: true })
limit: number | undefined;
@Field(() => Number, { nullable: true })
skip: number | undefined;
@Field(() => String, { nullable: true })
sessionId: string | undefined;
}
// ================== Return Types ==================
@ObjectType('ChatMessage')
class ChatMessageType implements Partial<ChatMessage> {
@Field(() => String)
role!: 'system' | 'assistant' | 'user';
@Field(() => String)
content!: string;
@Field(() => [String], { nullable: true })
attachments!: string[];
@Field(() => Date, { nullable: true })
createdAt!: Date | undefined;
}
@ObjectType('CopilotHistories')
class CopilotHistoriesType implements Partial<ChatHistory> {
@Field(() => String)
sessionId!: string;
@Field(() => String, {
description: 'An mark identifying which view to use to display the session',
})
action!: string;
@Field(() => Number, {
description: 'The number of tokens used in the session',
})
tokens!: number;
@Field(() => [ChatMessageType])
messages!: ChatMessageType[];
}
@ObjectType('CopilotQuota')
class CopilotQuotaType {
@Field(() => SafeIntResolver)
limit!: number;
@Field(() => SafeIntResolver)
used!: number;
}
// ================== Resolver ==================
@ObjectType('Copilot')
export class CopilotType {
@Field(() => ID, { nullable: true })
workspaceId!: string | undefined;
}
@Resolver(() => CopilotType)
export class CopilotResolver {
constructor(
private readonly permissions: PermissionService,
private readonly quota: QuotaService,
private readonly mutex: MutexService,
private readonly chatSession: ChatSessionService
) {}
@ResolveField(() => CopilotQuotaType, {
name: 'quota',
description: 'Get the quota of the user in the workspace',
complexity: 2,
})
async getQuota(@CurrentUser() user: CurrentUser) {
const quota = await this.quota.getUserQuota(user.id);
const limit = quota.feature.copilotActionLimit;
const actions = await this.chatSession.countUserActions(user.id);
const chats = await this.chatSession
.listHistories(user.id)
.then(histories =>
histories.reduce(
(acc, h) => acc + h.messages.filter(m => m.role === 'user').length,
0
)
);
return { limit, used: actions + chats };
}
@ResolveField(() => [String], {
description: 'Get the session list of chats in the workspace',
complexity: 2,
})
async chats(
@Parent() copilot: CopilotType,
@CurrentUser() user: CurrentUser
) {
if (!copilot.workspaceId) return [];
await this.permissions.checkCloudWorkspace(copilot.workspaceId, user.id);
return await this.chatSession.listSessions(user.id, copilot.workspaceId);
}
@ResolveField(() => [String], {
description: 'Get the session list of actions in the workspace',
complexity: 2,
})
async actions(
@Parent() copilot: CopilotType,
@CurrentUser() user: CurrentUser
) {
if (!copilot.workspaceId) return [];
await this.permissions.checkCloudWorkspace(copilot.workspaceId, user.id);
return await this.chatSession.listSessions(user.id, copilot.workspaceId, {
action: true,
});
}
@ResolveField(() => [CopilotHistoriesType], {})
async histories(
@Parent() copilot: CopilotType,
@CurrentUser() user: CurrentUser,
@Args('docId', { nullable: true }) docId?: string,
@Args({
name: 'options',
type: () => QueryChatHistoriesInput,
nullable: true,
})
options?: QueryChatHistoriesInput
) {
const workspaceId = copilot.workspaceId;
if (!workspaceId) {
return [];
} else if (docId) {
await this.permissions.checkCloudPagePermission(
workspaceId,
docId,
user.id
);
} else {
await this.permissions.checkCloudWorkspace(workspaceId, user.id);
}
return await this.chatSession.listHistories(
user.id,
workspaceId,
docId,
options
);
}
@Public()
@Mutation(() => String, {
description: 'Create a chat session',
})
async createCopilotSession(
@CurrentUser() user: CurrentUser,
@Args({ name: 'options', type: () => CreateChatSessionInput })
options: CreateChatSessionInput
) {
await this.permissions.checkCloudPagePermission(
options.workspaceId,
options.docId,
user.id
);
const lockFlag = `session:${user.id}:${options.workspaceId}`;
await using lock = await this.mutex.lock(lockFlag);
if (!lock) {
return new TooManyRequestsException('Server is busy');
}
const { limit, used } = await this.getQuota(user);
if (limit && Number.isFinite(limit) && used >= limit) {
return new PaymentRequiredException(
`You have reached the limit of actions in this workspace, please upgrade your plan.`
);
}
const session = await this.chatSession.create({
...options,
userId: user.id,
});
return session;
}
}
@Resolver(() => UserType)
export class UserCopilotResolver {
constructor(private readonly permissions: PermissionService) {}
@ResolveField(() => CopilotType)
async copilot(
@CurrentUser() user: CurrentUser,
@Args('workspaceId', { nullable: true }) workspaceId?: string
) {
if (workspaceId) {
await this.permissions.checkCloudWorkspace(workspaceId, user.id);
}
return { workspaceId };
}
}
@@ -11,6 +11,7 @@ import {
ChatMessageSchema,
getTokenEncoder,
PromptMessage,
PromptMessageSchema,
PromptParams,
} from './types';
@@ -105,37 +106,62 @@ export class ChatSession implements AsyncDisposable {
@Injectable()
export class ChatSessionService {
private readonly logger = new Logger(ChatSessionService.name);
constructor(
private readonly db: PrismaClient,
private readonly prompt: PromptService
) {}
private async setSession(state: ChatSessionState): Promise<void> {
await this.db.aiSession.upsert({
where: {
id: state.sessionId,
},
update: {
messages: {
create: state.messages.map((m, idx) => ({ idx, ...m })),
},
},
create: {
id: state.sessionId,
messages: { create: state.messages },
// connect
user: { connect: { id: state.userId } },
workspace: { connect: { id: state.workspaceId } },
doc: {
connect: {
id_workspaceId: {
id: state.docId,
private async setSession(state: ChatSessionState): Promise<string> {
return await this.db.$transaction(async tx => {
let sessionId = state.sessionId;
// find existing session if session is chat session
if (!state.prompt.action) {
const { id } =
(await tx.aiSession.findFirst({
where: {
userId: state.userId,
workspaceId: state.workspaceId,
docId: state.docId,
prompt: { action: { equals: null } },
},
select: { id: true },
})) || {};
if (id) sessionId = id;
}
await tx.aiSession.upsert({
where: {
id: sessionId,
userId: state.userId,
},
update: {
messages: {
// delete old messages
deleteMany: {},
create: state.messages.map(m => ({
...m,
params: m.params || undefined,
})),
},
},
prompt: { connect: { name: state.prompt.name } },
},
create: {
id: sessionId,
workspaceId: state.workspaceId,
docId: state.docId,
messages: {
create: state.messages.map(m => ({
...m,
params: m.params || undefined,
})),
},
// connect
user: { connect: { id: state.userId } },
prompt: { connect: { name: state.prompt.name } },
},
});
return sessionId;
});
}
@@ -171,6 +197,7 @@ export class ChatSessionService {
})
.then(async session => {
if (!session) return;
const messages = ChatMessageSchema.array().safeParse(session.messages);
return {
@@ -184,18 +211,58 @@ export class ChatSessionService {
});
}
async listHistories(
private calculateTokenSize(
messages: PromptMessage[],
model: AvailableModel
): number {
const encoder = getTokenEncoder(model);
return messages
.map(m => encoder?.encode_ordinary(m.content).length || 0)
.reduce((total, length) => total + length, 0);
}
async countUserActions(userId: string): Promise<number> {
return await this.db.aiSession.count({
where: { userId, prompt: { action: { not: null } } },
});
}
async listSessions(
userId: string,
workspaceId: string,
docId: string,
options: ListHistoriesOptions
options?: { docId?: string; action?: boolean }
): Promise<string[]> {
return await this.db.aiSession
.findMany({
where: {
userId,
workspaceId,
docId: workspaceId === options?.docId ? undefined : options?.docId,
prompt: {
action: options?.action ? { not: null } : null,
},
},
select: { id: true },
})
.then(sessions => sessions.map(({ id }) => id));
}
async listHistories(
userId: string,
workspaceId?: string,
docId?: string,
options?: ListHistoriesOptions
): Promise<ChatHistory[]> {
return await this.db.aiSession
.findMany({
where: {
userId,
workspaceId: workspaceId,
docId: workspaceId === docId ? undefined : docId,
prompt: { action: { not: null } },
id: options.sessionId ? { equals: options.sessionId } : undefined,
prompt: {
action: options?.action ? { not: null } : null,
},
id: options?.sessionId ? { equals: options.sessionId } : undefined,
},
select: {
id: true,
@@ -210,20 +277,33 @@ export class ChatSessionService {
},
},
},
take: options.limit,
skip: options.skip,
take: options?.limit,
skip: options?.skip,
orderBy: { createdAt: 'desc' },
})
.then(sessions =>
sessions
.map(({ id, prompt, messages }) => {
const ret = ChatMessageSchema.array().safeParse(messages);
if (ret.success) {
const encoder = getTokenEncoder(prompt.model as AvailableModel);
const tokens = ret.data
.map(m => encoder?.encode_ordinary(m.content).length || 0)
.reduce((total, length) => total + length, 0);
return { sessionId: id, tokens, messages: ret.data };
try {
const ret = PromptMessageSchema.array().safeParse(messages);
if (ret.success) {
const tokens = this.calculateTokenSize(
ret.data,
prompt.model as AvailableModel
);
return {
sessionId: id,
action: prompt.action || undefined,
tokens,
messages: ret.data,
};
} else {
this.logger.error(
`Unexpected message schema: ${JSON.stringify(ret.error)}`
);
}
} catch (e) {
this.logger.error('Unexpected error in listHistories', e);
}
return undefined;
})
@@ -238,8 +318,12 @@ export class ChatSessionService {
this.logger.error(`Prompt not found: ${options.promptName}`);
throw new Error('Prompt not found');
}
await this.setSession({ ...options, sessionId, prompt, messages: [] });
return sessionId;
return await this.setSession({
...options,
sessionId,
prompt,
messages: [],
});
}
/**
@@ -76,8 +76,9 @@ export type ChatMessage = z.infer<typeof ChatMessageSchema>;
export const ChatHistorySchema = z
.object({
sessionId: z.string(),
action: z.string().optional(),
tokens: z.number(),
messages: z.array(ChatMessageSchema),
messages: z.array(PromptMessageSchema.or(ChatMessageSchema)),
})
.strict();
@@ -104,8 +105,8 @@ export interface CopilotProvider {
export interface CopilotTextToTextProvider extends CopilotProvider {
generateText(
messages: PromptMessage[],
model: string,
options: {
model?: string,
options?: {
temperature?: number;
maxTokens?: number;
signal?: AbortSignal;
@@ -114,8 +115,8 @@ export interface CopilotTextToTextProvider extends CopilotProvider {
): Promise<string>;
generateTextStream(
messages: PromptMessage[],
model: string,
options: {
model?: string,
options?: {
temperature?: number;
maxTokens?: number;
signal?: AbortSignal;