@lota-sdk/core 0.1.16 → 0.1.18

package/src/runtime/team-consultation-orchestrator.ts

@@ -9,7 +9,6 @@ import type { ReadableUploadMetadataLike } from './chat-attachments'
 import type { RepoSectionName } from './indexed-repositories-policy'
 import { buildTeamConsultationFailureMessage } from './team-consultation-prompts'
 import { extractMessageText } from './workstream-chat-helpers'
-import type { ReasoningProfileName } from './workstream-routing-policy'
 
 export type DefaultRepoSections = RepoSectionName[]
 const TEAM_CONSULTATION_TIMEOUT_MS = 90_000
@@ -66,7 +65,6 @@ export interface TeamConsultationParticipantRunner {
     agentId: string,
     params: {
       task: string
-      reasoningProfile: ReasoningProfileName
       systemWorkspaceDetails?: string
       preSeededMemoriesSection?: string
       retrievedKnowledgeSection?: string
@@ -90,7 +88,6 @@ export interface CreateConsultTeamToolParams {
   latestUserMessageId: string
   availableUploads: ReadableUploadMetadataLike[]
   defaultRepoSectionsByAgent: Record<string, DefaultRepoSections | undefined>
-  reasoningProfile: 'fast' | 'standard' | 'deep'
   systemWorkspaceDetails?: string
   getPreSeededMemoriesSection: (agentId: string) => Promise<string | undefined>
   retrievedKnowledgeSection?: string
@@ -135,7 +132,6 @@ export function createConsultTeamTool(params: CreateConsultTeamToolParams) {
         try {
           const { agent, observer } = await params.participantRunner.buildParticipantAgent(agentId, {
             task,
-            reasoningProfile: params.reasoningProfile,
             systemWorkspaceDetails: params.systemWorkspaceDetails,
             preSeededMemoriesSection: await params.getPreSeededMemoriesSection(agentId),
             retrievedKnowledgeSection: params.retrievedKnowledgeSection,

package/src/services/agent-executor.service.ts

@@ -146,7 +146,6 @@ class AgentExecutorService {
       mode: dispatchMode,
       onboardingActive: false,
       linearInstalled: Boolean(linearInstallation),
-      reasoningProfile: 'standard',
       additionalInstructionSections: [
         buildOwnershipDispatchContextSection({
           node: params.nodeSpec,

package/src/services/document-chunk.service.ts

@@ -1,5 +1,3 @@
-import { createHash } from 'node:crypto'
-
 import { chunkMarkdownDocument, chunkPagedDocument, chunkPlainTextDocument } from '../document/org-document-chunking'
 import type { ParsedDocumentChunk } from '../document/org-document-chunking'
 import { getDefaultEmbeddings } from '../embeddings/provider'
@@ -56,7 +54,7 @@ export class DocumentChunkService {
   }
 
   hashContent(content: string): string {
-    return createHash('sha256').update(content).digest('hex')
+    return new Bun.CryptoHasher('sha256').update(content).digest('hex')
   }
 
   // Uses 4 chars/token (conservative estimate for document content which tends

package/src/services/index.ts

@@ -30,6 +30,7 @@ export * from './plugin-executor.service'
 export * from './recent-activity-title.service'
 export * from './recent-activity.service'
 export * from './skill-resolver.service'
+export * from './social-chat-history.service'
 export * from './system-executor.service'
 export * from './user.service'
 export * from './workstream-message.service'

package/src/services/memory.service.ts

@@ -722,6 +722,8 @@ class MemoryService {
     input,
     output,
     sourceId,
+    source = 'chat',
+    sourceMetadata,
     onboardStatus,
     agentName,
     historyMessages = [],
@@ -733,6 +735,8 @@ class MemoryService {
     input: string
     output: string
     sourceId?: string
+    source?: string
+    sourceMetadata?: Record<string, unknown>
     onboardStatus?: string
     agentName?: string
     historyMessages?: Array<{ role: 'user' | 'agent'; content: string; agentName?: string }>
@@ -789,7 +793,7 @@ class MemoryService {
         scopeId: orgScopeId,
         memoryType: ORG_MEMORY_TYPE,
         ...(typeof assessedImportance === 'number' ? { importance: assessedImportance } : {}),
-        metadata: { orgId, source: 'chat', ...(sourceId ? { sourceId } : {}), ...assessmentMetadata },
+        metadata: { orgId, source, ...(sourceId ? { sourceId } : {}), ...sourceMetadata, ...assessmentMetadata },
       },
     ]
 
@@ -803,8 +807,9 @@ class MemoryService {
           orgId,
           agentName: scopedAgentName,
           memoryScope: 'agent',
-          source: 'chat',
+          source,
           ...(sourceId ? { sourceId } : {}),
+          ...sourceMetadata,
           ...assessmentMetadata,
         },
       })

package/src/services/recent-activity.service.ts

@@ -1,5 +1,3 @@
-import { createHash } from 'node:crypto'
-
 import {
   RecentActivityDeepLinkSchema,
   RecentActivityEventInputSchema,
@@ -70,7 +68,7 @@ type RecentActivityEventRow = z.infer<typeof RecentActivityEventRowSchema>
 type RecentActivityRow = z.infer<typeof RecentActivityRowSchema>
 
 function buildDeterministicRecordId(table: string, key: string): RecordId {
-  const digest = createHash('sha256').update(key).digest('hex')
+  const digest = new Bun.CryptoHasher('sha256').update(key).digest('hex')
   return new RecordId(table, digest)
 }
 

package/src/services/social-chat-history.service.ts

@@ -0,0 +1,197 @@
+import { z } from 'zod'
+
+import { getRedisConnection } from '../redis'
+import type { LotaRuntimeBackgroundCursor, LotaRuntimeBackgroundCursorKind } from '../runtime/runtime-extensions'
+
+const DEFAULT_SOCIAL_CHAT_HISTORY_PREFIX = 'lota:social:history'
+
+const SocialChatMessageRoleSchema = z.enum(['user', 'assistant'])
+const SocialChatSourceSchema = z.literal('social')
+const SocialChatPlatformSchema = z.literal('slack')
+
+const SocialChatHistoryMessageSchema = z.object({
+  source: SocialChatSourceSchema,
+  sourceId: z.string().trim().min(1),
+  platform: SocialChatPlatformSchema,
+  workspaceId: z.string().trim().min(1),
+  channelId: z.string().trim().min(1),
+  threadId: z.string().trim().min(1),
+  messageId: z.string().trim().min(1),
+  role: SocialChatMessageRoleSchema,
+  parts: z.array(z.record(z.string(), z.unknown())),
+  metadata: z.record(z.string(), z.unknown()).optional(),
+  cursor: z.object({ createdAt: z.coerce.date(), id: z.string().trim().min(1) }),
+})
+
+export type SocialChatHistoryMessage = z.infer<typeof SocialChatHistoryMessageSchema>
+
+let socialChatHistoryPrefix = DEFAULT_SOCIAL_CHAT_HISTORY_PREFIX
+
+function trimConfiguredPrefix(value: string | undefined): string {
+  const normalized = value?.trim()
+  return normalized && normalized.length > 0 ? normalized : DEFAULT_SOCIAL_CHAT_HISTORY_PREFIX
+}
+
+function compareCursorOrder(left: LotaRuntimeBackgroundCursor, right: LotaRuntimeBackgroundCursor): number {
+  const timeDiff = left.createdAt.getTime() - right.createdAt.getTime()
+  if (timeDiff !== 0) return timeDiff
+  return left.id.localeCompare(right.id)
+}
+
+class SocialChatHistoryService {
+  configure(params?: { keyPrefix?: string }): void {
+    socialChatHistoryPrefix = trimConfiguredPrefix(params?.keyPrefix)
+  }
+
+  private messageStorageKey(message: {
+    platform: 'slack'
+    workspaceId: string
+    threadId: string
+    messageId: string
+  }): string {
+    return `${socialChatHistoryPrefix}:message:${message.platform}:${message.workspaceId}:${message.threadId}:${message.messageId}`
+  }
+
+  private threadIndexKey(workspaceId: string, threadId: string): string {
+    return `${socialChatHistoryPrefix}:thread:${workspaceId}:${threadId}`
+  }
+
+  private workspaceIndexKey(workspaceId: string): string {
+    return `${socialChatHistoryPrefix}:workspace:${workspaceId}`
+  }
+
+  private cursorKey(kind: LotaRuntimeBackgroundCursorKind, workspaceId: string): string {
+    return `${socialChatHistoryPrefix}:cursor:${kind}:${workspaceId}`
+  }
+
+  private serializeMessage(message: SocialChatHistoryMessage): string {
+    return JSON.stringify({
+      ...message,
+      cursor: { ...message.cursor, createdAt: message.cursor.createdAt.toISOString() },
+    })
+  }
+
+  private parseStoredMessage(value: string | null): SocialChatHistoryMessage | null {
+    if (!value) return null
+    try {
+      const parsed = SocialChatHistoryMessageSchema.safeParse(JSON.parse(value))
+      return parsed.success ? parsed.data : null
+    } catch {
+      return null
+    }
+  }
+
+  private serializeCursor(cursor: LotaRuntimeBackgroundCursor): string {
+    return JSON.stringify({ ...cursor, createdAt: cursor.createdAt.toISOString() })
+  }
+
+  private parseCursor(value: string | null): LotaRuntimeBackgroundCursor | null {
+    if (!value) return null
+    try {
+      const parsed = z.object({ createdAt: z.coerce.date(), id: z.string().trim().min(1) }).safeParse(JSON.parse(value))
+      return parsed.success ? parsed.data : null
+    } catch {
+      return null
+    }
+  }
+
+  async upsertMessages(messages: SocialChatHistoryMessage[]): Promise<SocialChatHistoryMessage[]> {
+    if (messages.length === 0) return []
+
+    const redis = getRedisConnection()
+    const normalizedMessages = messages.map((message) => SocialChatHistoryMessageSchema.parse(message))
+    const multi = redis.multi()
+
+    for (const message of normalizedMessages) {
+      const storageKey = this.messageStorageKey(message)
+      const score = message.cursor.createdAt.getTime()
+      multi.set(storageKey, this.serializeMessage(message))
+      multi.zadd(this.threadIndexKey(message.workspaceId, message.threadId), score, storageKey)
+      multi.zadd(this.workspaceIndexKey(message.workspaceId), score, storageKey)
+    }
+
+    await multi.exec()
+    return normalizedMessages
+  }
+
+  async listThreadMessages(params: { workspaceId: string; threadId: string }): Promise<SocialChatHistoryMessage[]> {
+    const redis = getRedisConnection()
+    const storageKeys = await redis.zrange(this.threadIndexKey(params.workspaceId, params.threadId), 0, -1)
+    if (storageKeys.length === 0) return []
+
+    const storedValues = await redis.mget(storageKeys)
+    return storedValues
+      .map((value) => this.parseStoredMessage(value))
+      .filter((message): message is SocialChatHistoryMessage => message !== null)
+      .sort((left, right) => compareCursorOrder(left.cursor, right.cursor))
+  }
+
+  async listWorkspaceMessages(params: {
+    workspaceId: string
+    cursor: LotaRuntimeBackgroundCursor | null
+    onboardingCutoff: Date | null
+  }): Promise<SocialChatHistoryMessage[]> {
+    const redis = getRedisConnection()
+    const indexKey = this.workspaceIndexKey(params.workspaceId)
+    const scoreStart =
+      params.cursor?.createdAt.getTime() ??
+      (params.onboardingCutoff ? params.onboardingCutoff.getTime() : Number.NEGATIVE_INFINITY)
+    const storageKeys =
+      params.cursor || params.onboardingCutoff
+        ? await redis.zrangebyscore(indexKey, scoreStart, '+inf')
+        : await redis.zrange(indexKey, 0, -1)
+
+    if (storageKeys.length === 0) return []
+
+    const storedValues = await redis.mget(storageKeys)
+    return storedValues
+      .map((value) => this.parseStoredMessage(value))
+      .filter((message): message is SocialChatHistoryMessage => message !== null)
+      .filter((message) => {
+        if (params.cursor) {
+          return compareCursorOrder(message.cursor, params.cursor) > 0
+        }
+        if (params.onboardingCutoff) {
+          return message.cursor.createdAt.getTime() > params.onboardingCutoff.getTime()
+        }
+        return true
+      })
+      .sort((left, right) => compareCursorOrder(left.cursor, right.cursor))
+  }
+
+  async hasWorkspaceMessages(params: {
+    workspaceId: string
+    cursor: LotaRuntimeBackgroundCursor | null
+    onboardingCutoff: Date | null
+  }): Promise<boolean> {
+    const messages = await this.listWorkspaceMessages({
+      workspaceId: params.workspaceId,
+      cursor: params.cursor,
+      onboardingCutoff: params.onboardingCutoff,
+    })
+    return messages.length > 0
+  }
+
+  async getBackgroundCursor(
+    kind: LotaRuntimeBackgroundCursorKind,
+    workspaceId: string,
+  ): Promise<LotaRuntimeBackgroundCursor | null> {
+    const redis = getRedisConnection()
+    return this.parseCursor(await redis.get(this.cursorKey(kind, workspaceId)))
+  }
+
+  async setBackgroundCursor(
+    kind: LotaRuntimeBackgroundCursorKind,
+    workspaceId: string,
+    cursor: LotaRuntimeBackgroundCursor,
+  ): Promise<void> {
+    const redis = getRedisConnection()
+    await redis.set(this.cursorKey(kind, workspaceId), this.serializeCursor(cursor))
+  }
+}
+
+export const socialChatHistoryService = new SocialChatHistoryService()
+
+export function configureSocialChatHistory(params?: { keyPrefix?: string }): void {
+  socialChatHistoryService.configure(params)
+}

package/src/services/workstream-message.service.ts

@@ -1,5 +1,3 @@
-import { createHash } from 'node:crypto'
-
 import { parseRowMetadata, recordIdSchema, requireTimestamp, withCreatedAtMetadata } from '@lota-sdk/shared'
 import type { ChatMessage } from '@lota-sdk/shared'
 import { RecordId, surql } from 'surrealdb'
@@ -30,7 +28,7 @@ function toMessageId(value: string | RecordIdRef): string {
  */
 function toWorkstreamMessageRowId(workstreamId: RecordIdRef, messageId: string): RecordId {
   const workstreamStr = recordIdToString(workstreamId, TABLES.WORKSTREAM)
-  const digest = createHash('sha256').update(`${workstreamStr}\0${messageId}`).digest('hex').slice(0, 32)
+  const digest = new Bun.CryptoHasher('sha256').update(`${workstreamStr}\0${messageId}`).digest('hex').slice(0, 32)
   return new RecordId(TABLES.WORKSTREAM_MESSAGE, digest)
 }
 

package/src/services/workstream-turn-preparation.service.ts

@@ -69,11 +69,6 @@ import {
   toHistoryMessages,
   toOptionalTrimmedString,
 } from '../runtime/workstream-chat-helpers'
-import {
-  classifyHighImpactResponse,
-  classifyPolicyClasses,
-  resolveReasoningProfile,
-} from '../runtime/workstream-routing-policy'
 import type { WorkstreamState } from '../runtime/workstream-state'
 import { chatRunRegistry } from '../services/chat-run-registry.service'
 import type { NormalizedWorkstream, WorkstreamRecord } from '../services/workstream.types'
@@ -274,7 +269,6 @@ interface StreamAgentResponseContext {
   onboardingActive: boolean
   linearInstalled: boolean
   githubInstalled: boolean
-  reasoningProfileName: string
   buildContextResult: Record<string, unknown> | null
   getExecutionPlanInstructionSections: () => Promise<string[] | undefined>
   getPreSeededMemoriesSection: (agentId: string) => Promise<string | undefined>
@@ -324,7 +318,6 @@ async function streamAgentResponse(
       onboardingActive: ctx.onboardingActive,
       linearInstalled: ctx.linearInstalled,
       githubInstalled: ctx.githubInstalled,
-      reasoningProfile: ctx.reasoningProfileName,
       skills: streamParams.skills,
       additionalInstructionSections: streamParams.additionalInstructionSections,
       context: ctx.buildContextResult,
@@ -345,7 +338,6 @@ async function streamAgentResponse(
     skills: streamParams.skills,
     onboardingActive: ctx.onboardingActive,
     linearInstalled: ctx.linearInstalled,
-    reasoningProfile: ctx.reasoningProfileName,
     systemWorkspaceDetails: ctx.promptContext.systemWorkspaceDetails,
     preSeededMemoriesSection,
     retrievedKnowledgeSection: ctx.retrievedKnowledgeSection,
@@ -753,15 +745,6 @@ export async function prepareWorkstreamRunCore(params: WorkstreamRunCoreParams):
     'getInstallationForOrganization',
   ])
 
-  const highImpactAssessment = classifyHighImpactResponse({ message: messageText })
-  const policyAssessment = classifyPolicyClasses({ message: messageText })
-  const reasoningProfile = resolveReasoningProfile({
-    message: messageText,
-    forceDeep: highImpactAssessment.classes.length > 0 || policyAssessment.classes.length > 0,
-    explicitProfile: onboardingActive ? 'standard' : undefined,
-  })
-  timer.step('reasoning-classification')
-
   const [linearInstallation, githubInstallation, indexedRepoContext, recentDomainEvents, promptSummary] =
     await Promise.all([
       getLinearInstallationByOrgId ? (getLinearInstallationByOrgId(orgRef) as Promise<unknown>) : Promise.resolve(null),
@@ -968,9 +951,6 @@ export async function prepareWorkstreamRunCore(params: WorkstreamRunCoreParams):
           const buildAgentMetadataPatch = (agentId: string, agentName: string): NonNullable<MessageMetadata> => ({
             agentId,
             agentName,
-            reasoningProfile: reasoningProfile.name,
-            highImpactClasses: highImpactAssessment.classes,
-            policyClasses: policyAssessment.classes,
             semanticTerminationReason: 'none',
           })
 
@@ -1007,7 +987,6 @@ export async function prepareWorkstreamRunCore(params: WorkstreamRunCoreParams):
             onboardingActive,
             linearInstalled,
             githubInstalled,
-            reasoningProfileName: reasoningProfile.name,
             buildContextResult,
             getExecutionPlanInstructionSections,
             getPreSeededMemoriesSection,
@@ -1137,7 +1116,6 @@ export async function prepareWorkstreamRunCore(params: WorkstreamRunCoreParams):
                 mode: 'fixedWorkstreamMode',
                 onboardingActive,
                 linearInstalled,
-                reasoningProfile: reasoningProfile.name,
                 systemWorkspaceDetails: promptContext.systemWorkspaceDetails,
                 preSeededMemoriesSection: specialistPreSeededMemories,
                 retrievedKnowledgeSection,
@@ -1234,7 +1212,6 @@ export async function prepareWorkstreamRunCore(params: WorkstreamRunCoreParams):
                   availableUploads: listReadableUploads(),
                   provideRepoTool: indexedRepoContext.provideRepoTool,
                   defaultRepoSectionsByAgent: indexedRepoContext.defaultSectionsByAgent as never,
-                  reasoningProfile: reasoningProfile.name,
                   systemWorkspaceDetails: promptContext.systemWorkspaceDetails,
                   getPreSeededMemoriesSection,
                   retrievedKnowledgeSection,

package/src/system-agents/context-compaction.agent.ts

@@ -1,6 +1,7 @@
 import { ToolLoopAgent } from 'ai'
 
-import { bifrostOpenRouterResponseHealingModel } from '../bifrost/bifrost'
+import { aiGatewayOpenRouterResponseHealingModel } from '../ai-gateway/ai-gateway'
+import { buildAiGatewayCacheHeaders } from '../ai-gateway/cache-headers'
 import {
   OPENROUTER_STRUCTURED_HELPER_MODEL_ID,
   OPENROUTER_HIGH_REASONING_PROVIDER_OPTIONS,
@@ -39,7 +40,8 @@ Return valid data for:
 export function createContextCompactionAgent(options: CreateHelperToolLoopAgentOptions) {
   return new ToolLoopAgent({
     id: 'context-compaction',
-    model: bifrostOpenRouterResponseHealingModel(OPENROUTER_STRUCTURED_HELPER_MODEL_ID),
+    model: aiGatewayOpenRouterResponseHealingModel(OPENROUTER_STRUCTURED_HELPER_MODEL_ID),
+    headers: buildAiGatewayCacheHeaders('context-compaction'),
     providerOptions: OPENROUTER_HIGH_REASONING_PROVIDER_OPTIONS,
     ...resolveHelperAgentOptions(options, { instructions: CONTEXT_COMPACTION_PROMPT }),
   })

package/src/system-agents/delegated-agent-factory.ts

@@ -17,6 +17,7 @@ interface DelegatedAgentDefinition {
   providerOptions?: AgentProviderOptions
   instructions: string
   tools?: ToolSet
+  headers?: Record<string, string>
   maxSteps?: number
   maxOutputTokens?: number
   temperature?: number
@@ -165,6 +166,7 @@ export function createDelegatedAgentTool(definition: DelegatedAgentDefinition):
               id: definition.id,
               model: resolveAgentModel(definition.model),
               ...(definition.providerOptions ? { providerOptions: definition.providerOptions } : {}),
+              ...(definition.headers ? { headers: definition.headers } : {}),
               instructions: buildDelegatedAgentInstructions(definition.instructions, agentTools),
               tools: agentTools,
               maxOutputTokens: definition.maxOutputTokens ?? DEFAULT_DELEGATED_AGENT_MAX_OUTPUT_TOKENS,
@@ -206,6 +208,7 @@ export function createDelegatedAgentToolWithContext<TContext>(
               id: definition.id,
               model: resolveAgentModel(definition.model),
               ...(definition.providerOptions ? { providerOptions: definition.providerOptions } : {}),
+              ...(definition.headers ? { headers: definition.headers } : {}),
               instructions: buildDelegatedAgentInstructions(definition.instructions, agentTools),
               tools: agentTools,
               maxOutputTokens: definition.maxOutputTokens ?? DEFAULT_DELEGATED_AGENT_MAX_OUTPUT_TOKENS,

package/src/system-agents/memory-reranker.agent.ts

@@ -1,9 +1,10 @@
 import { ToolLoopAgent } from 'ai'
 
-import { bifrostOpenRouterResponseHealingModel } from '../bifrost/bifrost'
+import { aiGatewayOpenRouterResponseHealingModel } from '../ai-gateway/ai-gateway'
+import { buildAiGatewayCacheHeaders } from '../ai-gateway/cache-headers'
 import {
   OPENROUTER_LOW_REASONING_PROVIDER_OPTIONS,
-  OPENROUTER_STRUCTURED_REASONING_MODEL_ID,
+  OPENROUTER_STRUCTURED_HELPER_MODEL_ID,
 } from '../config/model-constants'
 import type { CreateHelperToolLoopAgentOptions } from '../runtime/agent-types'
 import { resolveHelperAgentOptions } from './helper-agent-options'
@@ -31,7 +32,8 @@ Set every item.relevance as a string; use empty string when no reason is needed.
 export function createMemoryRerankerAgent(options: CreateHelperToolLoopAgentOptions) {
   return new ToolLoopAgent({
     id: 'memory-reranker',
-    model: bifrostOpenRouterResponseHealingModel(OPENROUTER_STRUCTURED_REASONING_MODEL_ID),
+    model: aiGatewayOpenRouterResponseHealingModel(OPENROUTER_STRUCTURED_HELPER_MODEL_ID),
+    headers: buildAiGatewayCacheHeaders('memory-reranker'),
     providerOptions: OPENROUTER_LOW_REASONING_PROVIDER_OPTIONS,
     ...resolveHelperAgentOptions(options),
   })

package/src/system-agents/memory.agent.ts

@@ -1,6 +1,7 @@
 import { ToolLoopAgent } from 'ai'
 
-import { bifrostOpenRouterResponseHealingModel } from '../bifrost/bifrost'
+import { aiGatewayOpenRouterResponseHealingModel } from '../ai-gateway/ai-gateway'
+import { buildAiGatewayCacheHeaders } from '../ai-gateway/cache-headers'
 import {
   OPENROUTER_STRUCTURED_HELPER_MODEL_ID,
   OPENROUTER_HIGH_REASONING_PROVIDER_OPTIONS,
@@ -51,7 +52,8 @@ Return only the schema fields with no extra formatting.
 export function createOrgMemoryAgent(options: CreateHelperToolLoopAgentOptions) {
   return new ToolLoopAgent({
     id: 'org-memory',
-    model: bifrostOpenRouterResponseHealingModel(OPENROUTER_STRUCTURED_HELPER_MODEL_ID),
+    model: aiGatewayOpenRouterResponseHealingModel(OPENROUTER_STRUCTURED_HELPER_MODEL_ID),
+    headers: buildAiGatewayCacheHeaders('org-memory'),
     providerOptions: OPENROUTER_HIGH_REASONING_PROVIDER_OPTIONS,
     ...resolveHelperAgentOptions(options),
   })

package/src/system-agents/recent-activity-title-refiner.agent.ts

@@ -1,6 +1,7 @@
 import { ToolLoopAgent } from 'ai'
 
-import { bifrostModel } from '../bifrost/bifrost'
+import { aiGatewayModel } from '../ai-gateway/ai-gateway'
+import { buildAiGatewayCacheHeaders } from '../ai-gateway/cache-headers'
 import { getLeadAgentDisplayName } from '../config/agent-defaults'
 import {
   OPENROUTER_STRUCTURED_HELPER_MODEL_ID,
@@ -78,7 +79,8 @@ Return only the title text. No quotes, labels, JSON, markdown, or explanation.
 export function createRecentActivityTitleRefinerAgent(options: CreateHelperToolLoopAgentOptions) {
   return new ToolLoopAgent({
     id: 'recent-activity-title-refiner',
-    model: bifrostModel(OPENROUTER_STRUCTURED_HELPER_MODEL_ID),
+    model: aiGatewayModel(OPENROUTER_STRUCTURED_HELPER_MODEL_ID),
+    headers: buildAiGatewayCacheHeaders('recent-activity-title-refiner'),
     providerOptions: OPENROUTER_HIGH_REASONING_PROVIDER_OPTIONS,
     ...resolveHelperAgentOptions(options, {
       instructions: buildRecentActivityTitleRefinerPrompt(),

package/src/system-agents/regular-chat-memory-digest.agent.ts

@@ -1,6 +1,7 @@
 import { ToolLoopAgent } from 'ai'
 
-import { bifrostOpenRouterResponseHealingModel } from '../bifrost/bifrost'
+import { aiGatewayOpenRouterResponseHealingModel } from '../ai-gateway/ai-gateway'
+import { buildAiGatewayCacheHeaders } from '../ai-gateway/cache-headers'
 import {
   OPENROUTER_STRUCTURED_HELPER_MODEL_ID,
   OPENROUTER_HIGH_REASONING_PROVIDER_OPTIONS,
@@ -65,7 +66,8 @@ Return only schema fields.
 export function createRegularChatMemoryDigestAgent(options: CreateHelperToolLoopAgentOptions) {
   return new ToolLoopAgent({
     id: 'regular-chat-memory-digest',
-    model: bifrostOpenRouterResponseHealingModel(OPENROUTER_STRUCTURED_HELPER_MODEL_ID),
+    model: aiGatewayOpenRouterResponseHealingModel(OPENROUTER_STRUCTURED_HELPER_MODEL_ID),
+    headers: buildAiGatewayCacheHeaders('regular-chat-memory-digest'),
     providerOptions: OPENROUTER_HIGH_REASONING_PROVIDER_OPTIONS,
     ...resolveHelperAgentOptions(options, {
       instructions: regularChatMemoryDigestPrompt,

package/src/system-agents/skill-extractor.agent.ts

@@ -1,7 +1,8 @@
 import { ToolLoopAgent } from 'ai'
 import { z } from 'zod'
 
-import { bifrostOpenRouterResponseHealingModel } from '../bifrost/bifrost'
+import { aiGatewayOpenRouterResponseHealingModel } from '../ai-gateway/ai-gateway'
+import { buildAiGatewayCacheHeaders } from '../ai-gateway/cache-headers'
 import {
   OPENROUTER_STRUCTURED_HELPER_MODEL_ID,
   OPENROUTER_HIGH_REASONING_PROVIDER_OPTIONS,
@@ -78,7 +79,8 @@ export type SkillCandidate = z.infer<typeof SkillCandidateSchema>
 export function createSkillExtractorAgent(options: CreateHelperToolLoopAgentOptions) {
   return new ToolLoopAgent({
     id: 'skill-extractor',
-    model: bifrostOpenRouterResponseHealingModel(OPENROUTER_STRUCTURED_HELPER_MODEL_ID),
+    model: aiGatewayOpenRouterResponseHealingModel(OPENROUTER_STRUCTURED_HELPER_MODEL_ID),
+    headers: buildAiGatewayCacheHeaders('skill-extractor'),
     providerOptions: OPENROUTER_HIGH_REASONING_PROVIDER_OPTIONS,
     ...resolveHelperAgentOptions(options, {
       instructions: skillExtractorPrompt,

package/src/system-agents/skill-manager.agent.ts

@@ -1,7 +1,8 @@
 import { ToolLoopAgent } from 'ai'
 import { z } from 'zod'
 
-import { bifrostOpenRouterResponseHealingModel } from '../bifrost/bifrost'
+import { aiGatewayOpenRouterResponseHealingModel } from '../ai-gateway/ai-gateway'
+import { buildAiGatewayCacheHeaders } from '../ai-gateway/cache-headers'
 import {
   OPENROUTER_STRUCTURED_HELPER_MODEL_ID,
   OPENROUTER_HIGH_REASONING_PROVIDER_OPTIONS,
@@ -68,7 +69,8 @@ export const SkillManagerOutputSchema = z.object({
 export function createSkillManagerAgent(options: CreateHelperToolLoopAgentOptions) {
   return new ToolLoopAgent({
     id: 'skill-manager',
-    model: bifrostOpenRouterResponseHealingModel(OPENROUTER_STRUCTURED_HELPER_MODEL_ID),
+    model: aiGatewayOpenRouterResponseHealingModel(OPENROUTER_STRUCTURED_HELPER_MODEL_ID),
+    headers: buildAiGatewayCacheHeaders('skill-manager'),
     providerOptions: OPENROUTER_HIGH_REASONING_PROVIDER_OPTIONS,
     ...resolveHelperAgentOptions(options, {
       instructions: skillManagerPrompt,

package/src/system-agents/title-generator.agent.ts

@@ -1,6 +1,7 @@
 import { ToolLoopAgent } from 'ai'
 
-import { bifrostModel } from '../bifrost/bifrost'
+import { aiGatewayModel } from '../ai-gateway/ai-gateway'
+import { buildAiGatewayCacheHeaders } from '../ai-gateway/cache-headers'
 import {
   OPENROUTER_FAST_REASONING_MODEL_ID,
   OPENROUTER_MINIMAL_REASONING_PROVIDER_OPTIONS,
@@ -32,7 +33,8 @@ Return only the title text. No quotes, no labels, no explanation.
 export function createWorkstreamTitleGeneratorAgent(options: CreateHelperToolLoopAgentOptions) {
   return new ToolLoopAgent({
     id: 'workstream-title-generator',
-    model: bifrostModel(OPENROUTER_FAST_REASONING_MODEL_ID),
+    model: aiGatewayModel(OPENROUTER_FAST_REASONING_MODEL_ID),
+    headers: buildAiGatewayCacheHeaders('workstream-title-generator'),
     providerOptions: OPENROUTER_MINIMAL_REASONING_PROVIDER_OPTIONS,
     ...resolveHelperAgentOptions(options, {
       instructions: WORKSTREAM_TITLE_GENERATOR_PROMPT,

package/src/tools/research-topic.tool.ts

@@ -1,4 +1,5 @@
-import { bifrostChatModel } from '../bifrost/bifrost'
+import { aiGatewayChatModel } from '../ai-gateway/ai-gateway'
+import { buildAiGatewayCacheHeaders } from '../ai-gateway/cache-headers'
 import {
   OPENROUTER_MEDIUM_REASONING_PROVIDER_OPTIONS,
   OPENROUTER_WEB_RESEARCH_MODEL_ID,
@@ -12,8 +13,9 @@ export const researchTopicTool = createDelegatedAgentTool({
   id: 'researchTopic',
   description:
     'Delegate a research task to a dedicated research agent that searches the web, fetches pages, and returns a synthesized markdown report. Call multiple instances in parallel for broad research across different topics.',
-  model: () => bifrostChatModel(OPENROUTER_WEB_RESEARCH_MODEL_ID),
+  model: () => aiGatewayChatModel(OPENROUTER_WEB_RESEARCH_MODEL_ID),
   providerOptions: OPENROUTER_MEDIUM_REASONING_PROVIDER_OPTIONS,
+  headers: buildAiGatewayCacheHeaders('researchTopic'),
   instructions: RESEARCHER_PROMPT,
   tools: { searchWeb: searchWebTool.create(), fetchWebpage: fetchWebpageTool.create() },
 })

package/src/tools/team-think.tool.ts

@@ -40,7 +40,6 @@ export function createTeamThinkTool(params: {
   availableUploads: ReadableUploadMetadata[]
   provideRepoTool: boolean
   defaultRepoSectionsByAgent: Record<string, DefaultRepoSections | undefined>
-  reasoningProfile: 'fast' | 'standard' | 'deep'
   systemWorkspaceDetails?: string
   getPreSeededMemoriesSection: (agentId: string) => Promise<string | undefined>
   retrievedKnowledgeSection?: string
@@ -59,7 +58,6 @@ export function createTeamThinkTool(params: {
         mode: 'fixedWorkstreamMode',
         onboardingActive: false,
         linearInstalled: false,
-        reasoningProfile: 'fast',
         systemWorkspaceDetails: runParams.systemWorkspaceDetails,
         preSeededMemoriesSection: runParams.preSeededMemoriesSection,
         retrievedKnowledgeSection: runParams.retrievedKnowledgeSection,
@@ -116,7 +114,6 @@ export function createTeamThinkTool(params: {
     latestUserMessageId: params.latestUserMessageId,
     availableUploads: params.availableUploads,
     defaultRepoSectionsByAgent: params.defaultRepoSectionsByAgent,
-    reasoningProfile: params.reasoningProfile,
     systemWorkspaceDetails: params.systemWorkspaceDetails,
     getPreSeededMemoriesSection: params.getPreSeededMemoriesSection,
     retrievedKnowledgeSection: params.retrievedKnowledgeSection,

package/src/workers/regular-chat-memory-digest.helpers.ts

@@ -2,7 +2,7 @@ import { isAgentName } from '../config/agent-defaults'
 import { compactWhitespace } from '../utils/string'
 
 interface DigestMessageForTranscript {
-  source: 'workstream'
+  source: 'workstream' | 'social'
   sourceId: string
   role: 'system' | 'user' | 'assistant'
   parts: Array<Record<string, unknown>>