@vellumai/assistant 0.8.12-dev.202606150824.4ddebe7 → 0.8.12-dev.202606151312.6aebfa4

package/openapi.yaml

@@ -8206,14 +8206,13 @@ paths:
   /v1/conversations/{id}/compaction:
     get:
       operationId: conversations_by_id_compaction_get
-      summary: Get the compaction trail leading up to an LLM call
+      summary: Get the compaction(s) attributed to an LLM call
       description:
-        Return the chronological list of compaction events that ran in the same agent turn as the call identified
-        by `callId`. Turn bounds are walked from the `messages` table (real user messages — tool-result user messages
-        are not boundaries). Served from the first-class compaction log when `compactionLogs.destination = "clickhouse"`
-        is configured, falling back to the legacy projection over `llm_request_logs` rows where `call_site =
-        "compactionAgent"`. When the conversation has no other messages around the selected call, every compaction
-        strictly before the call's `createdAt` is in scope. Drives the Inspector's Compaction tab.
+        Return the chronological list of compactions attributed to the call identified by `callId` — those that ran
+        strictly between the previous real (non-`compactionAgent`) LLM call and the selected call. Served from the
+        first-class compaction log when `compactionLogs.destination = "clickhouse"` is configured, falling back to the
+        legacy projection over `llm_request_logs` rows where `call_site = "compactionAgent"`. Usually empty or a single
+        compaction; the recovery cascade can fire several before one call lands. Drives the Inspector's Compaction tab.
       tags:
         - conversations
       responses:
@@ -8235,54 +8234,84 @@ paths:
                           type: string
                         createdAt:
                           type: number
-                        model:
+                        trigger:
                           anyOf:
                             - type: string
                             - type: "null"
-                        provider:
+                        compacted:
+                          anyOf:
+                            - type: boolean
+                            - type: "null"
+                        summaryFailed:
+                          anyOf:
+                            - type: boolean
+                            - type: "null"
+                        skipReason:
                           anyOf:
                             - type: string
                             - type: "null"
-                        inputTokens:
+                        contextTokensBefore:
                           anyOf:
                             - type: number
                             - type: "null"
-                        outputTokens:
+                        contextTokensAfter:
                           anyOf:
                             - type: number
                             - type: "null"
-                        durationMs:
+                        messagesBefore:
                           anyOf:
                             - type: number
                             - type: "null"
-                        responsePreview:
+                        messagesAfter:
                           anyOf:
-                            - type: string
+                            - type: number
                             - type: "null"
-                        requestMessageCount:
+                        compactedMessages:
                           anyOf:
                             - type: number
                             - type: "null"
-                        stopReason:
+                        preservedTailMessages:
+                          anyOf:
+                            - type: number
+                            - type: "null"
+                        durationMs:
+                          anyOf:
+                            - type: number
+                            - type: "null"
+                        summaryModel:
                           anyOf:
                             - type: string
                             - type: "null"
-                        estimatedCostUsd:
+                        summaryInputTokens:
                           anyOf:
                             - type: number
                             - type: "null"
+                        summaryOutputTokens:
+                          anyOf:
+                            - type: number
+                            - type: "null"
+                        summaryText:
+                          anyOf:
+                            - type: string
+                            - type: "null"
                       required:
                         - id
                         - createdAt
-                        - model
-                        - provider
-                        - inputTokens
-                        - outputTokens
+                        - trigger
+                        - compacted
+                        - summaryFailed
+                        - skipReason
+                        - contextTokensBefore
+                        - contextTokensAfter
+                        - messagesBefore
+                        - messagesAfter
+                        - compactedMessages
+                        - preservedTailMessages
                         - durationMs
-                        - responsePreview
-                        - requestMessageCount
-                        - stopReason
-                        - estimatedCostUsd
+                        - summaryModel
+                        - summaryInputTokens
+                        - summaryOutputTokens
+                        - summaryText
                       additionalProperties: false
                 required:
                   - conversationId
@@ -8303,7 +8332,7 @@ paths:
           required: true
           schema:
             type: string
-          description: ID of the selected LLM call from the rail. Defines the chronological cutoff for the trail.
+          description: ID of the selected LLM call from the rail. Defines the chronological cutoff for the attributed compactions.
   /v1/conversations/{id}/inference-profile:
     put:
       operationId: conversations_by_id_inferenceprofile_put

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vellumai/assistant",
-  "version": "0.8.12-dev.202606150824.4ddebe7",
+  "version": "0.8.12-dev.202606151312.6aebfa4",
   "license": "MIT",
   "type": "module",
   "exports": {

package/src/memory/compaction-log-store-clickhouse.ts

@@ -213,10 +213,12 @@ export class ClickHouseCompactionLogStore {
 
   /**
    * Read all compaction attempts whose `started_at` falls strictly inside
-   * the given window, pairing start/end rows by `compaction_id`. Mirrors
-   * the strict `>` / `<` predicate contract of
-   * `getCompactionLogsBetween` on the llm-request-log sources, so the
-   * route can apply the same ±1ms boundary shifts to both.
+   * the open window `(afterStartedAt, beforeStartedAt)`, pairing start/end
+   * rows by `compaction_id`. Mirrors the strict `>` / `<` predicate
+   * contract of `getCompactionLogsBetween` on the llm-request-log sources
+   * so the compaction route can scope both stores to the same call window
+   * (floor = previous real call's `createdAt`, ceiling = selected call's
+   * `createdAt`).
    */
   async getEventsBetween(
     conversationId: string,

package/src/memory/llm-request-log-source-clickhouse.ts

@@ -256,6 +256,34 @@ export class ClickHouseLlmRequestLogSource implements LlmRequestLogSource {
     return rows.map((r) => this.toLogRow(r));
   }
 
+  async getPreviousNonCompactionCallCreatedAt(
+    conversationId: string,
+    beforeCreatedAt: number,
+  ): Promise<number | null> {
+    const aid = await this.assistantId();
+    // `call_site != 'compactionAgent'` keeps the pre-migration-264 rows
+    // (stored as the empty-string default) in scope as real calls and
+    // excludes only the summarizer's own logs, matching the local
+    // store's `isNull(callSite) OR callSite != 'compactionAgent'`.
+    const sql = `SELECT
+        toUnixTimestamp64Milli(created_at) AS created_at
+      FROM ${this.tableRef()}
+      WHERE assistant_id = {assistant_id:String}
+        AND conversation_id = {conversation_id:String}
+        AND call_site != {call_site:String}
+        AND created_at < fromUnixTimestamp64Milli({before_created_at:Int64})
+      ORDER BY created_at DESC, id DESC
+      LIMIT 1
+      FORMAT JSONEachRow`;
+    const rows = await this.exec<{ created_at: string }>(sql, {
+      assistant_id: aid,
+      conversation_id: conversationId,
+      call_site: "compactionAgent",
+      before_created_at: String(beforeCreatedAt),
+    });
+    return rows[0] ? Number(rows[0].created_at) : null;
+  }
+
   async getCompactionLogsBetween(
     conversationId: string,
     afterCreatedAt: number | null,

package/src/memory/llm-request-log-source-local.ts

@@ -12,6 +12,7 @@ import type { LlmRequestLogSource } from "./llm-request-log-source.js";
 import {
   type CompactionAgentLogRow,
   getCompactionLogsBetween,
+  getPreviousNonCompactionCallCreatedAt,
   getRequestLogById,
   getRequestLogMetaById,
   getRequestLogsByConversationId,
@@ -39,6 +40,16 @@ export class LocalLlmRequestLogSource implements LlmRequestLogSource {
     return getRequestLogsByConversationId(conversationId);
   }
 
+  async getPreviousNonCompactionCallCreatedAt(
+    conversationId: string,
+    beforeCreatedAt: number,
+  ): Promise<number | null> {
+    return getPreviousNonCompactionCallCreatedAt(
+      conversationId,
+      beforeCreatedAt,
+    );
+  }
+
   async getCompactionLogsBetween(
     conversationId: string,
     afterCreatedAt: number | null,

package/src/memory/llm-request-log-source.ts

@@ -51,15 +51,32 @@ export interface LlmRequestLogSource {
    */
   getRequestLogsByConversationId(conversationId: string): Promise<LogRow[]>;
 
+  /**
+   * Return the `createdAt` of the most recent **non-`compactionAgent`**
+   * LLM call in the conversation that ran strictly before
+   * `beforeCreatedAt`, or `null` when the selected call is the first real
+   * call in the conversation.
+   *
+   * Drives the call-scoped floor for the Inspector's Compaction tab: a
+   * compaction is attributed to the next real call that ran after it, so
+   * the compactions for a given call are those that landed strictly
+   * between the previous real call and the selected call.
+   */
+  getPreviousNonCompactionCallCreatedAt(
+    conversationId: string,
+    beforeCreatedAt: number,
+  ): Promise<number | null>;
+
   /**
    * Fetch every `callSite = "compactionAgent"` log row in the conversation
    * whose `createdAt` falls in the **open window**
    * `(afterCreatedAt, beforeCreatedAt)`, ordered chronologically.
    *
-   * Drives the Inspector's Compaction tab. The route handler computes
-   * the window from the turn the selected call belongs to via
-   * `getTurnTimeBounds` in `memory/conversation-crud.ts`; this source
-   * method is bound-agnostic. See `getCompactionLogsBetween` in
+   * The legacy fallback for the Inspector's Compaction tab, used when the
+   * ClickHouse compaction log is unavailable. The route handler resolves
+   * the floor from `getPreviousNonCompactionCallCreatedAt` and the ceiling
+   * from the selected call's `createdAt`; this source method is
+   * bound-agnostic. See `getCompactionLogsBetween` in
    * `llm-request-log-store.ts` for the SQL.
    */
   getCompactionLogsBetween(

package/src/memory/llm-request-log-store.ts

@@ -9,6 +9,8 @@ import {
   isNull,
   lt,
   lte,
+  ne,
+  or,
   sql,
 } from "drizzle-orm";
 import { v4 as uuid } from "uuid";
@@ -413,6 +415,43 @@ function selectUnlinkedLogsInRange(
     .all();
 }
 
+/**
+ * Return the `createdAt` of the most recent **non-`compactionAgent`** LLM
+ * call in the conversation that ran strictly before `beforeCreatedAt`, or
+ * `null` when no such call exists (the selected call is the first real
+ * call in the conversation).
+ *
+ * Drives the call-scoped floor for the Inspector's Compaction tab. A
+ * compaction is attributed to the next real call that runs after it, so
+ * the trail for a given call is the set of compactions that landed
+ * strictly between the previous real call and the selected call. NULL
+ * `callSite` rows (pre-migration-264) are treated as real calls — only
+ * `compactionAgent` rows are excluded.
+ */
+export function getPreviousNonCompactionCallCreatedAt(
+  conversationId: string,
+  beforeCreatedAt: number,
+): number | null {
+  const db = getDb();
+  const row = db
+    .select({ createdAt: llmRequestLogs.createdAt })
+    .from(llmRequestLogs)
+    .where(
+      and(
+        eq(llmRequestLogs.conversationId, conversationId),
+        lt(llmRequestLogs.createdAt, beforeCreatedAt),
+        or(
+          isNull(llmRequestLogs.callSite),
+          ne(llmRequestLogs.callSite, "compactionAgent"),
+        ),
+      ),
+    )
+    .orderBy(desc(llmRequestLogs.createdAt), desc(llmRequestLogs.id))
+    .limit(1)
+    .get();
+  return row?.createdAt ?? null;
+}
+
 /**
  * Fetch every `compactionAgent` log row in the conversation whose
  * `createdAt` falls in the **open window** `(afterCreatedAt, beforeCreatedAt)`,