Workspaces lakehouse

The lakehouse collects usage information, document audit trails, Logic Pilot traces, and feedback from every workspace. This page shows where that data is saved, and which program reads it.

Follow a batch of data from the workspaces to the place that keeps it. Choose a type of data. The glowing cubes show the route.

Step 1 of 4

About 7 seconds after a burst of changes, each workspace sends them to the worker.

Document audit

Document changes are sent when events happen. About 7 seconds after a burst of changes, the workspace sends them to the Cloudflare worker. A poll every 5 minutes is a safety net, in case an event is missed. There is no D1 for this save.

The worker sends the batch into Cloudflare Pipelines. About a minute later the rows land in Iceberg document_audit.v1, in the R2 bucket workspace-events. That send also clears the KV cache for the document.

Saved in the lakehouse

• Document auditIceberg
• Document telemetryD1 and Iceberg
• Logic Pilot tracesD1 and Iceberg
• Workspace statsD1
• Logic Pilot statsD1
• Bug reportsD1
• Workspace feedbackD1
• Logic Pilot feedbackD1
• File storageD1

Terms used on this page

Programs and places

Term	Meaning
Workspace	One customer system. It has its own database.
Workspace backend	The program that sends data from each workspace. See logic-bee.
Worker	The program that receives the data and saves it. See bob-workspace-orchestration.
Builder	The program that reads the lakehouse data for charts. See bob-api.
D1	The hot store. A fast database. It answers queries about recent data.
Iceberg	The cold store. Files kept in the R2 bucket workspace-events. It keeps the long history.
R2	Cloudflare's file storage. It holds the Iceberg files and the workspace file buckets.
Cloudflare Pipelines	The Cloudflare service that writes the Iceberg files. Rows are ready about 1 minute after the worker sends them.

Lakehouse concepts

Term	Meaning
Hot store	A store built for fast, recent queries. On this page, that is D1.
Cold store	A store built to hold the long history. On this page, that is Iceberg.
Cleanup sweep	The job that runs every 6 hours and deletes D1 rows older than 28 days. Iceberg keeps its own copy.
Deployment	Another name for one workspace. Its id is the workspace id.
Company	One company inside a workspace. A workspace can hold more than one.
Business unit	A group inside a company.

Logic Pilot terms

Term	Meaning
Logic Pilot	The platform's AI agent. People talk with it in conversations, and each run can be traced as spans.
Conversation	One Logic Pilot chat thread. It has a mode, a model, and a message count.
Session	The document that holds a person's ratings on Logic Pilot messages, identified by sessionDocId.
Trace	One Logic Pilot run, from start to end.
Span	One step inside a trace, for example one model call.
Root span	The first span in a trace.

Other terms

Term	Meaning
Burst	A short period of rapid changes to one document.
Bucket	One Cloudflare R2 container for files. Each workspace has a public (cdn) bucket and a private (storage) bucket.
Class A / Class B operations	Cloudflare's two cost categories for R2 operations. Class A covers writes and lists. Class B covers reads.
JSON Patch	The format used in jsonPatch to describe one document change, as a list of operations.

The three programs

Three programs move this data.

Program	What it does	Repository
Workspace backend	Sends data from each workspace	logic-bee
Worker	Receives the data and saves it	bob-workspace-orchestration
Builder backend	Reads the data for charts	bob-api

A workspace is one customer system. It has its own database. There are many workspaces.

The workspace id in this store is the deployment id.

Local workspaces

A workspace on a developer machine does not push lakehouse data. NODE_ENV is not production, so the workspace returns before it calls the worker.

The scheduled sends still start. They stop before they read companies or open a request. A document change, a document count, a Logic Pilot span, a stats snapshot, a bug report, a feedback note, and a Logic Pilot rating stay on the local machine. They are not copied to D1 or Iceberg. An audit-trail read on that machine does not call the worker.

A production workspace does push this data. NODE_ENV is production.

File storage is separate. The worker collects those totals. The workspace does not send them in either environment.

What the lakehouse is

The lakehouse is one shared store. The business documents stay in the workspace. The lakehouse keeps copies of history, counts, reports, and file totals.

Two places save data:

• D1 is a database. It answers recent queries.
• Iceberg is files in the R2 bucket workspace-events. It keeps the long history.

Some types go to both places. Then D1 is the hot store. Iceberg is the long copy. D1 keeps that copy for 28 days. A query about those 28 days reads D1. An older query reads Iceberg. Every 6 hours, the worker deletes D1 rows that are older than 28 days.

Cloudflare Pipelines writes the Iceberg files. The rows are ready after about 1 minute.

Document telemetry and Logic Pilot traces use both places. Document audit uses Iceberg only. The other types use D1 only.

Fields in every row

Field	Meaning
eventId	The unique id of this row.
workspaceId	The id of the workspace that sent the row. This is the deployment id.
occurredAt	The time the event happened.

Most rows also carry companyId, the id of the company inside the workspace. Rows about one person also carry userId, and some also carry userEmail. Each data-type section below lists every field from the worker schema for that row.

See the route

This chart shows the same route, one type at a time. Choose a type. Drag the chart to move it. Use the controls to zoom.

Loading this route.

Document audit goes to Iceberg document_audit.v1. The workspace reads it back for that document. This save has no D1.

Document audit

Document audit is the history of one document. It records who changed the document, and what changed.

The workspace sends the changes about 7 seconds after a burst of changes stops. A check every 5 minutes sends a change that the first listen missed.

The worker saves the row in Iceberg document_audit.v1 only.

The workspace reads this history when a person opens the audit trail or the activity of that document. Builder does not chart this type.

Example row (click to expand)

This is the full row from the worker schema.

{
  "workspaceId": "d00232",
  "eventId": "evt_audit_001",
  "occurredAt": "2026-10-07T08:15:00.000Z",
  "auditId": "audit_001",
  "timestamp": "2026-10-07T08:15:00.000Z",
  "documentTimestamp": "2026-10-07T08:14:58.000Z",
  "serviceName": "logic-bee",
  "type": "document",
  "appId": "app_1",
  "appSlug": "finance",
  "appType": "workspace",
  "projectId": "proj_1",
  "deploymentId": "d00232",
  "bobPackage": "finance",
  "bobPackageVersion": "1.0.0",
  "environment": "production",
  "deployment": "d00232",
  "licenseId": "lic_1",
  "userId": "user_1",
  "userEmail": "ada@example.com",
  "roleType": "admin",
  "userRoleId": "role_1",
  "isSuperAdmin": "false",
  "userType": "user",
  "companyId": "co_1",
  "businessUnitId": "bu_1",
  "sessionId": "sess_1",
  "ipAddress": "203.0.113.10",
  "requestType": "http",
  "userAgent": "Mozilla/5.0",
  "complianceFlags": "[]",
  "retentionPolicy": "default",
  "retentionYears": 7,
  "dataClassification": "internal",
  "docId": "inv_1001",
  "mongoId": "6640a1",
  "docType": "invoice",
  "namespace": "finance",
  "docName": "invoice",
  "cfpPath": "finance/invoice",
  "status": "paid",
  "version": 3,
  "operationType": "update",
  "operation": "update",
  "jsonPatch": "[{\"op\":\"replace\",\"path\":\"/data/status\",\"value\":\"paid\"}]",
  "patchOperationsCount": 1,
  "fieldsAffected": "[\"/data/status\"]",
  "updatedFields": "[\"/data/status\"]",
  "removedFields": "[]",
  "truncatedArrays": "[]",
  "data": "{}",
  "previousDocument": "{\"status\":\"unpaid\"}",
  "currentDocument": "{\"status\":\"paid\"}",
  "currentDocumentHash": "abc123",
  "previousDocumentHash": "def456",
  "message": "Status changed from unpaid to paid",
  "activityMessage": "Ada marked the invoice paid",
  "isActivity": true,
  "userRequestId": "req_1",
  "transactionId": "txn_1",
  "podName": "logic-bee-1"
}

View all fields

Field	Meaning
workspaceId	The id of the workspace that sent the row. This is the deployment id.
eventId	The unique id of this row.
occurredAt	The time the event happened.
auditId	The id of the audit record. May be empty.
timestamp	The time the audit record was written.
documentTimestamp	The time stored on the document.
serviceName	The service that wrote the change.
type	The audit type.
appId	The id of the app.
appSlug	The slug of the app.
appType	The type of the app.
projectId	The id of the project.
deploymentId	The id of the deployment.
bobPackage	The package name.
bobPackageVersion	The package version.
environment	The environment, for example production.
deployment	The deployment name.
licenseId	The license id.
userId	The id of the person who made the change.
userEmail	The email of that person.
roleType	The role type of that person.
userRoleId	The id of that person's role.
isSuperAdmin	Whether that person is a super admin, as a string.
userType	The kind of user.
companyId	The id of the company inside the workspace.
businessUnitId	The id of the business unit.
sessionId	The id of the session.
ipAddress	The IP address of the request.
requestType	The kind of request.
userAgent	The user agent of the request.
complianceFlags	A JSON string of compliance flags.
retentionPolicy	The retention policy name.
retentionYears	How many years the row is kept.
dataClassification	The data classification label.
docId	The id of the document that changed.
mongoId	The database id of the document.
docType	The type of the document.
namespace	The namespace of the document.
docName	The name of the document, for display.
cfpPath	The CFP path of the document.
status	The document status after the change.
version	The document version after the change.
operationType	The kind of operation.
operation	The kind of change: create, update, or delete.
jsonPatch	The exact change, as a JSON Patch array stored in a string.
patchOperationsCount	The number of operations in jsonPatch.
fieldsAffected	A JSON string of the paths the change touched.
updatedFields	A JSON string of the paths that were updated.
removedFields	A JSON string of the paths that were removed.
truncatedArrays	A JSON string of arrays that were truncated.
data	A JSON string of the change data.
previousDocument	A JSON string of the document before the change.
currentDocument	A JSON string of the document after the change.
currentDocumentHash	A hash of the document after the change.
previousDocumentHash	A hash of the document before the change.
message	A short, human-readable summary of the change.
activityMessage	The text shown in the activity list.
isActivity	true when this row is also an activity item.
userRequestId	The id of the user request.
transactionId	The id of the transaction.
podName	The name of the pod that wrote the change.

If a copy does not arrive

The 5-minute check sends an audit change that the first listen missed. After the workspace prepares the history row, it does not send that change again.

Document telemetry

Document telemetry is a count of documents. It is not the documents.

Twice a day, each workspace sends the counts. Each row is one collection. count is the number of documents. countDeleted is the number of deleted documents.

The worker saves the same row in two places:

• D1 document_telemetry, table v1, for 28 days
• Iceberg document_telemetry.v1

Builder reads these counts for the document charts. The overview of one deployment uses them too.

Example row (click to expand)

This is the full row from the worker schema.

{
  "workspaceId": "d00232",
  "eventId": "evt_tel_001",
  "occurredAt": "2026-10-07T08:18:00.000Z",
  "auditId": "",
  "timestamp": "2026-10-07T08:18:00.000Z",
  "serviceName": "logic-bee",
  "type": "document-telemetry",
  "appId": "app_1",
  "appSlug": "finance",
  "appType": "workspace",
  "projectId": "proj_1",
  "deploymentId": "d00232",
  "bobPackage": "finance",
  "bobPackageVersion": "1.0.0",
  "environment": "production",
  "deployment": "d00232",
  "licenseId": "lic_1",
  "userId": "user_1",
  "userEmail": "ada@example.com",
  "roleType": "admin",
  "userRoleId": "role_1",
  "isSuperAdmin": "false",
  "userType": "user",
  "companyId": "co_1",
  "businessUnitId": "bu_1",
  "sessionId": "sess_1",
  "ipAddress": "203.0.113.10",
  "requestType": "cron",
  "userAgent": "",
  "complianceFlags": "[]",
  "retentionPolicy": "default",
  "retentionYears": 7,
  "dataClassification": "internal",
  "collectionName": "finance",
  "docType": "invoice",
  "namespace": "finance",
  "docName": "invoice",
  "cfpPath": "finance/invoice",
  "count": 1204,
  "countDeleted": 16,
  "version": 1
}

View all fields

Field	Meaning
workspaceId	The id of the workspace that sent the row. This is the deployment id.
eventId	The unique id of this row.
occurredAt	The time the count was taken.
auditId	May be empty.
timestamp	The time the row was written.
serviceName	The service that sent the count.
type	The telemetry type.
appId	The id of the app.
appSlug	The slug of the app.
appType	The type of the app.
projectId	The id of the project.
deploymentId	The id of the deployment.
bobPackage	The package name.
bobPackageVersion	The package version.
environment	The environment, for example production.
deployment	The deployment name.
licenseId	The license id.
userId	The id of the person, when the row has one.
userEmail	The email of that person.
roleType	The role type.
userRoleId	The id of the role.
isSuperAdmin	Whether the person is a super admin, as a string.
userType	The kind of user.
companyId	The id of the company inside the workspace.
businessUnitId	The id of the business unit.
sessionId	The id of the session.
ipAddress	The IP address, when the row has one.
requestType	The kind of request.
userAgent	The user agent, when the row has one.
complianceFlags	A JSON string of compliance flags.
retentionPolicy	The retention policy name.
retentionYears	How many years the row is kept.
dataClassification	The data classification label.
collectionName	The collection the count is for.
docType	The type of document counted.
namespace	The namespace of the collection.
docName	The display name for the document type.
cfpPath	The CFP path of the document type.
count	The number of documents in the collection.
countDeleted	The number of deleted documents in the collection.
version	The version of this count row.

If a copy does not arrive

The next twice-daily send sends a new snapshot.

Logic Pilot traces

A trace is one Logic Pilot run. A span is one step in that run.

The workspace waits until 512 spans are ready, or until 5 seconds pass. Then it sends the group. One group can contain spans from more than one run.

The worker saves the same row in two places:

• D1 mastra_spans, table v2, for 28 days
• Iceberg mastra_spans.v2

Builder reads recent lists from D1. Builder reads one trace from Iceberg. A short cache holds the latest trace. A new save clears that cache.

Example row (click to expand)

This is the full row from the worker schema. endedAt is left out when the span has no end time. Do not send an empty string for it.

{
  "workspaceId": "d00232",
  "eventId": "evt_span_001",
  "occurredAt": "2026-10-07T09:01:05.000Z",
  "startAt": "2026-10-07T09:01:04.160Z",
  "endedAt": "2026-10-07T09:01:05.000Z",
  "traceId": "trace_88",
  "spanId": "span_3",
  "parentSpanId": "span_1",
  "spanType": "model",
  "name": "answer",
  "model": "gpt-4.1",
  "provider": "openai",
  "durationMs": 840,
  "promptTokens": 120,
  "completionTokens": 40,
  "status": "ok",
  "companyId": "co_1",
  "userId": "user_1",
  "environment": "production",
  "conversationId": "conv_9",
  "messageId": "msg_9",
  "mode": "text",
  "inputJson": "{}",
  "outputJson": "{}",
  "attributesJson": "{}",
  "metadataJson": "{}",
  "errorJson": "",
  "entityType": "conversation",
  "entityId": "conv_9",
  "entityName": "Invoice help",
  "isEvent": false,
  "isRootSpan": false,
  "runId": "run_12",
  "tags": "[]"
}

View all fields

Field	Meaning
workspaceId	The id of the workspace that sent the row. This is the deployment id.
eventId	The unique id of this row.
occurredAt	The time the span was recorded.
startAt	The time the span started.
endedAt	The time the span ended. Omitted when there is no end time.
traceId	The id of the run this span belongs to.
spanId	The id of this step.
parentSpanId	The id of the parent span. May be empty.
spanType	The kind of step, for example model.
name	A short label for the step.
model	The AI model used in this step. May be empty.
provider	The provider of that model, for example openai. May be empty.
durationMs	How long the step took, in milliseconds.
promptTokens	Input tokens used by this step.
completionTokens	Output tokens used by this step.
status	ok, or an error state. May be empty.
companyId	The id of the company. May be empty.
userId	The id of the person. May be empty.
environment	The environment. May be empty.
conversationId	The Logic Pilot conversation. May be empty.
messageId	The message this span belongs to. May be empty.
mode	The conversation mode. May be empty.
inputJson	A JSON string of the span input.
outputJson	A JSON string of the span output.
attributesJson	A JSON string of span attributes.
metadataJson	A JSON string of span metadata.
errorJson	A JSON string of the error. Empty when there is no error.
entityType	The kind of entity this span is about. May be empty.
entityId	The id of that entity. May be empty.
entityName	The name of that entity. May be empty.
isEvent	true when this span is an event.
isRootSpan	true for the first span in the run.
runId	The id of the broader run that contains this trace. May be empty.
tags	A JSON string of tags.

If a copy does not arrive

A span group that fails to leave the workspace is dropped. Later spans go out in a new group.

Workspace stats

Workspace stats is a count of one company at one time. The count includes people, roles, companies, business units, invitations, API keys, two-factor use, and the autonomy level.

Twice a day, each workspace sends one row for each company. The worker saves it in D1 workspace_stats, table workspace_stats. If the same row is already saved, the worker leaves it.

Builder reads these rows to show change over time, and to show the latest counts.

Example row (click to expand)

This is the full row from the worker schema. eventId is workspaceId, companyId, and occurredAt.

{
  "eventId": "d00232:co_1:2026-10-07T07:26:00.000Z",
  "workspaceId": "d00232",
  "occurredAt": "2026-10-07T07:26:00.000Z",
  "companyId": "co_1",
  "users": 42,
  "deletedUsers": 1,
  "guests": 3,
  "deletedGuests": 0,
  "roles": 8,
  "deletedRoles": 0,
  "companies": 1,
  "deletedCompanies": 0,
  "businessUnits": 3,
  "superAdminUsers": 1,
  "adminUsers": 4,
  "otherUsers": 37,
  "apiKeys": 2,
  "deletedApiKeys": 0,
  "apiCredentials": 1,
  "invitesJson": "{\"pending\":2}",
  "guestInvitesJson": "{}",
  "pendingInvites": 2,
  "oldestPendingInviteAt": "2026-10-01T09:00:00.000Z",
  "invitesTotal": 10,
  "twoFactorRequired": 1,
  "usersWith2fa": 30,
  "rolesInUse": 6,
  "autonomyLevel": 2
}

View all fields

Field	Meaning
eventId	workspaceId, companyId, and occurredAt. A repeat of the same row is left as it is.
workspaceId	The id of the workspace that sent the row. This is the deployment id.
occurredAt	The time the snapshot was taken.
companyId	The id of the company. May be empty.
users	The number of people in the company.
deletedUsers	The number of deleted people.
guests	The number of guests.
deletedGuests	The number of deleted guests.
roles	The number of roles defined.
deletedRoles	The number of deleted roles.
companies	The number of companies in the workspace.
deletedCompanies	The number of deleted companies.
businessUnits	The number of business units.
superAdminUsers	The number of super admins.
adminUsers	The number of admins.
otherUsers	The number of people who are not admins.
apiKeys	The number of API keys.
deletedApiKeys	The number of deleted API keys.
apiCredentials	The number of API credentials.
invitesJson	A JSON string of invite counts by status.
guestInvitesJson	A JSON string of guest invite counts by status.
pendingInvites	The number of invitations not yet accepted.
oldestPendingInviteAt	The time of the oldest pending invitation. May be empty.
invitesTotal	The total number of invitations.
twoFactorRequired	1 if two-factor login is required, 0 if not.
usersWith2fa	The number of people with two-factor login on.
rolesInUse	The number of roles that are assigned.
autonomyLevel	The autonomy level set for the company.

If a copy does not arrive

The next twice-daily send sends a new row.

Logic Pilot stats

Logic Pilot stats is three counts. The workspace sends them twice a day.

1. One row for each conversation that changed in the last 72 hours.
2. One row of folder counts for each company.
3. One row of token totals for each model.

The 72 hours are longer than the time between the two daily sends. If one send fails, the next send still includes that conversation.

The worker saves the rows in D1 logic_pilot_stats. The tables are logic_pilot_conversation_stat, logic_pilot_folder_stat, and token_telemetry_stat.

Builder reads them for the conversation charts and the token charts.

Conversation rows

One row for each conversation that changed in the last 72 hours.

Example row (click to expand)

This is the full row from the worker schema. eventId is workspaceId, conversationId, and updatedAt. hasContext, isBookmarked, and inFolder are stored as true or false.

{
  "eventId": "d00232:conv_9:2026-10-06T18:00:00.000Z",
  "workspaceId": "d00232",
  "occurredAt": "2026-10-07T10:43:00.000Z",
  "updatedAt": "2026-10-06T18:00:00.000Z",
  "createdAt": "2026-10-01T09:00:00.000Z",
  "lastMessageAt": "2026-10-06T18:00:00.000Z",
  "firstMessageAt": "2026-10-01T09:05:00.000Z",
  "durationMs": 432000000,
  "companyId": "co_1",
  "userId": "user_1",
  "roleType": "admin",
  "conversationId": "conv_9",
  "name": "Invoice help",
  "nameLength": 12,
  "status": "active",
  "isBookmarked": false,
  "inFolder": true,
  "conversationType": "chat",
  "mode": "text",
  "modelId": "gpt-4.1",
  "agentId": "agent_1",
  "hasContext": true,
  "contextItemCount": 2,
  "eventsCount": 3,
  "messageCount": 14,
  "userMessageCount": 7,
  "assistantMessageCount": 7,
  "totalUserCharCount": 900,
  "totalUserWordCount": 160,
  "maxUserWordCount": 40,
  "voiceMessageCount": 0,
  "totalVoiceDurationMs": 0,
  "avgVoiceDurationMs": 0,
  "maxVoiceDurationMs": 0,
  "documentCount": 1,
  "spreadsheetCount": 0,
  "chartCount": 0
}

View all fields

Field	Meaning
eventId	workspaceId, conversationId, and updatedAt. A repeat send is ignored.
workspaceId	The id of the workspace that sent the row. This is the deployment id.
occurredAt	The time the snapshot was taken.
updatedAt	The time the conversation was last updated. May be empty.
createdAt	The time the conversation was created. May be empty.
lastMessageAt	The time of the last message. May be empty.
firstMessageAt	The time of the first message. May be empty.
durationMs	The length of the conversation, in milliseconds.
companyId	The id of the company. May be empty.
userId	The id of the person. May be empty.
roleType	The role type of that person. May be empty.
conversationId	The id of the conversation.
name	The conversation name. May be empty.
nameLength	The length of the name.
status	active, or another conversation state. May be empty.
isBookmarked	true when the conversation is bookmarked.
inFolder	true when the conversation is in a folder.
conversationType	The kind of conversation. May be empty.
mode	text, or another conversation mode. May be empty.
modelId	The AI model used in the conversation. May be empty.
agentId	The id of the agent. May be empty.
hasContext	true when the conversation has context items.
contextItemCount	The number of context items.
eventsCount	The number of events.
messageCount	The number of messages in the conversation.
userMessageCount	The number of messages from the person.
assistantMessageCount	The number of messages from the assistant.
totalUserCharCount	The total characters in the person's messages.
totalUserWordCount	The total words in the person's messages.
maxUserWordCount	The longest person message, in words.
voiceMessageCount	The number of voice messages.
totalVoiceDurationMs	The total voice length, in milliseconds.
avgVoiceDurationMs	The average voice length, in milliseconds.
maxVoiceDurationMs	The longest voice message, in milliseconds.
documentCount	The number of documents produced.
spreadsheetCount	The number of spreadsheets produced.
chartCount	The number of charts produced.

Folder rows

One row of folder counts for each company.

Example row (click to expand)

This is the full row from the worker schema. eventId is workspaceId, companyId, and occurredAt. One row is for one company, not for one folder.

{
  "eventId": "d00232:co_1:2026-10-07T10:43:00.000Z",
  "workspaceId": "d00232",
  "occurredAt": "2026-10-07T10:43:00.000Z",
  "companyId": "co_1",
  "folderCount": 6,
  "emptyFolderCount": 1,
  "usersWithFoldersCount": 4,
  "maxFoldersPerUser": 3,
  "avgFoldersPerUser": 1.5,
  "conversationsInFoldersCount": 20,
  "conversationsNotInFolderCount": 5,
  "maxConversationsInFolder": 8,
  "avgConversationsPerFolder": 3.3
}

View all fields

Field	Meaning
eventId	workspaceId, companyId, and occurredAt.
workspaceId	The id of the workspace that sent the row. This is the deployment id.
occurredAt	The time the snapshot was taken.
companyId	The id of the company. May be empty.
folderCount	The number of folders in the company.
emptyFolderCount	The number of folders with no conversations.
usersWithFoldersCount	The number of people who have at least one folder.
maxFoldersPerUser	The most folders one person has.
avgFoldersPerUser	The average number of folders per person.
conversationsInFoldersCount	The number of conversations placed in a folder.
conversationsNotInFolderCount	The number of conversations that are not in a folder.
maxConversationsInFolder	The most conversations in one folder.
avgConversationsPerFolder	The average number of conversations per folder.

Token rows

One row of token totals for each model. These numbers are totals since the counter started, not just for that day.

Example row (click to expand)

This is the full row from the worker schema.

{
  "eventId": "d00232:co_1:gpt-4.1:2026-10-07T10:43:00.000Z",
  "workspaceId": "d00232",
  "occurredAt": "2026-10-07T10:43:00.000Z",
  "companyId": "co_1",
  "model": "gpt-4.1",
  "input": 120000,
  "output": 40000,
  "cachedInput": 8000,
  "total": 160000
}

View all fields

Field	Meaning
eventId	The unique id of this row.
workspaceId	The id of the workspace that sent the row. This is the deployment id.
occurredAt	The time the snapshot was taken.
companyId	The id of the company. May be empty.
model	The AI model these totals are for.
input	Total input tokens sent to the model.
output	Total output tokens returned by the model.
cachedInput	Total input tokens served from cache.
total	input plus output.

If a copy does not arrive

The next twice-daily send sends a new snapshot. Conversation rows still cover the last 72 hours, so a missed send is not lost.

Bug reports

A person files a bug in the workspace. The workspace saves the bug first. Then it sends a copy to the worker.

The worker saves the copy in D1 feedbacks, table bug. Builder reads the copy.

Example row (click to expand)

This is the full row from the worker schema. A repeat uses bugDocId.

{
  "eventId": "evt_bug_001",
  "workspaceId": "d00232",
  "occurredAt": "2026-10-07T11:02:00.000Z",
  "environment": "production",
  "deploymentId": "d00232",
  "licenseId": "lic_1",
  "companyId": "co_1",
  "companyName": "Northwind",
  "userId": "user_1",
  "userEmail": "ada@example.com",
  "userName": "Ada",
  "roleType": "admin",
  "bugDocId": "bug_44",
  "pageUrl": "https://workspace.example/invoices",
  "screenshotUrl": "https://workspace.example/files/bug_44.png",
  "reportType": "something-doesnt-work",
  "description": "The invoice total does not update after a line change.",
  "userTimezone": "Europe/Chisinau",
  "date": "2026-10-07",
  "userLanguage": "en"
}

View all fields

Field	Meaning
eventId	The unique id of this row.
workspaceId	The id of the workspace that sent the row. This is the deployment id.
occurredAt	The time the bug was filed.
environment	The environment, for example production.
deploymentId	The id of the deployment.
licenseId	The license id.
companyId	The id of the company.
companyName	The name of the company. May be empty.
userId	The id of the person who filed the bug.
userEmail	The email of that person.
userName	The name of that person. May be empty.
roleType	The role type of that person. May be empty.
bugDocId	The id of the bug document in the workspace. A repeat of the same bug uses this id.
pageUrl	The page the person was on when they filed the bug. May be empty.
screenshotUrl	The URL of the screenshot. May be empty.
reportType	The kind of problem, for example something-doesnt-work. May be empty.
description	The person's own description of the problem. May be empty.
userTimezone	The person's time zone. May be empty.
date	The date the person entered. May be empty.
userLanguage	The person's language. May be empty.

If a copy does not arrive

The bug report stays in the workspace. Only the Builder copy is missing.

Workspace feedback

A person sends feedback in the workspace. The workspace saves it first. Then it sends a copy to the worker.

The worker saves the copy in D1 feedbacks, table workspace_feedback. Builder reads the copy.

Example row (click to expand)

This is the full row from the worker schema. A repeat uses feedbackDocId.

{
  "eventId": "evt_fb_001",
  "workspaceId": "d00232",
  "occurredAt": "2026-10-07T11:10:00.000Z",
  "environment": "production",
  "deploymentId": "d00232",
  "licenseId": "lic_1",
  "companyId": "co_1",
  "companyName": "Northwind",
  "userId": "user_1",
  "userEmail": "ada@example.com",
  "userName": "Ada",
  "roleType": "admin",
  "feedbackDocId": "fb_12",
  "workspaceUrl": "https://workspace.example",
  "description": "The home page loads slowly in the morning.",
  "date": "2026-10-07",
  "userLanguage": "en",
  "userTimezone": "Europe/Chisinau"
}

View all fields

Field	Meaning
eventId	The unique id of this row.
workspaceId	The id of the workspace that sent the row. This is the deployment id.
occurredAt	The time the feedback was sent.
environment	The environment, for example production.
deploymentId	The id of the deployment.
licenseId	The license id.
companyId	The id of the company.
companyName	The name of the company. May be empty.
userId	The id of the person who sent the feedback.
userEmail	The email of that person.
userName	The name of that person. May be empty.
roleType	The role type of that person. May be empty.
feedbackDocId	The id of the feedback document in the workspace. A repeat uses this id.
workspaceUrl	The workspace the feedback came from. May be empty.
description	The person's own description. May be empty.
date	The date the person entered. May be empty.
userLanguage	The person's language. May be empty.
userTimezone	The person's time zone. May be empty.

If a copy does not arrive

The feedback stays in the workspace. Only the Builder copy is missing.

Logic Pilot feedback

A person rates a Logic Pilot message. The rating stays on the session. When the session is saved, the workspace sends only new ratings.

The worker saves each new rating in D1 feedbacks, table logic_pilot_message. Builder reads the copy.

Example row (click to expand)

This is the full row from the worker schema. A repeat uses feedbackId. polarity is positive or negative.

{
  "eventId": "evt_lp_001",
  "workspaceId": "d00232",
  "occurredAt": "2026-10-07T11:20:00.000Z",
  "environment": "production",
  "deploymentId": "d00232",
  "licenseId": "lic_1",
  "companyId": "co_1",
  "companyName": "Northwind",
  "userId": "user_1",
  "userEmail": "ada@example.com",
  "userName": "Ada",
  "roleType": "admin",
  "feedbackId": "rate_7",
  "polarity": "negative",
  "typeOfIssue": "wrong_data",
  "details": "The answer used the wrong invoice.",
  "sessionDocId": "sess_3",
  "sessionName": "Invoice help",
  "messageId": "msg_9",
  "agentId": "agent_1",
  "mode": "text",
  "modelId": "gpt-4.1",
  "messageContents": "[]",
  "assistantVisualContents": "[]",
  "conversationId": "conv_9"
}

View all fields

Field	Meaning
eventId	The unique id of this row.
workspaceId	The id of the workspace that sent the row. This is the deployment id.
occurredAt	The time the rating was saved.
environment	The environment, for example production.
deploymentId	The id of the deployment.
licenseId	The license id.
companyId	The id of the company.
companyName	The name of the company. May be empty.
userId	The id of the person who rated the message.
userEmail	The email of that person.
userName	The name of that person. May be empty.
roleType	The role type of that person. May be empty.
feedbackId	The id of this rating. A repeat uses this id.
polarity	positive or negative.
typeOfIssue	The kind of problem, for example wrong_data. May be empty.
details	The person's own note about the rating. May be empty.
sessionDocId	The id of the session document.
sessionName	The name of the session. May be empty.
messageId	The id of the rated message.
agentId	The id of the agent. May be empty.
mode	The conversation mode. May be empty.
modelId	The AI model that produced the rated message. May be empty.
messageContents	A JSON string of the message contents.
assistantVisualContents	A JSON string of the assistant's visual contents.
conversationId	The id of the conversation. May be empty.

If a copy does not arrive

The rating stays on the session. Only the Builder copy is missing.

File storage

File storage is the size and the use of the workspace file buckets. The worker collects this data. The workspace does not send it.

Once a day, the worker asks Builder for deployments with status started or terminated. Then the worker reads the previous day's totals from Cloudflare. Each workspace has two buckets:

• {workspaceId}-cdn is the public bucket
• {workspaceId}-storage is the private bucket

The worker saves one row for each bucket for that day in D1 workspace_stats, table workspace_r2_stat. A later save for the same day replaces that row. The worker reads the totals. It does not open the files.

Builder reads these rows for the storage charts.

Loading this route.

Every workspace has two R2 buckets: one public, one private.

Example row (click to expand)

This is the full row from the worker schema. eventId is bucketName, then |, then the UTC day. occurredAt is the start of that day. Storage fields are the highest value of the day. Operation and bandwidth fields are totals for the day.

{
  "eventId": "d00232-cdn|2026-10-06",
  "workspaceId": "d00232",
  "occurredAt": "2026-10-06T00:00:00.000Z",
  "bucketName": "d00232-cdn",
  "bucketKind": "cdn",
  "payloadSize": 1073741824,
  "metadataSize": 1048576,
  "objectCount": 540,
  "uploadCount": 12,
  "classAOps": 120,
  "classBOps": 8000,
  "freeOps": 10,
  "errorOps": 2,
  "clientErrorOps": 1,
  "serverErrorOps": 1,
  "opsByTypeJson": "{\"PutObject\":12}",
  "bytesDownload": 524288000,
  "bytesUpload": 10485760
}

View all fields

Field	Meaning
eventId	bucketName, then |, then the UTC day YYYY-MM-DD.
workspaceId	The id of the workspace. This is the deployment id.
occurredAt	The start of that UTC day.
bucketName	The bucket this row is for.
bucketKind	cdn for the public bucket, storage for the private bucket.
payloadSize	The highest total bytes stored in the bucket that day.
metadataSize	The highest metadata bytes stored that day.
objectCount	The highest number of files in the bucket that day.
uploadCount	The number of uploads that day.
classAOps	Count of write-type operations that day.
classBOps	Count of read-type operations that day.
freeOps	Count of free operations that day.
errorOps	Count of error operations that day.
clientErrorOps	Count of client error operations that day.
serverErrorOps	Count of server error operations that day.
opsByTypeJson	A JSON string of request counts by Cloudflare action type.
bytesDownload	Total bytes downloaded that day.
bytesUpload	Total bytes uploaded that day.

This type carries no companyId or userId.

If a copy does not arrive

That day's group is skipped. A later save for the same day replaces the row.

Quick map

Data	Sent	Where it is saved	Kept for	Who reads it
Document audit	On each change, about 7 seconds after a burst stops	Iceberg document_audit.v1	Kept in Iceberg; no deletion rule given	The workspace
Document telemetry	Twice a day	D1 for 28 days, and Iceberg document_telemetry.v1	28 days in D1; kept in Iceberg after that	Builder
Logic Pilot traces	In groups, as spans are ready	D1 for 28 days, and Iceberg mastra_spans.v2	28 days in D1; kept in Iceberg after that	Builder
Workspace stats	Twice a day	D1 workspace_stats	Kept in D1; no deletion rule given	Builder
Logic Pilot stats	Twice a day	D1 logic_pilot_stats	Kept in D1; no deletion rule given	Builder
Bug reports	When a person files a bug	D1 feedbacks, table bug	Kept in D1; no deletion rule given	Builder
Workspace feedback	When a person sends feedback	D1 feedbacks, table workspace_feedback	Kept in D1; no deletion rule given	Builder
Logic Pilot feedback	When a person rates a message	D1 feedbacks, table logic_pilot_message	Kept in D1; no deletion rule given	Builder
File storage	Once a day (the worker reads it)	D1 workspace_r2_stat	Kept in D1; no deletion rule given	Builder

Data that is not in the lakehouse

Builder activity, build plans, and CFPs stay in Builder. Their charts read Builder.

The documents, sessions, and files that a person works on stay in the workspace.

Resources

This architecture needs three programs and the Cloudflare resources below.

The workspace backend already exists. It is logic-bee. It runs in each workspace and sends the data.

bob-api already exists. It is the Builder backend. It reads the lakehouse for charts.

The worker is bob-workspace-orchestration. It receives the data and saves it. Create that worker with the Cloudflare resources in the table below.

Programs

Program	Repository	What it does
Workspace backend	logic-bee	Sends data from each workspace. This program already exists.
bob-api	bob-api	Reads the lakehouse for Builder charts. This program already exists.
Worker	bob-workspace-orchestration	Receives the data and saves it. Its Cloudflare resources are in the next table.

Cloudflare

This table is the Cloudflare resources. Use these names. Do not invent new ones.

Iceberg types use a namespace and a version. The Iceberg table is {namespace}.{version}. The stream is {namespace}_{version}. The sink is {namespace}_{version}_sink. The pipeline is {namespace}_{version}_pipeline.

When that type also has a hot store, the D1 database name is the namespace and the D1 table name is the version (v1 or v2).

A D1-only type uses a database name and a table name that describe the data. Several tables can share one database.

Name	Provider	Type	Binding	Parent	Data	Documentation
bob-workspace-orchestration	Cloudflare	Worker	—	—	All lakehouse types	Workers
workspace-events	Cloudflare	R2 bucket	WORKSPACE_EVENTS	—	Iceberg files	R2
workspace_events_cache	Cloudflare	KV namespace	WORKSPACE_EVENTS_CACHE	bob-workspace-orchestration	Document audit and Logic Pilot traces	KV
document_telemetry	Cloudflare	D1 database	DOCUMENT_TELEMETRY	bob-workspace-orchestration	Document telemetry	D1
mastra_spans	Cloudflare	D1 database	MASTRA_SPANS	bob-workspace-orchestration	Logic Pilot traces	D1
feedbacks	Cloudflare	D1 database	FEEDBACKS	bob-workspace-orchestration	Bug reports, workspace feedback, and Logic Pilot feedback	D1
logic_pilot_stats	Cloudflare	D1 database	LOGIC_PILOT_STATS	bob-workspace-orchestration	Logic Pilot stats	D1
workspace_stats	Cloudflare	D1 database	WORKSPACE_STATS	bob-workspace-orchestration	Workspace stats and file storage	D1
v1	Cloudflare	D1 table	—	document_telemetry	Document telemetry	D1
v2	Cloudflare	D1 table	—	mastra_spans	Logic Pilot traces	D1
bug	Cloudflare	D1 table	—	feedbacks	Bug reports	D1
workspace_feedback	Cloudflare	D1 table	—	feedbacks	Workspace feedback	D1
logic_pilot_message	Cloudflare	D1 table	—	feedbacks	Logic Pilot feedback	D1
logic_pilot_conversation_stat	Cloudflare	D1 table	—	logic_pilot_stats	Logic Pilot stats	D1
logic_pilot_folder_stat	Cloudflare	D1 table	—	logic_pilot_stats	Logic Pilot stats	D1
token_telemetry_stat	Cloudflare	D1 table	—	logic_pilot_stats	Logic Pilot stats	D1
workspace_stats	Cloudflare	D1 table	—	workspace_stats	Workspace stats	D1
workspace_r2_stat	Cloudflare	D1 table	—	workspace_stats	File storage	D1
document_audit.v1	Cloudflare	Iceberg table	—	workspace-events	Document audit	R2 Data Catalog
document_telemetry.v1	Cloudflare	Iceberg table	—	workspace-events	Document telemetry	R2 Data Catalog
mastra_spans.v2	Cloudflare	Iceberg table	—	workspace-events	Logic Pilot traces	R2 Data Catalog
document_audit_v1	Cloudflare	Pipeline stream	DOCUMENT_AUDIT_STREAM	bob-workspace-orchestration	Document audit	Pipelines
document_audit_v1_sink	Cloudflare	Pipeline sink	—	workspace-events	Document audit	Pipelines
document_audit_v1_pipeline	Cloudflare	Pipeline	—	—	Document audit	Pipelines
document_telemetry_v1	Cloudflare	Pipeline stream	DOCUMENT_TELEMETRY_STREAM	bob-workspace-orchestration	Document telemetry	Pipelines
document_telemetry_v1_sink	Cloudflare	Pipeline sink	—	workspace-events	Document telemetry	Pipelines
document_telemetry_v1_pipeline	Cloudflare	Pipeline	—	—	Document telemetry	Pipelines
mastra_spans_v2	Cloudflare	Pipeline stream	MASTRA_SPANS_STREAM	bob-workspace-orchestration	Logic Pilot traces	Pipelines
mastra_spans_v2_sink	Cloudflare	Pipeline sink	—	workspace-events	Logic Pilot traces	Pipelines
mastra_spans_v2_pipeline	Cloudflare	Pipeline	—	—	Logic Pilot traces	Pipelines
0 2 * * *	Cloudflare	Cron	—	bob-workspace-orchestration	File storage	Cron Triggers
0 */6 * * *	Cloudflare	Cron	—	bob-workspace-orchestration	Document telemetry and Logic Pilot traces	Cron Triggers

Each workspace has its own file buckets. The worker only reads their totals. Those buckets are not in this list.

The worker vars DOCUMENT_AUDIT_ICEBERG_TABLE, DOCUMENT_TELEMETRY_ICEBERG_TABLE, and MASTRA_SPANS_ICEBERG_TABLE must match the three Iceberg table names.

Cloudflare limits

These limits are set by Cloudflare. They apply to the worker, the stores, the pipelines, and the reads on this page. The numbers were taken from the Cloudflare docs on 7 October 2026. Where a doc lists a Free plan and a Paid plan, the table below uses the Paid plan. The Free plan is lower. Pipelines and R2 SQL are in open beta, so those limits can change.

Worker

The worker receives every ingest and runs both crons. Workers limits.

Limit	Paid plan	What it means here
Memory	128 MB per isolate	A large ingest batch or a large query result must not be held in memory all at once.
CPU time, HTTP	30 seconds by default, up to 5 minutes	Waiting on D1, KV, R2, or fetch does not count as CPU time.
CPU time, cron shorter than 1 hour	30 seconds	This page has no cron that runs that often.
CPU time, cron of 1 hour or longer	15 minutes	The file-storage cron and the 28-day cleanup use this limit.
Cron wall time	15 minutes	A cron that is still running after 15 minutes is stopped.
Subrequests	10,000 per invocation	Each call to D1, KV, R2, or fetch counts.
Connections waiting for headers	6 at one time	A seventh call waits until one of the six receives headers.
Request body	100 MB on Free and Pro, 200 MB on Business, up to 5 GB on Enterprise	A body over the zone limit is rejected with HTTP 413.
waitUntil	30 seconds after the response	Work left running after the response can be stopped after 30 seconds.

An account can have 250 cron triggers on the Paid plan, and 5 on the Free plan. A cron change can take up to 15 minutes to reach the whole network. Cron time is UTC. Cron Triggers.

D1

D1 is the hot store. Each database is separate. D1 limits.

Limit	Paid plan	What it means here
Database size	10 GB	Cloudflare does not raise this. Five databases are used: document_telemetry, mastra_spans, feedbacks, logic_pilot_stats, and workspace_stats.
Queries per invocation	1,000	One cron run shares this cap across the databases it touches.
SQL duration	30 seconds	A delete of a huge set of rows in one statement can hit this. The 28-day cleanup deletes in batches.
Row size	2 MB	One row, including JSON strings, must stay under 2 MB.
SQL statement	100 KB
Bound parameters	100 per query
Columns per table	100
Concurrency	One query at a time per database	Extra queries wait. A full queue returns an overloaded error.

Pipelines

Pipelines carry document audit, document telemetry, and Logic Pilot traces into Iceberg. The product is in open beta. Pipelines limits.

Limit	Value	What it means here
Streams, sinks, and pipelines	20 of each per account	This page uses 3 of each.
Ingest request	5 MB	One send to a stream must stay under 5 MB.
Ingest rate	5 MB per second per stream
Stream schema	Fixed after creation	A new column needs a new stream, sink, pipeline, and Iceberg table. Logic Pilot traces did this for v2.
Bad events	Dropped while processing	An event that does not match the schema is accepted, then dropped.
Iceberg sink	Parquet only	A sink cannot be attached to an Iceberg table that already exists.
Roll interval	60 seconds minimum	Rows are ready about 1 minute after the worker sends them.

Streams. R2 Data Catalog sink.

R2 and Iceberg

Iceberg files live in the R2 bucket workspace-events. R2 limits. Table maintenance.

Limit	Value	What it means here
Bucket storage and object count	No stated maximum
One object	5 TiB
Writes to the same object name	1 per second
Catalog jurisdiction	Default jurisdiction only	A bucket in the EU or FedRAMP jurisdiction cannot use the catalog.
Compaction file size	64 MB to 512 MB	Compaction reads Parquet files only. Files that no snapshot points at are not deleted.
Snapshot expiration	Off unless enabled	If it is enabled, the default is to drop snapshots older than 30 days and keep the last 5.

R2 SQL

Builder and the workspace read Iceberg through R2 SQL. R2 SQL is read-only and in open beta. It reads Parquet only. R2 SQL limitations. R2 SQL troubleshooting.

Limit	Value	What it means here
Writes	Not allowed	New rows arrive through Pipelines, not through R2 SQL.
LIMIT	10,000 rows maximum
OFFSET	Not supported	A later page uses WHERE and ORDER BY, not OFFSET.
JSON inside a string column	No JSON path filter	jsonPatch and the other JSON strings are filtered in the application, or stored as their own columns.
Heavy queries	Can time out or return HTTP 400	A large COUNT(DISTINCT), a large sort, or a join of three big tables can be rejected.

KV

workspace_events_cache holds the short cache for a document audit and for one Logic Pilot trace. KV limits.

Limit	Paid plan	What it means here
Writes to the same key	1 per second	A burst of saves for one trace can hit this.
Operations per invocation	1,000
Value size	25 MiB
Key size	512 bytes

GraphQL Analytics

The file-storage cron reads bucket totals from the Cloudflare GraphQL Analytics API. It does not read the files. GraphQL API limits.

Limit	Value	What it means here
Requests	300 per 5 minutes for one user or token	The cron runs once a day, so this cap is wide for that job.
Accounts in one query	1