Overview - LangWatch

The Query API runs read-only LangWatchQL SELECT statements over your project’s analytics views. POST /api/v1/query runs one statement. GET /api/v1/query/schema lists the views, columns, and function names your key can query. The request body is the query itself, and a 200 is the result itself: no envelope, no wrapper object around it. A failure answers with the standard LangWatch error envelope and a real HTTP status, so one parser reads every response.

Endpoints

Method	Path	Description
`POST`	`/api/v1/query`	Run one LangWatchQL statement.
`GET`	`/api/v1/query/schema`	List queryable views, columns, permission gates, and function names.

Authentication

Send your LangWatch API key in any one of these forms:

Form	Header
Token header	`X-Auth-Token: <token>`
Bearer	`Authorization: Bearer <token>`
Basic	`Authorization: Basic base64(projectId:token)`

Any LangWatch API key works here: project, organization, or personal. Rows come from every project the key holds the analytics:view permission on: a project key reaches its one project, an organization or personal key reaches every project it can read. No project header is required. To run POST /api/v1/query over one project only, send its id as projectId in the request body: the run then reads that project’s rows alone, under that project’s own content protections. Without projectId the run covers every project the key can read. You can also filter inside the statement itself with WHERE TenantId = '<project id>', but the result is then redacted by the strictest protection across all the projects the key can read. A result that mixes rows from more than one project carries the MULTI_PROJECT_RESULT diagnostic, so a caller who expected a single project can tell.

Errors

Both endpoints answer with the standard LangWatch error envelope and a real HTTP status. No wrapper sits around it. See Error Envelope. Where a failure has a fix, the envelope’s meta carries the corrective fields for it. For example, FUNCTION_NOT_ALLOWED lists allowedFunctions.

Status	`code`	When
`400`	`validation_error`	The request body failed validation.
`400`	`lwql_unparseable`	The statement is not valid LangWatchQL.
`400`	`lwql_not_permitted`	The statement is valid but calls something the policy withholds, for example a function outside the allowlist, or a `LIMIT` above 10,000 rows (`meta.violations` names it `LIMIT_TOO_HIGH` or `LIMIT_REQUIRED_PER_BRANCH`).
`401`	`missing_credentials`	No supported authentication credential was sent.
`401`	`invalid_credentials`	The API key is unknown, revoked, or has the wrong secret.
`403`	`api_key_permission_denied`	The key lacks the `analytics:view` permission on any project.
`404`	`project_not_found`	The body’s `projectId` names a project the key cannot read with `analytics:view`, or one that does not exist. The two cases get the same answer.
`413`	`lwql_result_too_large`	The result body exceeds about 8,000,000 bytes.
`422`	`query_scan_limit_exceeded`	The statement is valid but would scan more data than the key allows.

Queryable views

Every view is a read-only projection over one ClickHouse or PostgreSQL table. GET /api/v1/query/schema is the live source of truth: the table below is a human-readable index of the same 47 views, grouped by what they’re for.

How the catalog is built

Every ClickHouse table is a view by default: the catalog is derived from a generated columns manifest, so a new table shows up here automatically once it’s catalogued, rather than needing a hand-written entry. Two column families are gated rather than freely selectable: free-text carriers (captured input, captured output, tool output, log bodies) sit behind the output gate (input for a few views that separate the two), and billed/derived cost columns sit behind the costs gate. A gate’s state comes from the project’s privacy settings, never from the query. Selecting a gated column your key’s project has not opened is refused outright with GATED_COLUMN, never silently dropped from the result.

Observability: traces, spans, logs

View	Description	Gates	Grain	Time column
`traces`	One row per trace, with its timing, token, cost and error rollups.	`input`, `output`, `costs`	one row per (TenantId, TraceId), latest version only	`OccurredAt`
`spans`	One row per span, with its timing, status and attributes.	`input`, `output`, `costs`	one row per (TenantId, TraceId, SpanId), latest version only	`StartTime`
`logs`	Canonical OpenTelemetry log records: requests, responses and provider events, correlated to their trace and span.	`input`, `output`	one row per (TenantId, TraceId, TimeUnixMs, RecordId)	`TimeUnixMs`
`log_ingestion_usage`	Per-tenant log ingestion usage estimates.	None	one row per (OrganizationId, RecordId), latest version only	`AcceptedAt`
`metric_points`	Individual metric data points with gauges, sums, histograms, and summaries	`output`	one row per (SeriesId, TimeUnixMs, PointId)	`TimeUnixMs`
`metric_series_definitions`	Metric series definitions from OpenTelemetry: name, unit, attributes, kind	`output`	one row per SeriesId	`LastSeenAt`
`metric_ingestion_usage`	Per-tenant metric ingestion usage estimates	None	one row per (OrganizationId, PointId), latest version only	`AcceptedAt`
`objects`	Stored file objects: what was captured, its size and hash.	None	one row per id	`created_at`
`annotations`	One row per human annotation of a trace, with the reviewer’s thumbs verdict.	None	one row per AnnotationId	`CreatedAt`
`langy_conversation_messages`	Langy conversation turns, one row per message.	`output`	one row per (TenantId, ConversationId, MessageId)	`CreatedAt`
`langy_usage_events`	Langy usage telemetry: tool calls, outcomes and durations.	None	one row per (TenantId, OccurredAt, EventId)	`OccurredAt`

Coding agents

View	Description	Gates	Grain	Time column
`coding_sessions`	One row per coding-agent session, with its shape, economics, git context and outcome. Deduplicated by session.	`input`, `costs`	one row per (TenantId, SessionId), latest version only	`StartedAt`
`coding_session_events`	One row per coding-agent session event: model calls, compactions, rate limits, tool runs and more, in event order.	`costs`	one row per (TenantId, SessionId, TimeUnixMs, RecordId)	`TimeUnixMs`
`coding_trace_sessions`	Correlates a trace to the coding-agent session it belongs to.	None	one row per (TenantId, TraceId)	`OccurredAt`
`coding_tool_results`	One row per coding-agent tool call, with the text it printed back to the agent.	`output`	one row per (TenantId, TraceId, SpanId), latest version only	`StartTime`
`judgments`	One row per Instant Eval run, trace and question, with the verdict the judge gave.	None	one row per (TenantId, RunId, TraceId, SpanId, QuestionId), latest version only	`CreatedAt`

coding_tool_results.OutputText is what the tool printed back to the agent. Read it to answer “what did this failed tool call print”:

SELECT ToolName, OutputText
FROM coding_tool_results
WHERE TenantId = '<project id>' AND Success = 0
ORDER BY StartTime DESC
LIMIT 20

Experiments & evaluations

View	Description	Gates	Grain	Time column
`evaluations`	One row per evaluation run, with its score and outcome.	`input`, `output`	one row per (TenantId, EvaluationId), latest version only	`ScheduledAt`
`simulations`	One row per simulation run, with its verdict and criteria.	`input`, `output`	one row per (TenantId, ScenarioRunId), latest version only	`StartedAt`
`simulation_trace_metrics`	Cost and latency metrics per simulated trace	`costs`	one row per (ScenarioRunId, TraceId)	`OccurredAt`
`experiments`	One row per experiment, with its display name and kind.	None	one row per ExperimentId	`CreatedAt`
`experiment_items`	Individual item results within an experiment run with evaluations	`input`, `output`, `costs`	one row per (RunId, ProjectionId)	`OccurredAt`
`experiment_run_results`	Experiment run results: completion counts, cost, and aggregate scores	`output`, `costs`	one row per (RunId, ExperimentId)	`StartedAt`
`dspy_optimizer_steps`	DSPy optimizer steps with prompts, examples, and LLM call details	`output`, `costs`	one row per (ExperimentId, RunId, StepIndex)	`CreatedAt`
`batch_evaluations`	One row per offline batch evaluation of a dataset row, with its score, outcome and cost.	`costs`	one row per BatchEvaluationId	`CreatedAt`
`test_suite_runs`	Test suite execution results with completion and pass counts	None	one row per (ScenarioSetId, BatchRunId)	`StartedAt`

Note: “dataset” above refers to an evaluation dataset (rows a batch evaluation runs over). Everything else queryable through LangWatchQL is a view.

Metrics & rollups

Rollup views (_by_minute, _rollups) return finalised values only: figures for a bucket that was still open when the row was written are absent, not partial. A metric a caller wants without touching per-row data belongs here.

View	Description	Gates	Grain	Time column
`trace_metrics`	One row per trace, time-sorted, carrying its metrics and the user, conversation, customer and origin dimensions. Count traces with uniqExact(TraceId).	`costs`	one row per (TenantId, OccurredAt, TraceId), latest version only; one row per trace wherever OccurredAt held still	`OccurredAt`
`trace_metrics_by_minute`	Trace and span metrics summed per minute. Every measure is a sum: divide by TraceCount for a per-trace average.	`costs`	one row per (TenantId, BucketStart), every measure summed	`BucketStart`
`model_usage_by_minute`	Span metrics summed per minute, per model and span type: span counts, costs and tokens.	`costs`	one merged row per (TenantId, BucketStart, Model, SpanType), every measure summed	`BucketStart`
`evaluation_metrics`	One row per evaluation, its latest state, time-sorted, carrying its outcome, cost and the trace dimensions it inherited.	`costs`	one row per (TenantId, EvaluationId), latest version only	`OccurredAt`
`evaluation_metrics_by_minute`	Evaluation outcomes summed per minute, per evaluator type and terminal status. Every measure is a sum: pass rate is PassCount over PassCount plus FailCount.	`costs`	one merged row per (TenantId, BucketStart, EvaluatorType, Status), every measure summed	`BucketStart`
`metric_rollups`	Time-bucketed rollups of a metric series: per-bucket min, max, sum, count and histogram buckets.	None	one row per (SeriesId, BucketStart)	`BucketStart`
`session_metrics`	Per-session time-series metrics with aggregate statistics	None	one row per (SessionId, SeriesId)	`AsOf`
`simulation_metric_rollups`	Per-trace simulation cost and latency, merged from the aggregating rollup: total cost and per-role cost and latency maps.	`costs`	one row per (TenantId, ScenarioRunId, TraceId), latest version only	`OccurredAt`

Gateway & billing

View	Description	Gates	Grain	Time column
`gateway_budget_ledger`	Budget ledger entries per request with token counts and spend	`costs`	one row per (BudgetId, GatewayRequestId)	`OccurredAt`
`gateway_budget_totals`	Per-scope budget totals, merged from the aggregating rollup: spend, token counts and request count per budget window.	`costs`	one row per (TenantId, Scope, ScopeId, Window, PeriodStart, BudgetId), latest version only	`PeriodStart`
`gateway_request_spend`	Individual API request spend and reconciliation details	`costs`, `output`	one row per GatewayRequestId	`OccurredAt`
`billing_events`	Billing events for customer invoice generation	None	one row per DeduplicationKeyHash	`EventTimestamp`

Governance & audit

Governance views are keyed by the organization’s hidden internal_governance project. A key scoped to an ordinary project sees zero rows here, not an error and not a gate refusal, until the caller’s project set includes the org-level governance project.

View	Description	Gates	Grain	Time column
`governance_cost_restatements`	Index of cost-rollup restatements, keyed by the org’s internal_governance project; visible only to a caller whose project set includes it.	`costs`	one row per (TenantId, RestatementKey)	`TenantId`
`governance_daily_cost_rollup`	Daily cost rollups per tenant, provider and model, keyed by the org’s internal_governance project; visible only to a caller whose project set includes it.	`costs`, `output`	one row per (TenantId, Day, CostSource, IngestionSourceId, Provider, Model, AgentId, CurrencyCode, RawActorId)	`Day`
`governance_hourly_kpis`	Hourly spend and token KPIs per source, keyed by the org’s internal_governance project; visible only to a caller whose project set includes it.	`costs`	one row per (TenantId, SourceId, HourBucket, TraceId)	`HourBucket`
`governance_security_events`	Raw OCSF-shaped security events, keyed by the org’s internal_governance project; visible only to a caller whose project set includes it.	`output`	one row per (TenantId, EventId)	`EventTime`

Automation firing history is not an analytics view. Read it per automation with GET /api/triggers/{id}/fires, langwatch trigger fires <id>, or the MCP tool platform_list_trigger_fires, each paged by cursor.

Legacy

Superseded by the canonical views above; kept for rolling-deployment reads and draining under their own TTL.

View	Description	Gates	Grain	Time column
`legacy_event_log`	Legacy append-only event log, superseded by the canonical fact tables	`output`	one row per (AggregateType, AggregateId, IdempotencyKey)	`EventOccurredAt`
`legacy_log_records`	Pre-canonical OpenTelemetry log record storage, superseded by log_records; retained for rolling-deployment reads and draining under its own TTL.	`output`	one row per (TraceId, SpanId, ProjectionId), latest version only	`TimeUnixMs`
`legacy_metric_records`	Pre-canonical OpenTelemetry metric record storage, superseded by the metric_data_points family; retained for rolling-deployment reads and draining under its own TTL.	None	one row per (TraceId, SpanId, MetricName, ProjectionId), latest version only	`TimeUnixMs`

Entities & reference data

View	Description	Gates	Grain	Time column
`projects`	The caller’s project, with its display name and slug.	None	one row per TenantId	`CreatedAt`
`prompts`	One row per prompt configuration, with the name its versions are known by.	None	one row per PromptId	`CreatedAt`
`prompt_versions`	One row per version of a prompt, carrying the version number a trace records.	None	one row per PromptVersionId	`CreatedAt`

Application tables (PostgreSQL)

Every tenant-scoped table of the application database, derived from the schema. Organization- and team-scoped tables appear once per project the caller can reach. Secret material and person identifiers are never exposed; free-text and JSON bodies are gated on the content permissions.

View	Description	Gates	Grain	Time column
`agents`	Rows of the Agent table, one row per AgentId.	`output`	one row per AgentId	`CreatedAt`
`ai_tool_entries`	Org-scoped catalog of AI tools surfaced on the personal home portal (/me) and managed by org admins at /governance/tool-catalog.	`output`	one row per (TenantId, AiToolEntryId); an organization-scoped row appears once per project	`CreatedAt`
`ai_tool_entry_departments`	Per-department scope binding for AiToolEntry.	None	one row per (TenantId, AiToolEntryDepartmentId); a row reached through its parent appears once per project	None
`ai_tool_entry_teams`	Per-team scope binding for AiToolEntry.	None	one row per (TenantId, AiToolEntryTeamId); a team-scoped row appears once per project	None
`analytics`	Rows of the Analytics table, one row per AnalyticsId.	`output`	one row per AnalyticsId	`CreatedAt`
`annotation_queue_items`	Rows of the AnnotationQueueItem table, one row per AnnotationQueueItemId.	None	one row per AnnotationQueueItemId	`CreatedAt`
`annotation_queue_scores`	Rows of the AnnotationQueueScores table, one row per (TenantId, AnnotationQueueId, AnnotationScoreId); a row reached through its parent appears once per project.	None	one row per (TenantId, AnnotationQueueId, AnnotationScoreId); a row reached through its parent appears once per project	None
`annotation_queues`	Rows of the AnnotationQueue table, one row per AnnotationQueueId.	`output`	one row per AnnotationQueueId	`CreatedAt`
`annotation_scores`	Rows of the AnnotationScore table, one row per AnnotationScoreId.	`output`	one row per AnnotationScoreId	`CreatedAt`
`anomaly_alerts`	Persisted anomaly detection.	`output`, `costs`	one row per (TenantId, AnomalyAlertId); an organization-scoped row appears once per project	`TriggerWindowStart`
`anomaly_rules`	Admin-defined anomaly detection rule.	`output`	one row per (TenantId, AnomalyRuleId); an organization-scoped row appears once per project	`CreatedAt`
`costs`	Rows of the Cost table, one row per CostId.	`costs`, `output`	one row per CostId	`CreatedAt`
`custom_graphs`	Rows of the CustomGraph table, one row per CustomGraphId. Rows placed on a dashboard set to Only me are not queryable.	`output`	one row per CustomGraphId	`CreatedAt`
`custom_llm_model_costs`	Rows of the CustomLLMModelCost table, one row per CustomLLMModelCostId.	`costs`, `output`	one row per CustomLLMModelCostId	`CreatedAt`
`dashboards`	Rows of the Dashboard table, one row per DashboardId. Dashboards set to Only me are not queryable.	None	one row per DashboardId	`CreatedAt`
`data_privacy_policies`	Rows of the DataPrivacyPolicy table, one row per (TenantId, DataPrivacyPolicyId); an organization-scoped row appears once per project.	`output`	one row per (TenantId, DataPrivacyPolicyId); an organization-scoped row appears once per project	`CreatedAt`
`dataset_records`	Rows of the DatasetRecord table, one row per DatasetRecordId.	`output`	one row per DatasetRecordId	`CreatedAt`
`datasets`	Rows of the Dataset table, one row per DatasetId.	`output`	one row per DatasetId	`CreatedAt`
`department_membership_histories`	One dated person → department link.	None	one row per (TenantId, DepartmentMembershipHistoryId); an organization-scoped row appears once per project	`CreatedAt`
`departments`	A department is a single-valued accounting dimension, separate from RBAC, for rolling up AI spend by org unit (Engineering, Marketing,…).	None	one row per (TenantId, DepartmentId); an organization-scoped row appears once per project	`CreatedAt`
`discovered_agents`	A non-human actor (a bot, agent or automation) discovered from a connected source, with the provider-native metadata the source published.	`output`	one row per (TenantId, DiscoveredAgentId); an organization-scoped row appears once per project	`CreatedAt`
`discovered_persons`	A name-text seen on a provider’s cost or audit rows: “the provider told us someone called m.silva exists”.	`output`, `costs`	one row per (TenantId, DiscoveredPersonId); an organization-scoped row appears once per project	`CreatedAt`
`email_suppressions`	Rows of the EmailSuppression table, one row per EmailSuppressionId.	`output`	one row per EmailSuppressionId	`CreatedAt`
`erased_identifier_suppressions`	Hashes of erased identifiers - the do-not-reimport list.	None	one row per (TenantId, ErasedIdentifierSuppressionId); an organization-scoped row appears once per project	`ErasedAt`
`evaluators`	Rows of the Evaluator table, one row per EvaluatorId.	`output`	one row per EvaluatorId	`CreatedAt`
`experiment_versions`	Rows of the ExperimentVersion table, one row per ExperimentVersionId.	`output`	one row per ExperimentVersionId	`CreatedAt`
`gateway_budget_bucket_boundaries`	Rows of the GatewayBudgetBucketBoundary table, one row per (TenantId, GatewayBudgetBucketBoundaryId); an organization-scoped row appears once per project.	None	one row per (TenantId, GatewayBudgetBucketBoundaryId); an organization-scoped row appears once per project	`CreatedAt`
`gateway_budget_ledgers`	Rows of the GatewayBudgetLedger table, one row per (TenantId, GatewayBudgetLedgerId); a row reached through its parent appears once per project.	`costs`	one row per (TenantId, GatewayBudgetLedgerId); a row reached through its parent appears once per project	`OccurredAt`
`gateway_budgets`	Rows of the GatewayBudget table, one row per (TenantId, GatewayBudgetId); an organization-scoped row appears once per project.	`output`, `costs`	one row per (TenantId, GatewayBudgetId); an organization-scoped row appears once per project	`CreatedAt`
`gateway_cache_rules`	Rows of the GatewayCacheRule table, one row per (TenantId, GatewayCacheRuleId); an organization-scoped row appears once per project.	`output`	one row per (TenantId, GatewayCacheRuleId); an organization-scoped row appears once per project	`CreatedAt`
`gateway_change_events`	Rows of the GatewayChangeEvent table, one row per Revision.	`output`	one row per Revision	`CreatedAt`
`gateway_guardrails`	Rows of the GatewayGuardrail table, one row per GatewayGuardrailId.	`output`	one row per GatewayGuardrailId	`CreatedAt`
`gateway_realtime_sessions`	Rows of the GatewayRealtimeSession table, one row per GatewayRealtimeSessionId.	`output`, `costs`	one row per GatewayRealtimeSessionId	`CreatedAt`
`github_branch_pull_request_checks`	The per-branch bookkeeping behind pull-request mapping: when a branch was last looked up on GitHub, how many pull requests it resolved to, and when it may be asked about again.	`output`	one row per (TenantId, GithubBranchPullRequestCheckId); an organization-scoped row appears once per project	`CreatedAt`
`github_installations`	The organization’s GitHub connection: a GitHub App installation mapped to a LangWatch organization.	`output`, `costs`	one row per (TenantId, GithubInstallationId); an organization-scoped row appears once per project	`CreatedAt`
`github_pull_requests`	A pull request discovered through the organization’s GitHub connection and mapped to the branch it was opened from, so a coding-agent session that knows only its branch can name the pull request it produced.	`output`, `input`	one row per (TenantId, GithubPullRequestId); an organization-scoped row appears once per project	`CreatedAt`
`governance_tenant_histories`	Every tenant id this organization has ever written governance rows under, appended on first use and never pruned.	None	one row per (TenantId, GovernanceTenantHistoryId); an organization-scoped row appears once per project	`FirstUsedAt`
`identity_match_suggestions`	Match candidates a background job computed and a human has yet to confirm.	None	one row per (TenantId, IdentityMatchSuggestionId); an organization-scoped row appears once per project	`ComputedAt`
`identity_matches`	A dated link from an external account to a platform user, opened when the match holds and closed when it stops being true.	None	one row per (TenantId, IdentityMatchId); an organization-scoped row appears once per project	`CreatedAt`
`ingestion_pull_run_projections`	Rebuildable per-source pull checkpoint and operator-facing run status.	`output`	one row per IngestionPullRunProjectionId	`CreatedAt`
`ingestion_sources`	Per-platform fleet configuration that connects a closed SaaS product’s audit / OTel / S3 stream to LangWatch’s Activity Monitor.	`output`, `costs`	one row per (TenantId, IngestionSourceId); a team-scoped row appears once per project	`CreatedAt`
`ingestion_templates`	Admin/platform-curated catalog row for upstream OTLP normalization.	`output`	one row per (TenantId, IngestionTemplateId); an organization-scoped row appears once per project	`CreatedAt`
`langy_active_turns`	Postgres-authoritative one-turn admission for a conversation. Shared Langy conversations only; a member’s private conversations are not queryable.	`output`	one row per LangyActiveTurnId	`CreatedAt`
`langy_conversation_projections`	Rebuildable operational state folded from Langy’s canonical event stream. Shared Langy conversations only; a member’s private conversations are not queryable.	`input`, `output`	one row per LangyConversationProjectionId	`CreatedAt`
`langy_conversation_turn_projections`	Rebuildable per-turn operational render document. Shared Langy conversations only; a member’s private conversations are not queryable.	`output`	one row per LangyConversationTurnProjectionId	`CreatedAt`
`langy_message_projections`	Rebuildable operational message fact. Shared Langy conversations only; a member’s private conversations are not queryable.	`output`	one row per LangyMessageProjectionId	`CreatedAt`
`langy_turn_requests`	Durable idempotency receipt for one logical Langy send. Shared Langy conversations only; a member’s private conversations are not queryable.	`output`	one row per LangyTurnRequestId	`CreatedAt`
`model_default_config_scopes`	Rows of the ModelDefaultConfigScope table, one row per (TenantId, ModelDefaultConfigScopeId); a row reached through its parent appears once per project.	None	one row per (TenantId, ModelDefaultConfigScopeId); a row reached through its parent appears once per project	`CreatedAt`
`model_default_configs`	Rows of the ModelDefaultConfig table, one row per (TenantId, ModelDefaultConfigId); an organization-scoped row appears once per project.	`output`	one row per (TenantId, ModelDefaultConfigId); an organization-scoped row appears once per project	`CreatedAt`
`model_provider_scopes`	Rows of the ModelProviderScope table, one row per (TenantId, ModelProviderScopeId); a row reached through its parent appears once per project.	None	one row per (TenantId, ModelProviderScopeId); a row reached through its parent appears once per project	`CreatedAt`
`model_providers`	Rows of the ModelProvider table, one row per (TenantId, ModelProviderId); an organization-scoped row appears once per project.	`output`	one row per (TenantId, ModelProviderId); an organization-scoped row appears once per project	`CreatedAt`
`monitors`	Rows of the Monitor table, one row per MonitorId.	`output`	one row per MonitorId	`CreatedAt`
`notifications`	Rows of the Notification table, one row per NotificationId.	`output`	one row per NotificationId	`CreatedAt`
`pinned_traces`	Rows of the PinnedTrace table, one row per PinnedTraceId.	`output`	one row per PinnedTraceId	`CreatedAt`
`process_manager_inboxes`	Idempotency marker for a source event consumed by a process definition.	None	one row per ProcessManagerInboxId	`ConsumedAt`
`process_manager_instances`	Durable state for a long-running event-sourced process.	`output`	one row per ProcessManagerInstanceId	`NextWakeAt`
`process_manager_outbox_attempts`	One row per failed outbox delivery attempt, so the table grows with delivery trouble rather than with traffic.	`output`	one row per ProcessManagerOutboxAttemptId	`OccurredAt`
`process_manager_outboxes`	Transactional intents emitted by a durable process transition.	`output`	one row per ProcessManagerOutboxId	`CreatedAt`
`prompt_tag_assignments`	Rows of the PromptTagAssignment table, one row per PromptTagAssignmentId.	None	one row per PromptTagAssignmentId	`CreatedAt`
`prompt_tags`	Rows of the PromptTag table, one row per (TenantId, PromptTagId); an organization-scoped row appears once per project.	None	one row per (TenantId, PromptTagId); an organization-scoped row appears once per project	`CreatedAt`
`retention_policies`	Rows of the RetentionPolicy table, one row per (TenantId, RetentionPolicyId); an organization-scoped row appears once per project.	`output`	one row per (TenantId, RetentionPolicyId); an organization-scoped row appears once per project	`CreatedAt`
`routing_policies`	Rows of the RoutingPolicy table, one row per (TenantId, RoutingPolicyId); an organization-scoped row appears once per project.	`output`	one row per (TenantId, RoutingPolicyId); an organization-scoped row appears once per project	`CreatedAt`
`routing_policy_scopes`	Rows of the RoutingPolicyScope table, one row per (TenantId, RoutingPolicyScopeId); a row reached through its parent appears once per project.	None	one row per (TenantId, RoutingPolicyScopeId); a row reached through its parent appears once per project	`CreatedAt`
`saved_views`	Rows of the SavedView table, one row per SavedViewId.	`output`	one row per SavedViewId	`CreatedAt`
`scenario_versions`	Rows of the ScenarioVersion table, one row per ScenarioVersionId.	`output`	one row per ScenarioVersionId	`CreatedAt`
`scenarios`	Rows of the Scenario table, one row per ScenarioId.	`output`	one row per ScenarioId	`CreatedAt`
`scheduled_jobs`	Rows of the ScheduledJob table, one row per ScheduledJobId.	`output`	one row per ScheduledJobId	`CreatedAt`
`share_links`	Rows of the ShareLink table, one row per ShareLinkId.	None	one row per ShareLinkId	`CreatedAt`
`simulation_suites`	Rows of the SimulationSuite table, one row per SimulationSuiteId.	`output`	one row per SimulationSuiteId	`CreatedAt`
`topic_clustering_run_history_projections`	Per-project topic clustering run history (audit read model).	`output`	one row per TopicClusteringRunHistoryProjectionId	`CreatedAt`
`topic_clustering_run_projections`	Per-project topic clustering run status: the rebuildable operational read model behind the settings page.	`output`	one row per TopicClusteringRunProjectionId	`CreatedAt`
`topic_model_projections`	Cursor row for the topic-model projection.	None	one row per TopicModelProjectionId	`CreatedAt`
`topics`	One row per topic, with the name a trace’s TopicId resolves to.	None	one row per TopicId	`CreatedAt`
`trace_edit_overlays`	Rows of the TraceEditOverlay table, one row per TraceEditOverlayId.	`output`	one row per TraceEditOverlayId	`CreatedAt`
`trigger_sents`	Rows of the TriggerSent table, one row per TriggerSentId.	None	one row per TriggerSentId	`CreatedAt`
`triggers`	Rows of the Trigger table, one row per TriggerId.	`output`	one row per TriggerId	`CreatedAt`
`virtual_key_scopes`	Rows of the VirtualKeyScope table, one row per (TenantId, VirtualKeyScopeId); a row reached through its parent appears once per project.	None	one row per (TenantId, VirtualKeyScopeId); a row reached through its parent appears once per project	`CreatedAt`
`virtual_keys`	Rows of the VirtualKey table, one row per (TenantId, VirtualKeyId); an organization-scoped row appears once per project.	`output`	one row per (TenantId, VirtualKeyId); an organization-scoped row appears once per project	`CreatedAt`
`webhook_endpoint_deliveries`	The per-ATTEMPT delivery log for BOTH webhook channels: the endpoints platform’s batches and the automations channel’s trigger fires.	`output`	one row per WebhookEndpointDeliveryId	`CreatedAt`
`workflow_versions`	Rows of the WorkflowVersion table, one row per WorkflowVersionId.	`output`	one row per WorkflowVersionId	`CreatedAt`
`workflows`	Rows of the Workflow table, one row per WorkflowId.	`output`	one row per WorkflowId	`CreatedAt`

Tables intentionally left out

A table off the catalog above is always a recorded decision, never an oversight: a coverage guard fails the build until every ClickHouse table is either catalogued or given a reason here:

Table	Reason
`lwql_api_key_tenant_map`	Access-control plumbing, not customer telemetry: holds key hashes the row policy reads to self-filter; the policy already scopes it, so exposing it would leak the isolation mechanism for no customer benefit.
`goose_db_version`	The goose migration-version table, engine-internal tooling state, no tenant column at all.
Any table ending in `_mv`, or naming a `.inner` target	Materialised-view internals: engine plumbing, not a table a caller would query; its data is already exposed through the table it feeds.

Supported functions

A query may call only the functions on the allowlist below. GET /api/v1/query/schema serves the same list under its functions field, so an agent can read it back at run time, and a FUNCTION_NOT_ALLOWED refusal carries it under allowedFunctions so a rejected query can be fixed without another round trip. Combinator forms of the listed aggregates, such as countIf, are also accepted.

abs, addDays, addHours, addMinutes, addMonths, addQuarters, addSeconds, addWeeks, addYears, and
any, anyLast, argMax, argMin, arrayAll, arrayAvg, arrayConcat, arrayCount, arrayDistinct, arrayElement
arrayEnumerate, arrayExists, arrayFilter, arrayFirst, arrayFirstIndex, arrayJoin, arrayLast, arrayMap, arrayMax, arrayMin
arrayReverseSort, arraySlice, arraySort, arrayStringConcat, arraySum, arrayUniq, assumeNotNull, avg, caseWithExpression, caseWithoutExpression
CAST, ceil, ceiling, coalesce, concat, concatWithSeparator, count, date_trunc, dateDiff, dateTrunc
dense_rank, divide, empty, endsWith, equals, exists, exp, extract, extractAll, first_value
floor, formatDateTime, fromUnixTimestamp, globalIn, globalNotIn, greater, greaterOrEquals, greatest, groupArray, groupUniqArray
has, hasAll, hasAny, if, ifNull, ilike, in, indexOf, intDiv, intDivOrZero
isDistinctFrom, isNotDistinctFrom, isNotNull, isNull, isValidJSON, JSONExtract, JSONExtractArrayRaw, JSONExtractBool, JSONExtractFloat, JSONExtractInt
JSONExtractKeys, JSONExtractKeysAndValues, JSONExtractRaw, JSONExtractString, JSONExtractUInt, JSONHas, JSONLength, JSONType, lagInFrame, lambda
last_value, leadInFrame, least, left, length, lengthUTF8, less, lessOrEquals, like, ln
log, log10, log2, lower, lowerUTF8, map, mapContains, mapFilter, mapKeys, mapValues
match, max, median, medianExact, min, minus, modulo, moduloOrZero, multiIf, multiply
negate, not, notEmpty, notEquals, notILike, notIn, notLike, now, now64, nth_value
ntile, nullIf, or, plus, position, positionCaseInsensitive, pow, power, quantile, quantileExact
quantiles, quantilesExact, quantilesTDigest, quantilesTiming, quantileTDigest, quantileTiming, rank, replaceAll, replaceOne, replaceRegexpAll
replaceRegexpOne, right, round, row_number, sign, simpleJSONExtractBool, simpleJSONExtractFloat, simpleJSONExtractInt, simpleJSONExtractString, simpleJSONExtractUInt
simpleJSONHas, splitByChar, splitByString, sqrt, startsWith, stddevPop, stddevSamp, substring, substringUTF8, subtractDays
subtractHours, subtractMinutes, subtractMonths, subtractQuarters, subtractSeconds, subtractWeeks, subtractYears, sum, toBool, toDate
toDate32, toDate32OrDefault, toDate32OrNull, toDate32OrZero, toDateOrDefault, toDateOrNull, toDateOrZero, toDateTime, toDateTime64, toDateTime64OrDefault
toDateTime64OrNull, toDateTime64OrZero, toDateTimeOrDefault, toDateTimeOrNull, toDateTimeOrZero, today, toDayOfMonth, toDayOfWeek, toDayOfYear, toDecimal128
toDecimal128OrDefault, toDecimal128OrNull, toDecimal128OrZero, toDecimal256, toDecimal256OrDefault, toDecimal256OrNull, toDecimal256OrZero, toDecimal32, toDecimal32OrDefault, toDecimal32OrNull
toDecimal32OrZero, toDecimal64, toDecimal64OrDefault, toDecimal64OrNull, toDecimal64OrZero, toFloat32, toFloat32OrDefault, toFloat32OrNull, toFloat32OrZero, toFloat64
toFloat64OrDefault, toFloat64OrNull, toFloat64OrZero, toHour, toInt128, toInt128OrDefault, toInt128OrNull, toInt128OrZero, toInt16, toInt16OrDefault
toInt16OrNull, toInt16OrZero, toInt256, toInt256OrDefault, toInt256OrNull, toInt256OrZero, toInt32, toInt32OrDefault, toInt32OrNull, toInt32OrZero
toInt64, toInt64OrDefault, toInt64OrNull, toInt64OrZero, toInt8, toInt8OrDefault, toInt8OrNull, toInt8OrZero, toIntervalDay, toIntervalHour
toIntervalMinute, toIntervalMonth, toIntervalQuarter, toIntervalSecond, toIntervalWeek, toIntervalYear, toISOWeek, toLastDayOfMonth, toMinute, toMonday
toMonth, topK, toQuarter, toSecond, toStartOfDay, toStartOfFifteenMinutes, toStartOfFiveMinutes, toStartOfHour, toStartOfInterval, toStartOfISOWeek
toStartOfMinute, toStartOfMonth, toStartOfQuarter, toStartOfSecond, toStartOfTenMinutes, toStartOfWeek, toStartOfYear, toString, toTimeZone, toUInt128
toUInt128OrDefault, toUInt128OrNull, toUInt128OrZero, toUInt16, toUInt16OrDefault, toUInt16OrNull, toUInt16OrZero, toUInt256, toUInt256OrDefault, toUInt256OrNull
toUInt256OrZero, toUInt32, toUInt32OrDefault, toUInt32OrNull, toUInt32OrZero, toUInt64, toUInt64OrDefault, toUInt64OrNull, toUInt64OrZero, toUInt8
toUInt8OrDefault, toUInt8OrNull, toUInt8OrZero, toUnixTimestamp, toUUID, toUUIDOrDefault, toUUIDOrNull, toUUIDOrZero, toWeek, toYear
trim, trimBoth, trimLeft, trimRight, trunc, truncate, tuple, tupleElement, uniq, uniqCombined
uniqExact, upper, upperUTF8, varPop, varSamp, yesterday

Response ceilings

A statement that names no LIMIT is capped at 10,000 rows: that LIMIT is appended before the query runs, so a large answer comes back as one page rather than a stream. To read the rest, page with LIMIT/OFFSET and an ORDER BY. A statement whose own LIMIT asks for more than 10,000 rows is refused with LIMIT_TOO_HIGH before it runs. When using UNION, every top-level branch must carry its own LIMIT clause of 10,000 rows or fewer, or the query is refused with LIMIT_REQUIRED_PER_BRANCH. A result whose body exceeds about 8,000,000 bytes is refused outright with lwql_result_too_large. Select fewer columns or a smaller LIMIT to bring it under the cap.

​Endpoints

​Authentication

​Errors

​Queryable views

​How the catalog is built

​Observability: traces, spans, logs

​Coding agents

​Experiments & evaluations

​Metrics & rollups

​Gateway & billing

​Governance & audit

​Legacy

​Entities & reference data

​Application tables (PostgreSQL)

​Tables intentionally left out

​Supported functions

​Response ceilings