---
title: "Query traces"
method: POST
path: "/api/v2/traces/query"
tags: ["runs"]
---

# Query traces

`POST /api/v2/traces/query`

## Request

- Base URL: `https://api.smith.langchain.com`
- URL: `https://api.smith.langchain.com/api/v2/traces/query`
- Auth: one of:
  - API key in header `X-API-Key` and API key in header `X-Tenant-Id`
  - HTTP bearer and API key in header `X-Tenant-Id`

Returns a paginated list of traces (root runs) for a single tracing project. Each item carries the trace's root run plus optional trace-wide aggregates (`total_tokens`, `total_cost`, `first_token_time`) under `trace_aggregates`, so clients never have to merge by `trace_id`.

Traces are scanned within a `start_time` window: `min_start_time` defaults to 24 hours before the request, `max_start_time` defaults to the request time. Set either explicitly to widen or narrow the window.

Supports filters (`trace_filter`, `tree_filter`), cursor pagination (`cursor`), and field projection (`selects`).

When `ai_search` is set, `Accept: text/event-stream` is required; requests without it return 406. AI search is unavailable on deployments that route queries to the v1 backend and returns 501 there.

Self-hosted deployments require LangSmith `v0.16` or later.

## Headers

- `Accept` string
- `Content-Type` string

## Request body

- QueryQueryTracesRequestBody
  - `ai_search` string: `ai_search` is a plain-language criterion evaluated against the messages from the agent trajectory scoped to the trace. AND-ed with the ordinary filters. Requires semantic filtering enabled for the deployment. Must contain nonempty text of at most 2000 UTF-8 bytes.
  - `cursor` string: `cursor` is the opaque string returned in a previous response's `next_cursor`.
  - `max_feedback_modified_at` string, date-time: `max_feedback_modified_at` is the exclusive upper bound (RFC3339) on feedback `modified_at`, applied to the feedback conditions in `trace_filter` or `tree_filter` (for example eq(feedback_key, "correctness")). Requires at least one feedback condition. Root runs must still fall within the `start_time` window.
  - `max_start_time` string, date-time: `max_start_time` is the exclusive upper bound for the root-run start time scan (RFC3339). Defaults to the request time when omitted.
  - `min_feedback_modified_at` string, date-time: `min_feedback_modified_at` is the inclusive lower bound (RFC3339) on feedback `modified_at`, applied to the feedback conditions in `trace_filter` or `tree_filter` (for example eq(feedback_key, "correctness")). Requires at least one feedback condition. Root runs must still fall within the `start_time` window.
  - `min_start_time` string, date-time: `min_start_time` is the inclusive lower bound for the root-run start time scan (RFC3339). Defaults to 24 hours before the request when omitted.
  - `page_size` integer: `page_size` is the maximum number of traces to return per page. Defaults to 20; must be between 1 and 100 when set.
  - `project_id` string, uuid: `project_id` is the UUID of the tracing project that owns the traces. Required.
  - `selects` QueryRunSelectField[]: `selects` lists which properties to include on each returned trace. Properties listed here are routed to the appropriate sub-object on each item: `total_tokens`, `total_cost`, and `first_token_time` appear under `trace_aggregates`; everything else appears under `root_run`. If omitted, only `id` is returned on `root_run`.
  - `trace_filter` string: `trace_filter` narrows results to traces whose root run matches this LangSmith filter expression. This filter targets root runs only — `is_root = true` is implied. See https://docs.langchain.com/langsmith/trace-query-syntax#filter-query-language for syntax.
  - `trace_ids` string[]: `trace_ids` is an optional fast-path restriction to a known set of trace UUIDs. Equivalent in result to including each UUID in a `trace_filter`, but more efficient at scale.
  - `tree_filter` string: `tree_filter` narrows results to traces containing at least one run anywhere in the run tree (root or descendant) that matches this LangSmith filter expression.

## Response `200`

OK

- QueryQueryTracesResponseBody
  - `items` QueryTrace[]: `items` is the page of traces.
    - `query_metadata` object, nullable: `query_metadata` describes why this trace matched the query, including `sem_filter_score` when a semantic filter is used. Returned automatically, or null when unavailable.
    - `root_run` QueryRunResponse
      - `app_path` string: `app_path` identifies the application code location that produced this run, if recorded.
      - `attachments` QueryRunAttachmentURLs
      - `completion_cost` number: `completion_cost` is estimated USD cost for the completion.
      - `completion_cost_details` QueryRunCompletionCostDetails
        - `raw` object: `raw` maps each category name to its estimated USD cost.
      - `completion_token_details` QueryRunCompletionTokenDetails
        - `raw` object: `raw` maps each category name to its completion-token count.
      - `completion_tokens` integer: `completion_tokens` is the completion-side token count.
      - `dotted_order` string: `dotted_order` is the hierarchical ordering key for trace trees.
      - `end_time` string, date-time: `end_time` is when the run ended (RFC3339 date-time). JSON null if the run has not finished yet.
      - `error` string: `error` is the error message when `status` indicates failure.
      - `error_preview` string: `error_preview` is a truncated plain-text error snippet.
      - `events` QueryRunEvent[]: `events` is the ordered list of run events (for example streaming tokens).
        - `kwargs` object: `kwargs` is the event payload — an opaque JSON object whose shape depends on `name` and on the emitting SDK. For example LangChain emits `{"token": {...}}` for `new_token` events, tool-call start/end details for tool events, and arbitrary user-defined payloads for custom events. Clients should treat `kwargs` as untyped JSON: do not assume specific keys exist for a given `name`, and tolerate additional unknown keys appearing over time.
        - `name` string: `name` is the event kind. Common values emitted by the LangChain/LangSmith tracer SDKs include `"start"`, `"end"`, and `"new_token"`, but applications may emit arbitrary strings for their own instrumentation.
        - `time` string, date-time: `time` is when the event occurred (RFC3339 date-time with millisecond precision).
      - `extra` object: `extra` is additional runtime JSON attached to the run.
      - `feedback_stats` QueryRunFeedbackStats
      - `first_token_time` string, date-time: `first_token_time` is when the first output token was produced (RFC3339 date-time), when recorded for streamed runs.
      - `id` string, uuid: `id` is this run's UUID.
      - `inputs` object: `inputs` is the run input payload (arbitrary JSON object).
      - `inputs_preview` string: `inputs_preview` is a truncated plain-text preview of inputs.
      - `is_in_dataset` boolean: `is_in_dataset` is true when this run is linked to a dataset example.
      - `is_root` boolean: `is_root` is true when this run has no parent (it is the trace root).
      - `last_queued_at` string, date-time: `last_queued_at` is the most recent time this run was added to an annotation queue.
      - `latency_seconds` number: `latency_seconds` is wall-clock duration from start to end in seconds.
      - `ls_user_id` string, uuid: `ls_user_id` identifies the LangSmith user whose credential traced the run. It is absent for runs traced with a service-account API key, which has no associated user.
      - `manifest` object: `manifest` is the serialized configuration of the traced component (for example the model parameters, prompt template, or pipeline definition), when recorded.
      - `metadata` object: `metadata` is arbitrary user-defined JSON metadata.
      - `name` string: `name` is a human-readable label for the run (for example the model name, function name, or step name chosen when the run was traced).
      - `outputs` object: `outputs` is the run output payload (arbitrary JSON object).
      - `outputs_preview` string: `outputs_preview` is a truncated plain-text preview of outputs.
      - `parent_run_ids` string[]: `parent_run_ids` lists ancestor run UUIDs from the trace root down to the direct parent.
      - `price_model_id` string, uuid: `price_model_id` identifies the pricing model UUID used for cost estimates, when recorded.
      - `project_id` string, uuid: `project_id` is the tracing project UUID this run was logged to.
      - `prompt_cost` number: `prompt_cost` is estimated USD cost for the prompt.
      - `prompt_cost_details` QueryRunPromptCostDetails
        - `raw` object: `raw` maps each category name to its estimated USD cost.
      - `prompt_token_details` QueryRunPromptTokenDetails
        - `raw` object: `raw` maps each category name to its prompt-token count.
      - `prompt_tokens` integer: `prompt_tokens` is the prompt-side token count.
      - `query_metadata` object, nullable: `query_metadata` describes this query result, including `sem_filter_score` when a semantic filter is used. Returned automatically by run queries, or null when unavailable.
      - `reference_dataset_id` string, uuid: `reference_dataset_id` is the dataset UUID for the reference example, if any.
      - `reference_example_id` string, uuid: `reference_example_id` is the dataset example UUID this run was compared against, if any.
      - `run_type` 'TOOL' | 'CHAIN' | 'LLM' | 'RETRIEVER' | 'EMBEDDING' | 'PROMPT' | 'PARSER'
      - `share_url` string: `share_url` is the fully-qualified URL of this run's public view, rooted at the deployment's LangSmith app origin (for example `https://smith.langchain.com/public/4f7a1b2c-8d9e-4a0b-9c1d-2e3f4a5b6c7d/r`). It is returned only when `SHARE_URL` is included in `selects`, and only when the run has been explicitly shared; the URL remains stable until the run is unshared. Anyone with this URL can view the run anonymously, so treat it as a secret and do not log it.
      - `start_time` string, date-time: `start_time` is when the run started (RFC3339 date-time).
      - `status` 'SUCCESS' | 'ERROR' | 'PENDING'
      - `tags` string[]: `tags` lists user-defined tags on this run.
      - `thread_evaluation_time` string, date-time: `thread_evaluation_time` is thread-level evaluation timing (RFC3339 date-time), when recorded.
      - `thread_id` string, uuid: `thread_id` is the conversation thread UUID this run belongs to, if any.
      - `total_cost` number: `total_cost` is total estimated USD cost (prompt plus completion).
      - `total_tokens` integer: `total_tokens` is prompt plus completion tokens.
      - `trace_id` string, uuid: `trace_id` is the root trace UUID; for a root run it matches `id`.
    - `trace_aggregates` QueryTraceAggregates
      - `first_token_time` string, date-time: `first_token_time` is when the first output token was produced anywhere in the trace (RFC3339), when recorded.
      - `total_cost` number: `total_cost` is total estimated USD cost across every run in the trace.
      - `total_tokens` integer: `total_tokens` is prompt plus completion tokens summed across every run in the trace.
  - `next_cursor` string: `next_cursor` is the opaque cursor for the next page. Null on the final page.

## Other responses

- `400`: bad request (malformed JSON or invalid parameters)
- `401`: missing or invalid authentication
- `403`: forbidden (insufficient permission)
- `404`: session not found
- `406`: not acceptable (ai_search requires Accept: text/event-stream)
- `422`: unprocessable entity (e.g. invalid UUID)
- `500`: internal server error
- `501`: V2 filter syntax or ai_search is unavailable for this deployment; use legacy function-style filters or set SMITHDB_QUERY_ENABLED=true
- `503`: service unavailable
- `504`: gateway timeout or deadline exceeded

## Changes

- **2026-10-19** `5b873ddde64b`: 2 info
  - added the new optional request property `max_feedback_modified_at`
  - added the new optional request property `min_feedback_modified_at`
- **2026-10-16** `1a8ee9358d81`: 1 info
  - endpoint added

[Operation history](https://skmtc.dev/langchain-ai/apis/langsmith/changes/api/v2/traces/query/post.md)

---

[API](https://skmtc.dev/langchain-ai/apis/langsmith.md) - [All operations](https://skmtc.dev/langchain-ai/apis/langsmith/llms.txt) - [OpenAPI document](https://skmtc.dev/langchain-ai/apis/langsmith/revisions/7a01c01f75e8?raw)