Self-hosted LangSmith changelog

Subscribe: Our changelog includes an RSS feed that can integrate with Slack, email, Discord bots like Readybot or RSS Feeds to Discord Bot, and other subscription tools.

Self-hosted LangSmith is an add-on to the Enterprise plan designed for our largest, most security-conscious customers. For more details, refer to Pricing. Contact our sales team if you want to get a license key to trial LangSmith in your environment.

​

2026-05-06

self-hosted

langsmith-0.15.0-rc.6

• Fixed “Enabled” column header that was being cut off in the Automations tab.
• Fixed Polly chat button icon getting stuck to the cursor when clicked near the button’s bounding box.
• Fixed a bug where creating or editing files in the file browser during an active “needs approval” interrupt caused the interrupt to disappear.
• Audit Logs: Self-hosted now uses go queue by default.
• Enforced LLM Auth Proxy URL allowlist for Polly.
• Composite feedback columns in the experiment view displayed as flat columns with sub-score visibility toggles in the column menu.
• Evaluator API responses now included a created_by field showing who created each evaluator, and the evaluators table UI displayed a “Created By” column.
• Migrated to add a fleet-admin:read permission.
• Auth interrupt dialog now showed two options: authenticate with all provider scopes (primary) or just the requested scope (secondary).
• Salesforce tool integration was gated behind a feature flag and hidden by default until enabled per-tenant.
• Evaluators table now supported filtering by feedback key and resource (project/dataset).
• Added an experimental async subagents feature allowing subagent threads to not block fleet agent executions.
• Fixed infinite scroll in data tables requiring a scroll-up-and-back-down to load subsequent pages.
• Added a feedback banner on the evaluators page to collect user feedback on the reusable evaluators launch.
• Fixed a bug where SSO users with multiple provider records received 403 errors when their identity provider session lacked SAML metadata.
• Fixed evaluator filter badge icons showing as black/white instead of blue in light/dark mode.
• Oversized feedback fields were now gracefully truncated or dropped instead of rejecting the entire feedback entry.
• Polly now read and respected the active filter on the runs page, merging new conditions rather than replacing user-set filters.
• Fixed evaluators table column dropdown showing “feedback_keys” instead of “Feedback Key”.
• Fixed Insights clustering jobs failing on retry with UniqueViolationError by making insert operations idempotent.
• Removed the “Re-run from here” button from the Agent Builder editor chat interface.
• Evaluators created from the playground now correctly tracked who created them.
• Feedback key filter on evaluators table now showed a searchable dropdown of existing keys instead of a freeform text input.
• Added instance admin endpoint (POST /v1/platform/ops/backfills/restart) to restart backfill jobs and introduced INSTANCE_ADMIN_EMAILS env var (replaces deprecated USAGE_EXPORT_ADMIN_EMAILS).
• Improved handling of custom log formats when streaming logs from the LangSmith Deployment data plane.
• Fixed a JS error when viewing experiment results in the onboarding quickstart flow.
• Added application tag support to evaluators.
• Fixed missing icons and display names for OAuth providers on non-standard deployments.
• Fixed a false “used in other resources” warning when editing a reused evaluator from the only resource it was attached to.
• The experiment comparison delta stream endpoint now supported per-session filters, enabling the comparison table to scope feedback deltas to a filtered subset of runs.
• Fixed Microsoft OAuth repeatedly prompting for consent on every connection attempt even after admin consent was granted.
• Fixed agent clone button not appearing for users with edit access to a shared agent.
• Enabled granular rollup and cleanup jobs that showed daily granularity for granular trace usage.
• Chat messages now scrolled to the top of the viewport when sent, matching the UX of ChatGPT and Claude.
• Evaluator pages now respected application tags consistently, including evaluators surfaced through tagged datasets and tracing projects.
• Google Calendar tools now accepted an optional calendar_id parameter to read or manage other users’ calendars within the same Google Workspace organization.
• Insights “Run now” failures (e.g., invalid API key) now surfaced as an error toast instead of silently disappearing.
• Fixed Microsoft integration accounts always showing “Additional permissions needed to lookup email” even after reauthorizing.
• The evaluator sample-data selector now only showed datasets and tracing projects in the currently selected application.
• Fixed a self-hosted issue where changing Fleet access controls could fail with a 405 error.
• Fixed a bug where users with “can run” permissions on an agent could modify the agent’s files through the hub backend.
• Enabled SSRF protection on all Playground touchpoints. Disabled by default in self-hosted environments.
• Added X (Twitter) integration with BYO OAuth — 17 tools for posting, searching, monitoring mentions, managing follows, and more. Existing API-key-based Twitter tools are preserved but deprecated.
• Added a public Go endpoint for fetching comparison view runs on shared datasets, enabling the shared dataset comparison table to use the Go backend when the experiments_use_smithdb flag is on.
• Clicking a workspace agent card now opened the chat view instead of the editor.
• Fixed multiple-resource filter on the Evaluators page to use AND logic (only show evaluators attached to all selected resources).
• Agents could now have custom icons and colors, selectable from the editor header.
• Fixed a bug where SCIM user updates from IDPs with different email casing were incorrectly rejected as email change attempts.
• Added fleet-admin:read permission to gate the new Fleet Admin section.
• The batch trace messages endpoint (POST /v2/traces/messages) added feedback_stats per trace, eliminating the need to also query the runs endpoint for feedback data.
• Improved log signal-to-noise ratio in the Deployments UI (filtered out logs from an internal server process).
• Improved playground lag on keystroke caused by per-keystroke re-renders of the entire prompt editor component tree.
• Agentify banner in the agent builder chat could again be dismissed with an X button.
• Fixed “View source run” on evaluator traces showing an Invalid trace ID error for dataset-scoped evaluators.
• Improved Agent Builder chat streaming by migrating it to the new React SDK and normalizing streamed LangGraph message handling.
• Markdown tables in Fleet chat rendered with rounded corners, styled headers, and copy/download-as-CSV buttons.
• “View source run” on evaluator traces now worked for dataset-scoped evaluators (previously it produced an Invalid trace ID page).
• Replaced the pre-run Python fetch script with direct CLI calls driven by the agent. Requires langsmith-cli ≥ v0.2.17.
• Cleaner styling and responsive behavior on the issue detail page, with a link-out to GitHub for diff file paths in the Proposed Fix section.
• Running sandbox snapshots now preserved live filesystem changes when restoring a new sandbox.
• Self-hosted: enabled ABAC and audit-logs.
• Sandbox uploads larger than 10 MiB now streamed through the router instead of failing with Failed to read request body.
• Cron schedule triggers now supported attaching skills via the same / menu used in chat, so agents run specific skills on each trigger.
• Fixed: annotation queue run rules no longer write runs to queues in a different workspace.
• Issues-board “Create Evaluator” now opened the same Configure Evaluator pane as the rest of the app, with the issue’s suggested name and prompt prefilled.
• Introduced new inline filter design.
• Fixed evaluator detail page traces/runs toggle reverting to “Traces” when a row is clicked.
• Fleet agents could now generate HTML slide decks that render natively in the in-app presentation viewer, no setup required.
• Issues-board-ao prompt now guided the agent toward fewer, broader subagent calls — screening inline on small scans, consolidating investigation subagents across related categories — reducing cost on large scans.
• GitHub App integration for agent builder — connected GitHub accounts via App installations for repo access.
• Fixed: the “Run a skill” browser in cron schedule dialogs now lets you pick a skill.
• Fixed a bug where newly created tracing projects could be missing from dashboard project selection when Go sessions routing was enabled.
• Slide presentation mode for HTML files is now available for all FLEET users.
• Rolled back the issues-board-ao Phase 1b prompt changes that caused the agent to undersample traces.
• Fixed LLM evaluators getting stuck with llm_evaluator.commit_hash_or_tag not found on save when the stored commit reference was a short commit-hash prefix.
• Evaluator prompt handles no longer included the parent dataset or session name, so moving an evaluator between containers did not leave a stale reference in its prompt handle.
• Showed a clear message when a user without workspaces:manage clicked Install GitHub App, instead of failing silently.
• Issues board now supported BYOK Anthropic API key — configured your own key via the settings pane.
• Allowed workspace members to manage their own GitHub App connection (disconnect, switch actor preference) without requiring workspaces:manage.
• Issues-board-ao Phase 1b screening now used the new trace messages trajectory (input/output preview + step summary) as the first read per trace, only opening the full conversation when trajectory-level signals were ambiguous, reducing screening cost without changing coverage.
• Fixed issue update API to accept name and description fields, allowing the CLI --title and --description flags to work correctly.
• Fixed service account API keys listing extra workspaces the parent service account could reach.
• Subagent cards now rendered at the point they were invoked in the trace view instead of being grouped after all tool calls.
• Optimized sampled run rule queries for self-hosted customers.
• Fixed layout of the few-shot corrections config on the evaluator page so the template description read as one flowing sentence instead of breaking into stacked columns.
• Fixed Playground model chip not updating after changing provider in Model Configuration.
• Fixed agent-builder chat freezing and repeated “Maximum update depth exceeded” errors when multiple sub-agents streamed in parallel on long threads.
• Fixed an issue where reloading the threads page with a selected thread did not correctly scroll the tree to the correct turn and did not show the correct turn number in the trace/turn view.
• Optimized run facet queries for self-hosted users via pre-aggregation.
• Memory updates showed semantic labels with distinct icons, and the approval card supported per-file collapsible diffs with an “Always approve” option.
• Correcting a feedback wouldn’t update the chip, just the popover. This fix updated both.
• User-flagged traces in projects with an issues agent were now automatically added to a project-specific and tenant-global annotation queue for review.
• Google Drive and Google Meet integration cards now appeared on the Agent Builder integrations page.
• Fixed smithdb ingestion stream protocol error (h2 protocol error: error reading a body from connection) caused by abrupt stream cancellation in asynq workers.
• Added Thread Id (if present) to the trace id click to copy tooltip.
• Prevented dataset-attached evaluators from being saved with thread grouping in the frontend.
• Fixed a 500 error on example creation.

Download the Helm chart: langsmith-0.15.0-rc.6.tgz

​

2026-05-05

self-hosted

langsmith-0.15.0-rc.5

• Fixed the “Enabled” column header being cut off in the Automations tab.
• Fixed the Polly chat button icon getting stuck to the cursor when clicking near the button’s bounding box.
• Fixed a bug where creating or editing files in the file browser during an active “needs approval” interrupt would cause the interrupt to disappear.
• Audit Logs: Self-hosted now uses go queue by default.
• Enforced LLM Auth Proxy URL allowlist for Polly.
• Composite feedback columns in the experiment view now displayed as flat columns with sub-score visibility toggles in the column menu.
• Evaluator API responses now included a created_by field showing who created each evaluator, and the evaluators table UI displayed a “Created By” column.
• Migration added a fleet-admin:read permission.
• Auth interrupt dialog now showed two options: authenticate with all provider scopes (primary) or just the requested scope (secondary).
• Salesforce tool integration is now gated behind a feature flag and hidden by default until enabled per-tenant.
• Evaluators table now supported filtering by feedback key and resource (project/dataset).
• Added an experimental async sub-agent feature that allows sub-agent threads to not block fleet agent executions.
• Fixed infinite scroll in data tables requiring a scroll-up-and-back-down to load subsequent pages.
• Added a feedback banner on the evaluators page to collect user feedback on the reusable evaluators launch.
• Fixed a bug where SSO users with multiple provider records received 403 errors when their identity provider session lacked SAML metadata.
• Fixed evaluator filter badge icons showing as black/white instead of blue in light/dark mode.
• Oversized feedback fields are now gracefully truncated or dropped instead of rejecting the entire feedback entry.
• Polly now read and respected the active filter on the runs page, merging new conditions rather than replacing user-set filters.
• Fixed the evaluators table column dropdown showing “feedback_keys” instead of “Feedback Key”.
• Fixed Insights clustering jobs failing on retry with UniqueViolationError by making insert operations idempotent.
• ABAC: Added remaining /sessions/{session_id}/ endpoints.
• Removed the “Re-run from here” button from the Agent Builder editor chat interface.
• Evaluators created from the playground now correctly tracked who created them.
• Feedback key filter on evaluators table now showed a searchable dropdown of existing keys instead of a freeform text input.
• Added instance admin endpoint (POST /v1/platform/ops/backfills/restart) to restart backfill jobs. Introduced INSTANCE_ADMIN_EMAILS env var (replacing deprecated USAGE_EXPORT_ADMIN_EMAILS).
• Improved handling of custom log formats when streaming logs from the LangSmith Deployment data plane.
• Fixed a JS error when viewing experiment results in the onboarding quickstart flow.
• Added application tag support to evaluators.
• Fixed missing icons and display names for OAuth providers on non-standard deployments.
• Fixed a false “used in other resources” warning that appeared when editing a reused evaluator from the only resource it was attached to.
• The experiment comparison delta stream endpoint now supports per-session filters, enabling the comparison table to scope feedback deltas to a filtered subset of runs.
• Fixed Microsoft OAuth repeatedly prompting for consent on every connection attempt even after admin consent was granted.
• Fixed agent clone button not appearing for users with edit access to a shared agent.
• Enabled granular rollup and cleanup jobs that show daily granularity for granular trace usage.
• Chat messages now scroll to the top of the viewport when sent, matching the UX of ChatGPT and Claude.
• Evaluator pages now respect application tags consistently, including evaluators surfaced through tagged datasets and tracing projects.
• Google Calendar tools now accept an optional calendar_id parameter to read or manage other users’ calendars within the same Google Workspace organization.
• Insights “Run now” failures (e.g., invalid API key) now surface as an error toast instead of silently disappearing.
• Fixed Microsoft integration accounts always showing “Additional permissions needed to lookup email” even after reauthorizing.
• The evaluator sample-data selector now only shows datasets and tracing projects in the currently selected application.
• Fixed a self-hosted issue where changing Fleet access controls could fail with a 405 error.
• Fixed a bug where users with “can run” permissions on an agent could modify the agent’s files through the hub backend.
• Enabled SSRF protection on all Playground touchpoints, disabled by default in self-hosted environments.
• Added X (Twitter) integration with BYO OAuth — 17 tools for posting, searching, monitoring mentions, managing follows, and more. Existing API-key-based Twitter tools are preserved but deprecated.
• Added a public Go endpoint for fetching comparison view runs on shared datasets, enabling the shared dataset comparison table to use the Go backend when the experiments_use_smithdb flag is on.
• Clicking a workspace agent card now opens the chat view instead of the editor.
• Fixed multiple-resource filter on the Evaluators page to use AND logic, showing only evaluators attached to all selected resources.
• Agents can now have custom icons and colors, selectable from the editor header.
• Fixed a bug where SCIM user updates from IDPs with different email casing were incorrectly rejected as email change attempts.
• Added fleet-admin:read permission to gate the new Fleet Admin section.
• Fixed an error when creating custom code evaluators from a dataset that use perform_eval(run, example) with two arguments.
• CSV export for the experiment comparison view now routes through the Go backend when the experiments_use_smithdb flag is enabled, streaming up to 10,000 rows with per-session and feedback key columns.
• Fixed workspace-agent navigation so read-only users can reach the editor and clone shared agents, and added hover-panel action buttons for agent cards.
• Agent Builder: HTML file previews now detect slides and offer a presentation mode with thumbnail navigation and fullscreen viewing.
• Added projects:increase-trace-tier and projects:decrease-trace-tier permissions replacing projects:update-retention for granular control over project trace retention changes.
• Unified files sidebar is now enabled for all tenants in Agent Builder chat.
• CSV export for experiments now includes per-annotator feedback detail (score, value, comment, user) when the “include annotator detail” option is selected.
• Filters in progress on the tracing and threads pages are now preserved when clicking away from the filter popover and reopening it.
• Hamburger menu is now hidden when inbox history is open.
• Hardened built-in integrations access controls so ABAC-allowed tools stay available and unauthorized tool execution is blocked.
• Fixed CSV file upload when creating a dataset in self-hosted and split data plane deployments.
• Fleet usage dashboard now tracks agent names on run metadata for display alongside agent IDs.
• Fleet usage API now returns agent names alongside agent IDs.
• Issue board linked traces now show a short summary of why each trace is evidence for the issue.
• Fixed a bug where numeric metadata values were silently dropped during ingestion for large metadata values when blob storage was enabled, causing metadata filtering to return no results for affected runs.
• Clicking the back arrow on evaluator details now returns to the tracing project or dataset you came from, instead of always going to the global evaluators list.
• Fixed a bug where hiding the Outputs column had no effect on single-experiment views without saved column configuration.
• Context canceled errors on run query endpoints are now classified as 499 (client closed) instead of 500, reducing false alerts.
• Users can now view and assign resource tags to integrations and MCP servers in Settings.
• Issues board proposed fix now renders diff blocks with GitHub-style line highlighting instead of plain text.
• Issues agent now requires at least two example traces before creating an issue and will look further back in history to find corroborating evidence when only one trace is found.
• Issues board suggested evaluators now include both Python and JavaScript versions, and switching languages in the evaluator editor preserves the suggested code.
• New GET /v1/platform/orgs/current/members endpoint returning paginated org members (active and pending) with per-workspace roles and glob filtering.
• Fleet usage dashboard now displays agent names alongside agent IDs.
• Code evaluators that return {"score": value} or {"value": value} (with optional “comment”) without a named key now display the evaluator name as their feedback key instead of “score” or “value.”
• Added a setup-guide banner in the Salesforce integration dialog to help users prepare for the admin install step before connecting.
• Fixed a bug where rows with no feedback results in the experiment comparison table would show a loading spinner indefinitely instead of showing “No feedback.”
• Fixed the workspace integrations access-control message to reference the correct workspaces:manage permission.
• Fixed evaluator configuration pane back arrow showing prebuilt evaluators instead of closing.
• Improved issue board UI with better empty states, loading skeletons, consistent timestamps, and visual polish.
• POST /workspaces/current/members now requires role_id for RBAC-enabled organizations. Requests without it return 400 instead of defaulting to WORKSPACE_ADMIN.
• Issue board issues now always include a suggested evaluator.
• Users can now flag traces for issues agent review directly from the trace detail view.
• After forking an evaluator, you are now redirected to the new evaluator’s page with a confirmation toast.
• Added filtering (by priority, status, tags) and sorting (by severity, last updated, created) to the issue board.
• Audit logs now capture dataset download events (download_dataset) separately from generic dataset reads, applying to all export formats (CSV, JSONL, OpenAI, OpenAI FT).
• Fixed duplicate tracing projects appearing in the projects list.
• Salesforce tool integration is now available to all tenants (no longer gated behind a feature flag).
• Fixed dataset experiment LLM calls failing when the customer’s auth proxy requires a non-empty JWT sub claim.
• Fixed a bug where the Hallucination prebuilt evaluator would crash with a KeyError when its prompt template contained variables not covered by the variable mapping.
• Traces flagged for issues agent review are now automatically added to a project-specific annotation queue for team review and test harness creation.
• GitHub links in the datasets table now support right-click to open in a new tab or copy the link address.
• Issues now record an open_pr event when a user opens a PR, enabling full PR lifecycle tracking in the issue activity timeline.
• Members API (GET /v1/platform/orgs/current/members) now includes linked_login_methods (auth provider info) and scim_groups (SCIM group memberships) on each member.
• ABAC for runs:create on the POST /runs endpoint.
• Skills Library can now be served from the template API behind the fleet_template_api feature flag. No user-visible behavior change on either side of the flag — the flag is OFF by default.
• Polly now uses the built-in time range picker for time-based queries instead of embedding absolute timestamps in the filter string.
• Agent Templates page now loads templates from the new template API and auto-installs any required skills when creating an agent from a template — both from the templates page directly and from agents generated through the GPA chat. Behind feature flag during rollout.
• Enforce directional trace tier permissions: users need projects:increase-trace-tier to raise retention and projects:decrease-trace-tier to lower it.
• Dataset description tooltips in the table now only appear when the text is actually truncated, and show formatted GitHub links when they do.
• Dataset detail page description tooltips now render GitHub links with a clickable icon, matching the table view.
• Bulk export destinations now support config_kwargs_s3 for configuring S3 addressing style, fixing compatibility with S3-compatible services that require virtual-hosted addressing (e.g., Volcengine TOS).
• Annotation queues list view now supports sorting by name, total runs, updated at, and created at columns.
• Enterprise SSO users can now log in from the main login page by entering their email, without needing their organization’s SSO URL.
• ABAC: runs:create for POST /runs/batch.
• The batch trace messages endpoint (POST /v2/traces/messages) now returns feedback_stats per trace, eliminating the need to also query the runs endpoint for feedback data.
• Improved log signal-to-noise ratio in the Deployments UI (filters out logs from an internal server process).
• Improves playground lag on keystroke caused by per-keystroke re-renders of the entire prompt editor component tree.
• Agentify banner in agent builder chat can again be dismissed with an X button.
• Fixed “View source run” on evaluator traces showing an Invalid trace ID error for dataset-scoped evaluators.
• Improved Agent Builder chat streaming by migrating it to the new React SDK and normalizing streamed LangGraph message handling.
• Markdown tables in Fleet chat now render with rounded corners, styled headers, and copy/download-as-CSV buttons.
• “View source run” on evaluator traces now works for dataset-scoped evaluators (previously it produced an Invalid trace ID page).
• Replaced the pre-run Python fetch script with direct CLI calls driven by the agent. Requires LangSmith-CLI ≥ v0.2.17.
• Cleaner styling and responsive behavior on the issue detail page, with a link-out to GitHub for diff file paths in the Proposed Fix section.
• Running sandbox snapshots now preserve live filesystem changes when restoring a new sandbox.
• Sandbox uploads larger than 10 MiB now stream through the router instead of failing with Failed to read request body.
• Cron schedule triggers now support attaching skills via the same / menu used in chat, so agents run specific skills on each trigger.
• Annotation queue run rules no longer write runs to queues in a different workspace.
• Issues Board “Create Evaluator” now opens the same Configure Evaluator pane as the rest of the app, with the issue’s suggested name and prompt prefilled.
• Introduced new inline filter design.
• Fix evaluator detail page traces/runs toggle reverting to “Traces” when a row is clicked.
• Fleet agents can now generate HTML slide decks that render natively in the in-app presentation viewer, no setup required.
• Issues-board-ao prompt now guides the agent toward fewer, broader sub-agent calls — screening inline on small scans, consolidating investigation sub-agents across related categories, reducing cost on large scans.
• GitHub App integration for Agent Builder — connect GitHub accounts via App installations for repo access.
• The “Run a skill” browser in cron schedule dialogs now lets you actually pick a skill.
• Fixed a bug where newly created tracing projects could be missing from dashboard project selection when Go sessions routing was enabled.
• Slide presentation mode for HTML files is now available for all FLEET users.
• Fixed LLM evaluators getting stuck with llm_evaluator.commit_hash_or_tag not found on save when the stored memory reference is a short commit-hash prefix.
• Evaluator prompt handles no longer include the parent dataset or session name, so moving an evaluator between containers does not leave a stale reference in its prompt handle.
• Showed a clear message when a user without workspaces:manage clicks Install GitHub App, instead of failing silently.
• Issues board now supports BYOK Anthropic API key — configure your own key via the settings pane.
• Allow workspace members to manage their own GitHub App connection (disconnect, switch actor preference) without requiring workspaces:manage.
• Issues-board-ao Phase 1b screening now uses the new trace messages trajectory (input/output preview + step summary) as the first read per trace, only opening the full conversation when trajectory-level signals are ambiguous, reducing screening cost without changing coverage.
• Fixed issue update API to accept name and description fields, allowing the CLI --title and --description flags to work correctly.
• Fixed service account API keys listing extra workspaces the parent service account could reach.
• Sub-agent cards now render at the point they were invoked in the trace view instead of being grouped after all tool calls.
• Optimized sampled run rule queries for self-hosted customers.
• Fixed layout of the few-shot corrections config on the evaluator page so the template description reads as one flowing sentence instead of breaking into stacked columns.
• Fixed Playground model chip not updating after changing provider in Model Configuration.
• Fixed agent-builder chat freezing and repeated “Maximum update depth exceeded” errors when multiple sub-agents stream in parallel on long threads.
• Fixed an issue where reloading the threads page with a selected thread did not correctly scroll the tree to the correct turn and did not show the correct turn number in the trace/turn view.
• Optimized run facet queries for self-hosted users via pre-aggregation.
• Memory updates show semantic labels with distinct icons, and the approval card supports per-file collapsible diffs with an “Always approve” option.
• Correcting feedback now updates both the chip and the popover.
• User-flagged traces in projects with an issues agent are now automatically added to a project-specific and tenant-global annotation queue for review.
• Fixed smithdb ingestion stream protocol error (h2 protocol error: error reading a body from connection) caused by abrupt stream cancellation in asynq workers.
• Added Thread Id (if present) to the trace ID click-to-copy tooltip.
• Prevented dataset-attached evaluators from being saved with thread grouping in the frontend.
• Fixed a 500 error on example creation.
• Fixed a bug where issues board suggested Python code evaluators failed with AttributeError due to incorrect attribute-style access on run dicts.

Download the Helm chart: langsmith-0.15.0-rc.5.tgz

​

2026-05-04

self-hosted

langsmith-0.15.0-rc.4

• Added support for new UI features including a redesigned evaluator settings section and issue board interface.
• Improved performance and UI responsiveness across various components including the evaluator interfaces and tracing projects.
• Fixed several UI bugs affecting the display and behavior of feedback, evaluator tabs, tracing identifiers, amongst others.
• Added new integrations including support for Google workspace tools and Anthropic APIs within the agent builder.
• Enhanced security and authentication measures with features like audit logging for sensitive data and improved SSO login functionality.
• Introduced new API endpoints to support features like org-based exclusion for time window enforcement and managing self-hosted licenses.
• Enabled performance improvements through optimizations in session syncing and experiments sync processes.
• Added functionality to track consumption and spending within fleet projects, including an interactive usage dashboard with CSV export capabilities.
• Released updates to allow custom, tenant-specific API keys and OAuth flows for platform integrations and execution contexts.

Download the Helm chart: langsmith-0.15.0-rc.4.tgz

​

2026-05-01

self-hosted

langsmith-0.14.5

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.14.5.tgz

​

2026-04-30

self-hosted

langsmith-0.15.0-rc.3

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.15.0-rc.3.tgz

​

2026-04-30

self-hosted

langsmith-0.14.4

• Fixed an issue in the playground where the environment was being polluted by an older version of a dependency.

Download the Helm chart: langsmith-0.14.4.tgz

​

2026-04-29

self-hosted

langsmith-0.14.3

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.14.3.tgz

​

2026-04-24

self-hosted

langsmith-0.15.0-rc.2

• Improved the ‘Enabled’ column in the automations table to prevent header truncation.
• Fixed the “polly” button icon from sticking to the cursor on click.
• Made the MCP server filter dropdown scrollable in the fleet.
• Added sorting to the evaluator’s table with ‘created’ and ‘updated at’ fields.
• Implemented fallback to LANGSMITH_AUTH_ENDPOINT for workspace in “polly”.
• Fixed file creation from clearing the active interrupt state in the fleet.
• Added URL allowlist enforcement for JWT injection in “polly”.
• Added user cost table and agent/user toggle in the fleet.
• Added new fleet-admin:read permission.
• Improved loading performance by removing unnecessary joins from session stats queries.
• Added analytics tracking for template/card selection in the frontend.
• Added timestamps and improved layout for issue board UI.
• Fixed bug causing the evaluator reuse playground to malfunction.
• Added PostgreSQL mTLS support for migrations.
• Added support for mCP servers with composable databases.
• Enhanced security by handling non-SSO login for users with multiple SSO provider records.
• Made evaluator source-run link resilient against invalid IDs.
• Added evaluator details page UX improvements.
• Added mobile-friendly login and installable PWA support.
• Improved overall UX in the evaluator tab with a carousel of suggested evaluators.
• Enabled audit logging for sensitive data access endpoints.
• Added session/dataset activity tracking in Redis for run rules to SmithDB.
• Added tracing project and usage dashboard for fleet projects.
• Added “create agent manually” button to the navbar in the Agent Builder section.
• Fixed rendering issue of tool calls in message view.
• Improved session sync performance to prevent idle sessions.
• Added new model support (Opus 4.7, GLM5, and minimax 2.5) for agent evaluation.
• Improved evaluator fork and usage UX in the frontend.
• Reduced unnecessary re-renders in the playground to improve performance.
• Enabled ‘wrap experiment quickstart’ with ComparisonStoreProvider to address UI bugs.
• Introduced new endpoint to update self-hosted licenses.
• Optimized idle session processing with performance improvements.
• Added audit logging for all remaining Go write endpoints.
• Enhanced “byoc” support with new verification mechanisms and separation of Python service routes.

Download the Helm chart: langsmith-0.15.0-rc.2.tgz

​

2026-04-24

self-hosted

langsmith-0.15.0-rc.1

• Fixed truncation issue by widening the ‘Enabled’ column in the automations table for better header visibility.
• Updated details view header for improved user experience.
• Improved performance by removing dead run_stats_facets join from session stats queries.
• Fixed MCP server filter dropdown to make it scrollable.
• Added ‘user cost’ table and toggle for agent/user view in the fleet.
• Enhanced security by adding URL allowlist enforcement for JWT injection.
• Added ability to sort evaluators by creation and update time in the backend.
• Added ‘Feedback Key’ filtering in the evaluators table for more precise searches.
• Showed back button on the full-page trace view to enhance navigation.
• Fixed audit-logs and various performance improvements for lower latency.
• Enhanced UI by showing ‘Feedback Key’ in evaluator column dropdown.
• Added session insights, views, metadata, and dashboard endpoints to improve data accessibility.
• Fixed file creation issues to prevent active interrupt state clearance.
• Provided async support for agents in the fleet to improve reliability.
• Fixed issues with agent cloning that previously caused flow issues.
• Added spend limit enforcement and the ability to track usage, enhancing cost management features.
• Made more efficient use of resources with new skill memory-store mirror updates in fleets.
• Improved evaluator reuse UX for users with better management of evaluator actions on issue generation.
• Enhanced security features by preventing sub-agents from triggering unauthorized actions.
• Optimized memory and resource management in the agent builder chat.
• Improved evaluator trace detail navigation by preserving search model in URLs.
• Upgraded per-environment favicon colors for clarity in staging and dev environments.
• Performance improvements in session sync reducing resource usage.
• Added default agent name support in usage dashboard for clarity in report generation.
• Made UI improvements to evaluator details for a smoother experience.
• Enabled auto-wake and auto-stop for Sandbox environments to save resources.
• Integrated tracing tool functionality in issue creation for better context and reliability.
• Added “create agent manually” button to navigation for easier agent management.
• Enhanced memory management tools UI for better approval process visualization.
• Fixed tool usage table and improved its performance for a better UX.
• Enabled auditing for sensitive data access endpoints for enhanced security compliance.
• Improved tracing project name display in usage dashboard for better project clarity.
• Introduced keyboard shortcuts in the editor page for rapid interaction.
• Set default MSP cron schedule as standard to reduce manual setup effort.
• Added integration user flow with prompt messages for smooth operation and understanding.
• Fixed default tracing project selection to Fleet to prevent inconsistencies.

Download the Helm chart: langsmith-0.15.0-rc.1.tgz

​

2026-04-20

self-hosted

langsmith-0.14.2

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.14.2.tgz

​

2026-04-20

self-hosted

langsmith-0.14.1

• Fixed an issue where the system looked back further to determine Redis eligibility, improving reliability in self-hosted environments.

Download the Helm chart: langsmith-0.14.1.tgz

​

2026-04-19

self-hosted

langsmith-0.14.0

LangSmith Self-Hosted v0.14 brings Polly (our AI assistant for traces and runs) to self-hosted, takes ABAC and audit logs GA (on by default), and enables the LLM Auth Proxy by default with URL allowlisting and richer JWT claims. Admins get unified model configurations shared across Agent Builder, Polly, Insights, Playground, and Evaluators, and fine-grained Prompt Owners for locking down who can promote or delete individual prompts. Evaluators gain multi-modal support and workspaces can now set cost alerts on tracing projects. Playground model support expands (Anthropic via Vertex AI, custom Azure models, Bedrock inference profiles, Gemini 3.1 Pro, GPT-5.3 / 5.4, Baseten + GLM-5), and new agent tools and triggers land for Google Sheets & Docs, Outlook, Teams, and Salesforce SOQL. On the infrastructure side, v0.14 adds GCS Workload Identity support for blob storage, Valkey as a drop-in Redis replacement, and a pre-upgrade migration hook for safer rollouts.

Follow the upgrade instructions to get access to everything. To book time with LangChain support for your upgrade, contact [e-mail].

Breaking changes

• Fixed an issue where host-backend wasn’t picking up commonEnv. This may result in duplicate environment variables that need to be removed.

Infrastructure changes

• Migrations now run as a Pre-upgrade hook prior to image versions rolling out. This will prevent issues when migrations fail.
• GCS Workload Identity support — authenticate to GCS blob storage using cloud-native workload identity instead of long-lived credentials.
• Valkey support — Valkey can now be used as a drop-in replacement for Redis.

New features

• Polly on self-hosted — our AI assistant for understanding traces, runs, and evaluator feedback is now available in self-hosted.
• ABAC and audit logs GA — Attribute-Based Access Control and audit logs are enabled by default for self-hosted deployments.
• LLM Auth Proxy on by default — URL allowlist prevents credential forwarding to unintended hosts, and JWTs now carry organization_name and workspace_name claims.
• Unified model configurations — Agent Builder, Polly, Insights, Playground, and Evaluators now share a single set of model configs, with workspace-admin controls over model access across all AI features.
• Prompt Owners — designate a specific group of users with fine-grained permission to promote or delete individual prompts, without granting broader org access.
• Multi-modal evaluators — pass attachments and base64 content (images, audio, PDFs) directly into evaluators.
• Cost alerts on tracing projects — set alerts on tracing project-level costs alongside existing LangSmith alerts.
• Expanded playground model support — Anthropic via Vertex AI, custom Azure models, Bedrock inference profiles and configurable base URLs, Gemini 3.1 Pro, GPT-5.3 / 5.4 (now default), and Baseten + GLM-5.
• New agent tools and triggers — Google Sheets and Docs, Outlook mail and calendar, Microsoft Teams, Salesforce SOQL, Gmail OAuth v2 with refresh tokens, and an Outlook Trigger.
• Insights enhancements — scheduled Insights reports, categories trending over time, full feedback comments in analysis, and a lower minimum job interval (6h → 1h).
• Annotation and review upgrades — required reviewers per queue, pairwise queues that honor reviewer_access_mode, an “Assigned to me” filter, per-annotator CSV export, and bulk table actions.
• Prompt Hub and tool registry — commit tag search, model select in template creation, a workspace-scoped tool registry API, and a private registry UI.
• Evaluator workflow improvements — prebuilt LLM evaluators use strict structured outputs by default, evaluators support tagging and reuse, retries no longer lose scores, and a new API runs playground experiments programmatically.
• Custom iframe output renderer — drag-to-resize HTML chart outputs in experiments and trace views.
• Thread and inbox UX — auto-generated thread titles, redesigned run details in threads, session-level feedback stats, keyboard shortcuts, and filtering out internal helper threads.

Admin changes

• Granular usage reporting — granular billable usage APIs that allow you to retrieve detailed trace usage data broken down by workspace, project, user, or API key.

​

2026-04-17

self-hosted

langsmith-0.13.43

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.43.tgz

​

2026-04-14

self-hosted

langsmith-0.13.42

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.42.tgz

​

2026-04-14

self-hosted

langsmith-0.13.41

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.41.tgz

​

2026-04-09

self-hosted

langsmith-0.13.40

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.40.tgz

​

2026-04-07

self-hosted

langsmith-0.13.39

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.39.tgz

​

2026-04-03

self-hosted

langsmith-0.13.38

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.38.tgz

​

2026-04-01

self-hosted

langsmith-0.13.37

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.37.tgz

​

2026-03-30

self-hosted

langsmith-0.13.36

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.36.tgz

​

2026-03-27

self-hosted

langsmith-0.13.35

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.35.tgz

​

2026-03-27

self-hosted

langsmith-0.13.34

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.34.tgz

​

2026-03-27

self-hosted

langsmith-0.13.33

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.33.tgz

​

2026-03-27

self-hosted

langsmith-0.13.32

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.32.tgz

​

2026-03-23

self-hosted

langsmith-0.13.31

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.31.tgz

​

2026-03-23

self-hosted

langsmith-0.13.30

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.30.tgz

​

2026-03-21

self-hosted

langsmith-0.13.29

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.29.tgz

​

2026-03-21

self-hosted

langsmith-0.13.28

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.28.tgz

​

2026-03-18

self-hosted

langsmith-0.13.27

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.27.tgz

​

2026-03-13

self-hosted

langsmith-0.13.26

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.26.tgz

​

2026-03-12

self-hosted

langsmith-0.13.25

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.25.tgz

​

2026-03-10

self-hosted

langsmith-0.13.24

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.24.tgz

​

2026-03-07

self-hosted

langsmith-0.13.23

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.23.tgz

​

2026-03-06

self-hosted

langsmith-0.13.21

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.21.tgz

​

2026-03-06

self-hosted

langsmith-0.13.20

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.20.tgz

​

2026-03-06

self-hosted

langsmith-0.13.19

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.19.tgz

​

2026-03-05

self-hosted

langsmith-0.13.18

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.18.tgz

​

2026-03-03

self-hosted

langsmith-0.13.17

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.17.tgz

​

2026-02-26

self-hosted

langsmith-0.13.16

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.16.tgz

​

2026-02-26

self-hosted

langsmith-0.13.15

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.15.tgz

​

2026-02-24

self-hosted

langsmith-0.13.14

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.14.tgz

​

2026-02-14

self-hosted

langsmith-0.13.13

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.13.tgz

​

2026-02-12

self-hosted

langsmith-0.13.12

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.12.tgz

​

2026-02-12

self-hosted

langsmith-0.13.11

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.11.tgz

​

2026-02-10

self-hosted

langsmith-0.13.10

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.10.tgz

​

2026-02-09

self-hosted

langsmith-0.13.9

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.9.tgz

​

2026-02-06

self-hosted

langsmith-0.13.7

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.7.tgz

​

2026-02-05

self-hosted

langsmith-0.13.6

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.6.tgz

​

2026-02-05

self-hosted

langsmith-0.13.5

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.5.tgz

​

2026-02-04

self-hosted

langsmith-0.13.4

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.4.tgz

​

2026-01-26

self-hosted

langsmith-0.13.3

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.3.tgz

​

2026-01-21

self-hosted

langsmith-0.13.2

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.2.tgz

​

2026-01-16

self-hosted

langsmith-0.13.1

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.1.tgz

​

2026-01-16

self-hosted

langsmith-0.13.0

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.13.0.tgz

​

2026-01-12

self-hosted

langsmith-0.12.37

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.37.tgz

​

2026-01-09

self-hosted

langsmith-0.12.36

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.36.tgz

​

2026-01-08

self-hosted

langsmith-0.12.35

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.35.tgz

​

2025-12-26

self-hosted

langsmith-0.12.34

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.34.tgz

​

2025-12-20

self-hosted

langsmith-0.12.33

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.33.tgz

​

2025-12-12

self-hosted

langsmith-0.12.32

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.32.tgz

​

2025-12-11

self-hosted

langsmith-0.12.31

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.31.tgz

​

2025-12-09

self-hosted

langsmith-0.12.30

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.30.tgz

​

2025-12-08

self-hosted

langsmith-0.12.29

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.29.tgz

​

2025-12-05

self-hosted

langsmith-0.12.28

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.28.tgz

​

2025-12-04

self-hosted

langsmith-0.12.27

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.27.tgz

​

2025-12-02

self-hosted

langsmith-0.8.30

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.8.30.tgz

​

2025-12-01

self-hosted

langsmith-0.12.25

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.25.tgz

​

2025-11-27

self-hosted

langsmith-0.12.24

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.24.tgz

​

2025-11-26

self-hosted

langsmith-0.12.23

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.23.tgz

​

2025-11-26

self-hosted

langsmith-0.12.22

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.22.tgz

​

2025-11-26

self-hosted

langsmith-0.12.21

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.21.tgz

​

2025-11-24

self-hosted

langsmith-0.12.20

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.20.tgz

​

2025-11-24

self-hosted

langsmith-0.12.19

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.19.tgz

​

2025-11-20

self-hosted

langsmith-0.12.18

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.18.tgz

​

2025-11-19

self-hosted

langsmith-0.12.17

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.17.tgz

​

2025-11-19

self-hosted

langsmith-0.12.16

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.16.tgz

​

2025-11-17

self-hosted

langsmith-0.12.15

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.15.tgz

​

2025-11-17

self-hosted

langsmith-0.12.14

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.14.tgz

​

2025-11-13

self-hosted

langsmith-0.12.13

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.13.tgz

​

2025-11-13

self-hosted

langsmith-0.12.12

• Internal improvements and maintenance updates

Download the Helm chart: langsmith-0.12.12.tgz

Additional Helm chart releases are available in the langchain-ai/helm GitHub repository.

---

Connect these docs to Claude, VSCode, and more via MCP for real-time answers.

Edit this page on GitHub or file an issue.