Self-hosting releases

​

Data plane v2.0.0

• Topics for automated log insights
• Sandboxes for agent evals
• Evaluation parameters
• Log indexing and full-text search
• Trace-level scorers
• Trace-level filters in charts
• estimated_cost() SQL function
• Image rendering security controls
• Single span filters with aggregations
• LIMIT with aggregations

​

Terraform AWS module releases

• v5.1.0 — No data plane version change. Required for the no-PostgreSQL mode step of the v2.0 upgrade.
  • skip_pg_for_brainstore_objects now defaults to "all". Existing deployments upgrading from any v4.x or v5.0.0 module version will have no-PostgreSQL mode activated automatically unless this variable is explicitly overridden. This change is one-way. Rolling back requires downtime. Pin skip_pg_for_brainstore_objects = "" before applying if you are not ready to enable this.
  • Production example cleaned up: explicit skip_pg_for_brainstore_objects = "" block removed, reflecting the new default.
• v5.0.0 — Ships data plane v2.0.0 image tags. Required for the data plane v2.0 upgrade.
  • brainstore_wal_footer_version now defaults to "v3". Existing deployments upgrading from v4.x must ensure all Brainstore nodes are running data plane v2.0 before applying, or must explicitly set brainstore_wal_footer_version = "" to preserve prior behavior. Applying WAL v3 while any node is still on a v1.x image causes read failures.
  • Sandbox example updated: skip_pg_for_brainstore_objects defaults to "all" in the sandbox config; explicit brainstore_wal_footer_version pin removed.
• v4.5.0 — No data plane version change.
  • Added skip_pg_for_brainstore_objects variable to enable no-PostgreSQL mode. Setting any non-empty value automatically fans out to BRAINSTORE_ASYNC_SCORING_OBJECTS and BRAINSTORE_LOG_AUTOMATIONS_OBJECTS on Brainstore EC2 nodes. Required before enabling Topics.
  • BRAINSTORE_WAL_USE_EFFICIENT_FORMAT is now automatically enabled when either brainstore_wal_footer_version or skip_pg_for_brainstore_objects is set.
  • When kms_key_arn is configured, all managed S3 buckets now enforce blocked_encryption_types = ["NONE"], preventing unencrypted uploads. This appears as a policy addition in terraform plan when upgrading.
  • Added x-bt-use-gateway to the AI Proxy Lambda function URL CORS allowed headers, enabling browser clients to send this header without a preflight rejection.
  • Added API Gateway routes for Topics API, storage-settings endpoints (/brainstore/storage-settings/object/{object_id}, .../backfill, .../status), and additional automation endpoints (/brainstore/automation/get-object-cursors, .../execute, .../upsert-object-cursor). Self-hosted deployments must upgrade to v4.5.0 to use these features — routes absent from the API Gateway spec return 403, which browsers report as a CORS error.

​

Terraform GCP module releases

• v1.5.1 — No data plane version change.
  • The API GCS bucket now includes a 1-day object lifecycle rule on the brainstore-cache/ prefix. Brainstore writes ephemeral cache data here during query execution; objects are automatically deleted after 1 day. No action is required on upgrade.
  • Storage paths in the API bucket are code-bundle/ and brainstore-cache/. The previously documented response/ path has been replaced by brainstore-cache/.
• v1.5.0 — No data plane version change. Required before upgrading to data plane v2.0.
  • The Workload Identity binding now covers both the brainstore and braintrust-api Kubernetes service accounts, enabling API pods to access GCS directly. Required for data plane v2.0. Note: google_service_account_iam_binding is authoritative — any IAM members added manually outside Terraform for roles/iam.workloadIdentityUser on this service account will be removed on the next terraform apply.
  • SQL users now set deletion_policy = "ABANDON" when postgres_deletion_protection = false, preventing Terraform errors on destroy in environments where Cloud SQL user deletion is not supported.
• v1.4.0 — No data plane version change.
  • Added brainstore_impersonation_targets variable to grant the Brainstore Kubernetes service account the ability to impersonate one or more Google Cloud service accounts (same or cross-project). Useful for Brainstore deployments that need to access GCS buckets or other GCP resources restricted to a specific service account identity. See Service account impersonation for configuration details.

​

Terraform Azure module releases

• v1.1.0 — No data plane version change. Required before upgrading to data plane v2.0.
  • Added existing_resource_group_name variable to deploy into a pre-existing Azure Resource Group rather than having the module create one. Includes a state migration (moved.tf) for users upgrading from v1.0.0 — the resource group Terraform address changed from azurerm_resource_group.main to azurerm_resource_group.main[0]. Confirm your plan shows the resource as moved, not destroyed.
  • Added custom_tags variable to apply user-defined tags to all provisioned resources. All resources also receive an automatic BraintrustDeploymentName tag.
  • Added aks_brainstore_pool_min_count and aks_services_pool_min_count variables (default: 2) to configure AKS autoscaler minimum node counts. Previously hardcoded to 2.
  • Added helm_object_storage_values output that returns all Azure object storage configuration values in the shape expected by the Braintrust Helm chart.

​

Helm chart releases

• v6.1.0 — GCP/Azure only. No data plane version change. Required for the no-PostgreSQL mode step of the v2.0 upgrade.
  • skipPgForBrainstoreObjects now defaults to "all". Existing deployments upgrading from chart 6.0.0 without an explicit override will have no-PostgreSQL mode activated automatically. This change is one-way. Rolling back requires downtime. Set skipPgForBrainstoreObjects: "" in helm-values.yaml before upgrading if you are not ready to enable this.
  • brainstoreWalFooterVersion now defaults to "v3". Deployments still running Data Plane 1.x images must pin this to "v1" or "" before upgrading.
• v6.0.0 — GCP/Azure only. Ships data plane v2.0.0 image tags. Required for the data plane v2.0 upgrade.
• v5.1.0 — GCP/Azure only. No data plane version change.
  • Added brainstoreWalFooterVersion chart value to configure the Brainstore WAL footer format. Defaults to "" (unset). Set to "v1" as part of the v2.0 upgrade sequence. Do not set to "v3" until after successfully deploying data plane v2.0.
  • Added skipPgForBrainstoreObjects chart value to enable no-PostgreSQL mode. Setting any non-empty value automatically fans out to BRAINSTORE_ASYNC_SCORING_OBJECTS and BRAINSTORE_LOG_AUTOMATIONS_OBJECTS on reader, writer, and fastreader pods. Required before enabling Topics.
  • BRAINSTORE_WAL_USE_EFFICIENT_FORMAT is now automatically enabled on the API pod when either brainstoreWalFooterVersion or skipPgForBrainstoreObjects is set.
  • BRAINSTORE_RESPONSE_CACHE_URI and BRAINSTORE_CODE_BUNDLE_URI are now auto-derived from the objectStorage configuration in values.yaml. No explicit configuration required. If you previously set these via extraEnvVars, remove those overrides to avoid conflicts.
  • ConfigMap checksum annotations added to all deployments (API, reader, writer, fastreader). Pods now automatically restart when their ConfigMap changes, so configuration-only Helm upgrades no longer require a manual kubectl rollout restart.

​

Terraform AWS module releases

• v4.4.0 — No data plane version change.
  • Added brainstore_wal_footer_version variable to control the Brainstore WAL footer format. Defaults to "" (unset). Set to "v1" as part of the v2.0 upgrade sequence. Do not set to "v3" until after successfully deploying data plane v2.0.
  • Added api_handler_memory_limit and ai_proxy_memory_limit variables to configure Lambda memory allocation for the API Handler and AI Proxy. Both default to 10240 MB (no change from previous behavior).
  • Brainstore now validates instance type local NVMe storage at terraform plan time. Instance types without local storage produce a descriptive plan-time error. Compatible families: c8gd, c5d, m5d, i3, i4i.
  • Added POST /v1/{object_type}/{id}/restore and POST /v1/{object_type}/{id}/restore/preview API Gateway routes. Self-hosted deployments must upgrade to v4.4.0 to use object restore features.
  • Added sandbox deployment example (examples/braintrust-data-plane-sandbox) for low-cost infrastructure testing with easy teardown.
• v4.3.0 — No data plane version change.
  • Lambda environment variables for automation_cron and billing_cron are now encrypted at rest using the deployment KMS key. Applied automatically on terraform apply; no configuration changes required.
  • The managed VPC’s public subnet no longer auto-assigns public IPv4 addresses on instance launch (map_public_ip_on_launch = false). Workloads that previously relied on auto-assigned public IPs in this subnet should migrate to Elastic IPs.
• v4.2.0 — No data plane version change.
  • All three S3 buckets (Brainstore, code-bundle, Lambda responses) now enforce HTTPS-only access via DenyInsecureTransport bucket policies. When upgrading from v4.1.0, these appear as new aws_s3_bucket_policy additions in terraform plan — non-destructive.
  • Removed unnecessary 0.0.0.0/0 egress rules from RDS and ElastiCache security groups. When upgrading from v4.1.0, these appear as deletions in terraform plan — expected and safe.
  • Added database_authorized_security_groups and redis_authorized_security_groups variables for supplying pre-existing security group IDs for RDS and ElastiCache (advanced use cases).
  • Fixed EKS IRSA trust policy for api_handler_role and brainstore_role: the :sub condition now uses StringLike to support wildcard namespace matching. Resolves IRSA authentication failures when eks_namespace is unset.

​

Terraform GCP module releases

• v1.3.0 — No data plane version change.
  • The Braintrust API service account is now granted roles/iam.serviceAccountTokenCreator on itself, enabling it to generate short-lived tokens and signed GCS URLs without exporting static credentials.
  • The Brainstore service account is now granted roles/storage.objectAdmin and roles/storage.legacyBucketReader on the API GCS bucket (in addition to its existing access to the Brainstore bucket). This is required for Brainstore to access shared objects in the API bucket.

​

Terraform Azure module releases

• v1.0.0 — No data plane version change. This is the first production-ready release of the Azure Terraform module, introducing Front Door ingress and a restructured AKS node pool configuration. See the Azure deployment guide for setup instructions.
  This release contains breaking changes for users upgrading from v0.9.0:

  • Node pool variables renamed: aks_user_pool_vm_size and aks_user_pool_max_count have been removed. Replace them with aks_brainstore_pool_vm_size + aks_brainstore_pool_max_count (for Brainstore nodes) and aks_services_pool_vm_size + aks_services_pool_max_count (for application nodes). The node pools will be destroyed and recreated during the upgrade — plan for workload migration.
  • azurerm provider upgrade required: The module now requires azurerm ~> 4.0 and Terraform >= 1.10.0. Run terraform init -upgrade before applying.
  • brainstore_license_key is now required: The module will fail at plan time if this variable is not supplied.
  • Added Azure Front Door Premium ingress via Private Link Service. Deployment follows a staged workflow: deploy base infrastructure and Helm chart first, then enable Front Door. The private link service requires manual approval in the Azure Portal.
  • AKS node pools are now split into a dedicated brainstore pool (requires a local-SSD VM SKU, e.g. Standard_D32ds_v6) and a services pool (general-purpose, no local SSD required). The Azure Container Storage extension (acstor) is automatically installed to configure RAID0 on the Brainstore pool.
  • Added a dedicated Braintrust workload identity with scoped Key Vault and Storage RBAC permissions. The workload_identity_client_id output is required for Helm chart configuration.

​

Helm chart releases

• v5.0.1 — GCP/Azure only. No data plane version change.
  • Brainstore resources blocks for reader, writer, and fastreader now accept any valid Kubernetes resource spec. Limits can be omitted entirely or set to {} to remove resource constraints. See Brainstore resource configuration for details.
  • Fixed AWS SDK compatibility with Google Cloud Storage when using S3-compatible mode (enableGcsAuth: false). The chart now automatically sets AWS_REQUEST_CHECKSUM_CALCULATION and AWS_RESPONSE_CHECKSUM_VALIDATION to WHEN_REQUIRED. No configuration changes required.
• v5.0.0 — GCP/Azure only. No data plane version change.
  • Fast readers are now enabled by default. Fast readers are isolated Brainstore nodes that handle common queries powering the Braintrust UI, keeping it responsive while resource-intensive ad-hoc queries are routed to standard reader nodes. See Brainstore fast reader configuration for details and hardware requirements for sizing guidance.
  • If you have customized brainstore.reader settings in values.yaml, mirror those customizations to the new brainstore.fastreader block before upgrading.
  • Readiness probe defaults for the API and Brainstore have been updated to use the same path (/) and port as liveness probes, with extended initialDelaySeconds (API: 30s, Brainstore: 60s) and periodSeconds (10s).
    If you have custom brainstore.readinessProbe overrides pointing to /status, remove them before upgrading. The /status readiness endpoint has a bug where it never recovers after a failure, which can permanently mark Brainstore nodes as not ready. The new default probe path (/) does not have this issue.
• v4.1.0 — GCP/Azure only. No data plane version change.
  • Added support for custom sidecars on Braintrust pods.

​

Data plane v1.1.32

• Improved ephemeral WAL memory management
• Added support for large payloads at ingest time
• Added support for routing BTQL queries to a separate fast reader pool for improved query performance
• SQL improvements: added HAVING clause support, ordered phrase queries; fixed json_extract argument validation and ternary operator handling for nulls
• Added support for pydantic-style JSON schema values in tools
• MCP server now exposes SQL query tool instead of BTQL, and adds SDK installation resource

​

Data plane v1.1.31

• Improve performance of S3 export (this is an ongoing effort, and we are continuing to make improvements)
• Added native GCS authentication for object storage
• SQL improvements: added PIVOT/UNPIVOT syntax, array identifier, string similarity functions in group by, negative array indexing; fixed MATCH syntax, .user identifier parsing, and positional aggregates for sort
• Improved async batch invoke performance with configurable batch sizes
• Improved proxy error responses with proper status codes
• Fixed attachment parsing for root-level attachments

​

Helm v3.1.0

• Added support for GCP Workload Identity on the API service, enabling native GCS authentication without HMAC keys. See GCS authentication options.

​

Data plane v1.1.30

• Added support for parsing prompts using Nunjucks/Jinja syntax in the playground
• Added date_trunc and json_extract functions to BTQL and SQL
• Added support for alerting on prompt environment updates
• Added support for File type in newer Vercel AI SDK versions
• Improved tags functionality
• Improved async scoring performance with batched payload requests
• Improved WAL compression and ephemeral data handling
• Fixed MCP OAuth flow when initial auth server discovery returns 401
• Fixed implicit aliasing support in BTQL queries
• Auto-correct double quotes to string literals in BTQL parser
• Automatically convert Google inline base64 attachments to Braintrust attachments

​

Data plane v1.1.29

• Added local shared read cache for Brainstore writers
• Added support for SQL as a frontend to BTQL
• Added braintrust.origin attribute support for OTEL logs
• Added capability to run API, realtime, and brainstore containers as a non-root user “braintrust”
• Fixed Azure Blob metadata key formatting (hyphen → underscore)

​

Data plane v1.1.28

• Fixed enforcement of BRAINSTORE_QUERY_TIMEOUT_SECONDS parameter.
• Added support for enforcing access to only certain data plane URL patterns.
• Added custom annotation view support.
• Enable adding MCP servers to prompts. After authenticating with OAuth, these can be used in the UI in Prompt Chat, Playgrounds, and Experiments.

​

Helm v3.0.0 (breaking change for Azure)

• Added Azure Container Storage driver support and updated Azure-specific configuration.

​

Data plane v1.1.27

• Added ability to share log and experiment traces publicly
• A built-in rate limit for API queries. You can enable this by setting RATELIMIT_BTQL_DEFAULT, which controls how many BTQL queries are allowed per object per minute, per object.
• Better query cancellation for long-running aggregation queries.
• Add comments, audit_data, and _async_scoring_state to the BTQL schema. You can now query for these fields alongside existing ones.

​

Data plane v1.1.26

• Faster performance for indexing (compaction and merging)
• Improved I/O utilization for queries that read a long time horizon
• Fix a longstanding deadlock bug that can occur with high query volume
• Many small improvements to query performance

​

Data plane v1.1.25

• Faster real-time queries (no “excluded docs” in most cases)
• Fix thinking for Mistral models
• Significantly faster indexing for large payloads
• Fix a bug with floating point division in queries
• Fix MCP OAuth flow for self-hosted deployments
• More iterations in hosted tools (100)
• Improve performance for high selectivity filter queries
• Fix gemini tool calls that included $defs and $refs in the schema
• Fix bug in /feedback endpoint when updating non-root spans

​

Data plane v1.1.22

• Added ability to create and edit custom charts in the monitor dashboard
• Added support for more Grok models and improved model refresh handling in /invoke endpoint
• Added support for IN clause in BTQL queries
• Improved processing of pydantic-ai OpenTelemetry spans with tool names in span names and proper input/output field mapping
• Added OpenAI Agents logs formatter for better span rendering in the UI
• Added retention support for Postgres WAL and object WAL (write-ahead logs)
• Add S3 lifecycle policies to reclaim additional space from bucket
• Added authentication support for remote evaluation endpoints
• Improved ability to fetch all datasets efficiently
• New MAX_LIMIT_FOR_QUERIES parameter to set the maximum allowable limit for BTQL queries. Larger result sets can still be queried through pagination

​

Data plane v1.1.21

• Process pydantic-ai OTel spans
• AI proxy now supports temperature > 1 for models which allow it
• Preview of data retention on logs, datasets, and experiments

​

Data plane v1.1.20

• Brainstore vacuum is enabled by default. This will reclaim space from object storage. As a bonus, vacuum also cleans up more data (segment-level write-ahead logs)
• AI proxy now dynamically fetches updates to the model registry
• Performance improvements to summary, IS NOT NULL, and != NULL queries
• Handle cancelled BTQL queries earlier and optimize schema inference queries
• Added a REST API for managing service tokens. See docs
• Support custom columns on the experiments page
• Aggregate custom metrics and include more built-in agent metrics in experiments and logs
• Preview of data retention on logs. You can define a per-project policy on logs which will be deleted on a schedule and no longer available in the UI and API

​

Data plane v1.1.19

• Add support for GPT-5 models
• OTel tracing support for Google Agent Development Kit
• OTel support for deleting fields
• Fix binder error handling for malformed BTQL queries
• Enable environment tags on prompt versions

​

Data plane v1.1.18

• Improve performance for non-selective searches. Eg make foo != 'bar' faster
• Improve performance for score filters. Eg make scores.correctness = 0 faster
• Improve group by performance. This should make the monitor page and project summary page significantly faster
• Add syntax for explicit casting. You can now use explicit casting functions to cast data to any datatype. e.g. to_number(input.foo), to_datetime(input.foo), etc
• Fix ILIKE queries on nested json: ILIKE queries previously returned incorrect results on nested json objects. ILIKE now works as expected for all json objects
• Improve backfill performance. New objects should get picked up faster
• Improve compaction latency. Indexing should kick in much faster, and in particular, this means data gets indexed a lot faster
• Improved support for OTel mappings, including the new GenAI Agent conventions and strands framework
• Add Gemini 2.5 Flash-Lite GA, GPT-OSS models on several providers, and Claude Opus 4.1

​

Data plane v1.1.15

• Add ability to run scorers as tasks in the playground
• You can now use object storage, instead of Redis, as a locks manager
• Support async python in inline code functions
• Don’t re-trigger online scoring on existing traces if only metadata fields like tags change

​

Data plane v1.1.14

• Switch the default query shape from traces to spans in the API. This means that btql queries will now return 1 row per span, rather than per trace. This change also applies to the REST API
• Service tokens with scoped, user-independent credentials for system integrations
• Fix a bug where very large experiments (run through the API) would drop spans if they could not flush data fast enough
• Support built-in OTel metrics (contact your account team for more details)
• New parallel backfiller improves performance of loading data into Brainstore across many projects

​

Data plane v1.1.13

• Fix support for COALESCE with variadic arguments
• Add option to select logs for online scoring with a BTQL filter
• Add ability to test online scoring configuration on existing logs
• Mmap based indexing optimization enabled by default for Brainstore

​

Data plane v1.1.11

• Add support for LLaMa 4 Scout for Cerebras
• Turn on index validation (which enables self-healing failed compactions) in the Cloudformation by default

​

Data plane v1.1.7

• Improve performance of error count queries in Brainstore
• Automatically heal segments that fail to compact
• Add support for new models including o3 pro
• Improve error messages for LLM-originated errors in the proxy

​

Data plane v1.1.6

• Patch a bug in 1.1.5 related to the realtime_state field in the API response

​

Data plane v1.1.5

• Default query timeout in Brainstore is now 32 seconds
• Auto-recompact segments which have been rendered unusable due to an S3-related issue
• Gemini 2.5 models

​

Data plane v1.1.4

• Optimize “Activity” (audit log) queries, which reduces the query workload on Postgres for large traces (even if you are using Brainstore)
• Automatically convert base64 payloads to attachments in the data plane
• Improve AI proxy errors for status codes 401->409
• Increase real-time query memory limit to 10GB in Brainstore

​

Data plane v0.0.65

• Improve error messages when trying to insert invalid unicode
• Backend support for appending messages

​

SDK version 0.0.197

• Fix a bug in init_function in the Python SDK which prevented the input argument from being passed to the function correctly when it was used as a scorer
• Support setting description and summarizeScores/summarize_scores in Eval(...)

• Many improvements to the playground experience:
  • Fixed many crashes and infinite loading spinner states
  • Improved performance across large datasets
  • Better support for running single rows for the first time
  • Fixed re-ordering prompts
  • Fixed adding and removing dataset rows
  • You can now re-run specific prompts for individual cells and columns
• You can now do “does not contain” filters for tags in experiments and datasets
• When you invoke() a function, inline base64 payloads will be automatically logged as attachments
• Add a strict mode to evals and functions which allows you to fail test cases when a variable is not present in a prompt
• Add Fireworks’ DeepSeek V3 03-24 and DeepSeek R1 (Basic), along with Qwen QwQ 32B in Fireworks and Together.ai, to the playground and AI proxy
• Fix bug that prevented Databricks custom provider form from being submitted without toggling authentication types
• Unify Vertex AI, Azure, and Databricks custom provider authentication inputs
• Add Llama 4 Maverick and Llama 4 Scout models to Together.ai, Fireworks, and Groq providers in the playground and AI proxy
• Add Mistral Saba and Qwen QwQ 32B models to the Groq provider in the playground and AI proxy
• Add Gemini 2.5 Pro Experimental and Gemini 2.0 Flash Thinking Mode models to the Vertex provider in the playground and AI proxy
• Add OpenAI’s o1-pro model to the playground and AI proxy
• Support OpenAI Responses API in the AI proxy
• Add support for the Gemini 2.5 Pro Experimental model in the playground and AI proxy
• Option to disable the experiment comparison auto-select behavior
• Add support for Databricks custom provider as a default cloud provider in the playground and AI proxy
• Allow supplying a base API URL for Mistral custom providers in the playground and AI proxy
• Support pushed code bundles larger than 50MB
• The OTEL endpoint now understands structured output calls from the Vercel AI SDK
• Added support for concat, lower, and upper string functions in BTQL
• Correctly propagate Bedrock streaming errors through the AI proxy and playground
• Online scoring supports sampling rates with decimal precision
• Added support for OpenAI GPT-4o Search Preview and GPT-4o mini Search Preview in the playground and AI proxy
• Add support for making Anthropic and Google-format requests to corresponding models in the AI proxy
• Fix bug in model provider key modal that prevents submitting a Vertex provider with an empty base URL
• Add column menu in grid layout with sort and visibility options
• Enable logging the origin field through the REST API
• Add support for “image” pdfs in the AI proxy
• Fix issue in which code function executions could hang indefinitely
• Add support for custom base URLs for Vertex AI providers
• Add dataset column to experiments table
• Add python3.13 support to user-defined functions
• Fix bug that prevented calling Python functions from the new unified playground

​

Data plane v0.0.64

• Brainstore is now set as the default storage option
• Improved backfilling performance and overall database load
• Enabled relaxed search mode for ClickHouse to improve query flexibility
• Added strict mode option to prompts that fails when required template arguments are missing
• Enhanced error reporting for missing functions and eval failures
• Fixed streaming errors that previously resulted in missing cells instead of visible error states
• Abort evaluations on server when stopped from playground
• Added support for external bucket attachments
• Improved handling of large base64 images by converting them to attachments
• Fixed proper handling of UTF-8 characters in attachment filenames
• Added the ability to set telemetry URL through admin settings

​

Data plane v0.0.63

• Support for Claude 3.7 Sonnet, Gemini 2.0 Flash-Lite, and several other models in the proxy
• Stability and performance improvements for ETL processes
• A new /status endpoint to check the health of Braintrust services

​

Data plane v0.0.61

• Upgraded to Node.js 22 in Docker containers

​

Data plane v0.0.60

• Make PG_URL configuration more uniform between nodeJS and python clients

​

Data plane v0.0.59

• Fix permissions bug with updating org-scoped env vars
• Support uploading file attachments
• You can now export OpenTelemetry (OTel) traces to Braintrust

​

Data plane v0.0.56

• Hosted tools are now available in the API
• Environment variables are now supported in the API
• Automatically backfill function_data for prompts created via the API

​

Data plane v0.0.54

• Support for bundled eval uploads
• The PATCH endpoint for prompts now supports updating the slug field
• Don’t fail insertion requests if realtime broadcast fails
• Performance optimizations to filters on scores, metrics, and created fields
• Performance optimizations to filter subfields of metadata and span_attributes

​

Data plane v0.0.53

• The API now supports running custom LLM and code (TypeScript and Python) functions

​

Data plane v0.0.51

• The proxy is now a first-class citizen in the API service, which simplifies deployment
• The proxy is now accessible at https://api.braintrust.dev/v1/proxy
• If you are self-hosting, the proxy is now bundled into the API service