v1.5.12

​

✨ Features

• Vertex AI Files & Batches API - Added support for the Vertex AI Files and Batches endpoints
• Vault Backends for Secrets - Sensitive config fields can now be stored in AWS Secrets Manager, GCP Secret Manager, or HashiCorp Vault as an alternative to AES encryption
• Per-Alias Provider Overrides - Key aliases now support alias-level Azure endpoint/API version/Anthropic version, Bedrock region/ARN, Vertex project/region, and Replicate deployments-endpoint overrides
• MCP Server Library - New browsable MCP server catalog with background sync, search and filters, install sheet, custom entries with soft-delete, and a multi-harness agent connect sheet
• Complexity Router - Route requests by prompt complexity using complexity_tier CEL expressions with a configurable analyzer (config UI, DB, and API included)
• Per-Model Usage in Quota API - The virtual key quota API now reports usage broken down per model
• OTEL HTTP Metrics & Span Filtering - The OTEL connector now emits HTTP-level metrics, and plugin spans can be filtered per connector via plugin_span_filters
• Canonical Model Name in Logs - Added canonical_model_name and alias_model_family columns to logs, and request metadata is now included in object-storage log exports
• Routing Audit Trail - Responses and errors now carry RoutingInfo extra fields with a retry/fallback audit trail from the core routing engine
• key_ids in Provider Config - Providers can be scoped to specific keys via the new key_ids field in the config schema and Helm chart
• Datadog Env Vars in Helm - Added support for DD environment variables in the Helm chart
• Anthropic Fable Compatibility - Added support for Anthropic Fable models, including fast mode pricing fixes

​

🐞 Fixed

• “Allow All” in vk provider config now properly  routes to all allowed models in key configurations
• Postgres Logstore Filters - Fixed metadata filters and pagination total_count for the Postgres logstore (thanks @zbloss!)
• Vertex Embeddings API Key Auth - The Vertex Embedding method now supports API key authentication (thanks @TransactCharlie!)
• Bedrock Cohere Usage - Cohere embed/rerank usage on Bedrock is now filled from the response header (thanks @Alishark14!)
• OpenAI File Upload - Fixed expires_at fields in OpenAI file uploads
• Virtual Key Handling - Generate a UUID when a virtual key is created without an ID, propagate the VK in GenAI file upload sessions, stamp the VK tool allowlist when the include-clients filter is present, and enforce the VK tool-grant boundary on caller-provided x-bf-mcp-include-* headers
• Governance Log Mappings - Fixed teams and customers name mappings on logs, the customer FK column issue, and added a unique-name constraint migration on the customer table
• DeepSeek v4 Reasoning - Fixed max reasoning effort handling for DeepSeek v4
• Gemini Tool Responses - Fixed parts handling in Gemini tool responses
• OpenRouter Cache Control - cache_control blocks are now preserved in OpenRouter chat requests
• Trace Attributes - Refactored tracers to correctly set trace-level attributes
• Provider Config Preservation - Use the in-file provider config when preserving a failed provider config instead of the existing runtime config

• feat: added Vertex AI Files and Batches API support
• feat: added vault backends (aws-secrets-manager, gcp-secret-manager, hashicorp-vault) as alternative to AES encryption for sensitive config fields
• feat: added per-alias provider overrides (Azure endpoint/API version/Anthropic version, Bedrock region/ARN, Vertex project/region, Replicate deployments endpoint) by promoting KeyAliases values to AliasConfig
• feat: added PreRequestHook to the LLMPlugin interface for once-per-request processing
• feat: added RoutingInfo to response/error extra fields with retry and fallback audit trail
• feat: scoped ListModels to a single key when KeyID is set
• feat: added datasheet store to model catalog
• feat: added Anthropic Fable model compatibility
• feat: added plugin span filtering for multiple connectors
• feat: tagged MCP health check requests via BifrostContextKeyMCPHealthCheck
• feat: refactored tracers to correctly set trace-level attributes
• refactor: extracted provider resolution into the modelcatalogresolver plugin
• fix: max reasoning effort handling for DeepSeek v4
• fix: virtual key propagation in GenAI file upload sessions
• fix: fast mode pricing for Anthropic
• fix: drop reasoning effort “none” for Vertex
• fix: added API key auth support to the Vertex Embedding method (thanks @TransactCharlie!)
• fix: fill Cohere embed/rerank usage from the response header on Bedrock (thanks @Alishark14!)
• fix: preserve cache_control blocks in OpenRouter chat requests
• fix: Gemini parts handling in tool responses
• fix: plugin name sanitization for live-loaded plugins for connectors

• feat: added canonical_model_name and alias_model_family columns to logs
• feat: added request metadata to LLM logs in object storage
• feat: added MCP server library (mcp_library table, catalog sync, paginated listing, custom entries with soft-delete)
• feat: added complexity analyzer config DB and API support
• feat: added vault backend support for sensitive config fields
• feat: added live model catalog cache store and per-key config aggregation (keyconfig.Store)
• feat: added unique-name constraint migration on the customer table
• feat: refactored tracers to correctly set trace-level attributes
• fix: metadata filters and pagination total_count for the Postgres logstore (thanks @zbloss!)
• fix: customer FK column issue
• fix: enforce auth on inference in the auth middleware
• fix: fast mode pricing for Anthropic
• chore: removed vault encryption hooks from certain GORM tables in favor of AES-only encryption
• chore: upgraded core to v1.5.19

• chore: implemented no-op PreRequestHook for the new LLMPlugin interface
• chore: upgraded core to v1.5.19 and framework to v1.3.19

• feat: added complexity router with complexity_tier CEL routing and configurable analyzer
• feat: enforce VK tool-grant boundary on caller-provided x-bf-mcp-include-* headers
• fix: stamp virtual key tool allowlist when the include-clients filter is present
• fix: teams and customers name mappings on logs
• fix: complexity router keyword merging and removal via config.json changes
• fix: parse provider-prefixed model strings in pre-request routing
• fix: exempt cache creation requests from model checks
• chore: upgraded core to v1.5.19 and framework to v1.3.19

• chore: implemented no-op PreRequestHook for the new LLMPlugin interface
• chore: upgraded core to v1.5.19 and framework to v1.3.19

• feat: added canonical_model_name and alias_model_family columns to logs
• chore: upgraded core to v1.5.19 and framework to v1.3.19

• chore: implemented no-op PreRequestHook for the new LLMPlugin interface
• chore: upgraded core to v1.5.19 and framework to v1.3.19

• chore: implemented no-op PreRequestHook for the new LLMPlugin interface
• chore: upgraded core to v1.5.19 and framework to v1.3.19

• feat: initial release - provider resolution extracted from core into a standalone plugin, with routing allowlist enforcement and provider-prefixed model parsing

• feat: added HTTP metrics
• feat: added plugin span filtering for multiple connectors
• feat: refactored tracers to correctly set trace-level attributes
• chore: upgraded core to v1.5.19 and framework to v1.3.19

• chore: implemented no-op PreRequestHook for the new LLMPlugin interface
• chore: upgraded core to v1.5.19 and framework to v1.3.19

• chore: implemented no-op PreRequestHook for the new LLMPlugin interface
• chore: upgraded core to v1.5.19 and framework to v1.3.19

• chore: implemented no-op PreRequestHook for the new LLMPlugin interface
• chore: upgraded core to v1.5.19 and framework to v1.3.19