Changelog

All notable changes to the bRRAIn platform are documented here. Releases follow Semantic Versioning.

v2.0.0 — 2026-04-01

bRRAIn 2.0: The Intelligence Update

This major release introduces the Agent framework, hybrid search, and a redesigned security model. It includes breaking changes to the API response format and authentication flow.

Features

Agent Framework: Autonomous agents that continuously enrich memory objects with entity extraction, relationship mapping, deduplication, and summarization. Agents run in the background and can be configured per-workspace.
Hybrid Search: New mode=hybrid parameter on the retrieve endpoint combines semantic vector search with keyword matching for improved recall. The fusion algorithm weights results based on query characteristics.
Customer-Managed Keys (BYOK): Enterprise customers can now bring their own encryption keys using AWS KMS, Google Cloud KMS, Azure Key Vault, or HashiCorp Vault. See the Vaults documentation for setup instructions.
Workspace Sharing: Share memory objects across workspace boundaries using Shared Collections, Memory Links, and Broadcast Memories. All sharing is audited and revocable.
Batch API: New POST /v1/memory/batch endpoint for storing up to 100 memory objects in a single request with atomic semantics.
OpenTelemetry Support: Native distributed tracing with OpenTelemetry. Configure a collector endpoint to receive spans from all eight zones.
Helm Chart: Official Kubernetes Helm chart for production self-hosted deployments with autoscaling, health checks, and pod disruption budgets.

Improvements

Retrieval latency reduced by 40% through query plan optimization and embedding cache warming
Memory store throughput increased to 5,000 objects per second per node
Dashboard redesign with real-time activity feeds and workspace-level analytics
API error responses now include a request_id field for easier debugging
Rate limit headers added to all responses (X-RateLimit-Limit, X-RateLimit-Remaining, X-RateLimit-Reset)

Bug Fixes

Fixed a race condition in the Ingestion Zone that could cause duplicate memory objects under high concurrent load
Fixed pagination cursor encoding that failed for memory objects with special characters in metadata
Fixed an issue where vault deletion did not properly clean up background agent tasks
Fixed timezone handling in created_after and created_before query filters

Breaking Changes

API response format: Error responses now use a nested error object with code, message, request_id, and status fields. Previously, errors returned a flat {"error": "message"} structure.
Authentication header: The X-API-Key header is no longer supported. Use Authorization: Bearer <key> instead. A 6-month migration period was provided in v1.1.0.
Memory object IDs: New objects use the mem_ prefix format. Existing objects retain their old IDs but are aliased to the new format.
Minimum API version: The /v0/ API prefix is no longer available. All clients must use /v1/.

Migration Guide

Update error handling code to parse the new nested error format
Replace X-API-Key headers with Authorization: Bearer headers
Update any hardcoded ID parsing logic to handle the mem_ prefix
Test all integrations against the /v1/ endpoints

v1.1.0 — 2025-12-15

Improved Search and Enterprise Features

Features

Semantic search improvements: New embedding model delivers 25% better relevance scores for domain-specific queries
SAML 2.0 SSO: Enterprise single sign-on support with automatic user provisioning and group-to-workspace mapping
Audit log streaming: Real-time audit log delivery to Splunk, Datadog, and Elastic via webhook or syslog
IP allowlisting: Restrict vault access to specific IP ranges at the vault policy level
Memory object TTL: Set a ttl_seconds field when storing memory objects for automatic archival

Improvements

Search results now include highlights with matched terms wrapped in <em> tags
API key management page shows last-used timestamp and request count
Documentation site launched at docs.brrain.io with full API reference
Go SDK updated to v0.5.0 with context support and connection pooling
Python SDK updated to v0.4.0 with async support

Bug Fixes

Fixed memory leak in the Processing Zone when handling malformed UTF-8 content
Fixed an issue where workspace deletion returned 500 instead of 404 for non-existent workspaces
Fixed CORS preflight handling for custom headers
Fixed rate limiter not resetting correctly at the minute boundary

Deprecations

The X-API-Key header is deprecated in favor of Authorization: Bearer. It will be removed in v2.0.0. Both methods work during the migration period.

v1.0.0 — 2025-09-01

Initial Release

The first generally available release of the bRRAIn platform.

Features

Memory Store and Retrieve: Core API endpoints for storing and retrieving knowledge with semantic search
Vault encryption: AES-256-GCM encryption at rest with platform-managed keys
Workspace isolation: Multi-tenant workspace model with role-based access control (Owner, Admin, Editor, Viewer)
REST API: Full REST API at api.brrain.io/v1 with JSON request/response format
Dashboard: Web-based dashboard for managing vaults, workspaces, API keys, and viewing stored memories
Go SDK: Official Go client library (github.com/Qosil/brrain-go) with full API coverage
Python SDK: Official Python client library (pip install brrain) with sync and async interfaces
Docker deployment: Official Docker image for self-hosted deployments with Docker Compose
Health monitoring: /healthz and /readyz endpoints for infrastructure health checks

Infrastructure

SOC 2 Type II certification completed
Data residency available in US (Virginia, Oregon) and EU (Ireland, Frankfurt)
99.95% uptime SLA for enterprise customers
Prometheus metrics endpoint for operational monitoring

Known Limitations

Maximum memory object size is 1 MB
Rate limit of 60 requests per minute on the free tier
Embedding generation may take up to 30 seconds for large documents
Self-hosted deployments require a license key (free for up to 3 users)