Kaizen Changelog | Kamiwaza Docs

On this page

This document tracks major changes, feature additions, and improvements in the Kaizen agent platform.

Version 1.4.0

Major Features

1. Skills-First Agent Workflows

• Reusable skills: Expanded Kaizen's built-in skill system with richer prebuilt capabilities, including chart, map, PDF, PPTX, and skill-builder workflows.
• Skills over bundles: Continued the shift away from older bundle-centric behavior toward reusable, agent-native skills and defaults.
• Skill visibility: Improved agent creation and conversation UX so active skills and skill configuration are easier to inspect and manage.

2. Context-Aware Document Processing

• Context Service integration: Added first-class support for collections, pipeline jobs, retrieval, and semantic search through Kamiwaza Context Service.
• Ingestion workflows: Agents can submit files for processing, monitor progress, and use retrieved context in downstream work.
• Operational guardrails: Automatic context-search tool injection was temporarily removed while internal authentication issues are stabilized.

3. Better Model, Tool, and Enterprise Integrations

• Kamiwaza model access: Kaizen now centers on Kamiwaza-managed model deployments and custom endpoints instead of direct Anthropic or Claude-specific setup.
• Tool Garden and MCP improvements: Improved Tool Garden MCP discovery, authentication, PAT fallback, host resolution, and error handling.
• Microsoft 365 and SharePoint: Expanded M365 and SharePoint browsing, download handling, schemas, and timeout behavior for more reliable enterprise file access.

4. Workroom-Aware and App Garden-Ready Deployment

• Workroom preservation: Kaizen now preserves workroom context across deployments and related API calls.
• App Garden compatibility: Improved App Garden deployment compatibility, runtime defaults, and route/base-path behavior in packaged deployments.
• Release metadata: Updated extension metadata, preview image handling, and compatibility to target Kamiwaza >=0.10.0.

5. Sandbox and Runtime Controls

• LLM tuning controls: Added configurable timeout, retry, and consumer-hardware defaults for agent model calls.
• Sandbox storage options: Added configurable sandbox storage sizing and persistence controls for workspace durability.
• Runtime hardening: Improved Kubernetes sandbox/runtime support, health checks, and container security context handling.

6. UX and Reliability Improvements

• Conversation experience: Improved approval cards, error toasts, file previews, citations, and conversation reliability under failure conditions.
• File handling: Added support for image attachments and larger uploads, with frontend middleware updates for bigger payloads.
• Workspace polish: Improved agent creation wizard layout, workspace panels, and context-processing progress surfaces.

Technical Improvements

• Security: Hardened error sanitization, upstream error handling, TLS/SSL behavior, and container/runtime defaults.
• Platform resilience: Added platform event logging, HTTP pool improvements, better health endpoints, and more robust SSE/conversation handling.
• Dependency maintenance: Upgraded vulnerable dependencies, refreshed SDK patches, and improved test coverage across context, MCP, runtime-default, and sandbox flows.

Version 1.3.2 vs 0.9.0

Major Features

1. App Garden Integration

• Deployment: Now fully deployable as a Kamiwaza App Garden extension.
• Configuration: Auto-generated docker-compose and metadata for seamless platform integration.
• Streaming: Migrated from WebSockets to Server-Sent Events (SSE) for better compatibility with corporate proxies and HTTP-only environments.

2. Advanced Skills System

• Custom Knowledge: Inject domain expertise into agents.
• Skill Types:
  • Repository Skills: Always-active context (policies, business rules).
  • Knowledge Skills: Triggered by specific keywords.
  • Task Skills: Pre-defined workflows with inputs.

3. MCP (Model Context Protocol) Integration

• External Tools: Connect agents to external APIs and services (Zapier, Google Workspace, Slack).
• Security: Encrypted storage for authentication headers and keys.
• Management: UI for configuring and testing MCP server connections.

4. Enterprise Security & Governance

• Action Confirmation: Configurable policies for tool execution:
  • Never Confirm: Autonomous execution.
  • Confirm Risky: Pauses for high-risk actions (file deletion, network calls).
  • Always Confirm: Manual approval for every step.
• Sandbox Controller: Optional secure mode for strict Docker isolation (Phase 8.5).

5. Enhanced Workspace & File Management

• File Uploads: "Paperclip" button to inject files directly into the agent's workspace.
• File Downloads: Browse and download generated files (reports, code) from the side panel.
• Activity Stream: Real-time "Thinking", "Tool Usage", and "Result" blocks for full visibility into agent logic.

6. UX Improvements

• ChatGPT-Style Sidebar: Time-grouped conversation history (Today, Yesterday, Last 7 Days).
• Inline Confirmation: Approve or reject agent actions directly within the chat stream.
• Stuck Detection: If an agent gets stuck in a loop or repetitive behavior, Kaizen automatically detects this and can help recover, ensuring your agent continues making progress on your tasks.

7. Microsoft 365 Integration

• Cloud Connectors: Connect directly to SharePoint and OneDrive to import documents into the workspace.
• User-Delegated Auth: Secure OAuth flow ensures users only access files they have permission to see.

Technical Improvements

• Backend: Migrated to FastAPI + Python 3.12.
• Frontend: Upgraded to Next.js 15.
• Database: Robust PostgreSQL storage with Alembic migrations for agents, secrets, and conversation history.
• SDK: Updated to OpenHands SDK 1.8.2 with custom SSL patching for internal service communication.

Historical Version Summary