Platform Changelog

​

AI models & documentation

​

Answer API model list

• New models — Added Claude Sonnet 4.6, Claude Opus 4.6, DeepSeek V3.2, and Qwen3 Next 80B A3B to the supported model catalog, with clearer short descriptions for when to use each model.
• Retired from docs — Removed older catalog entries (Claude Sonnet 4, Sonnet 4.5, Opus 4.5, Llama 3.3 70B) from the public model tables so the docs match what we support today.
• Credits — DeepSeek V3.2 is priced at 2 credits per use (see Generate AI Answer and Python SDK).

​

Python SDK docs

• Aligned with API reference — The Python SDK Generate AI Answer page uses the same model table and updated examples (ai_model defaults and samples).

​

Security

• Dependency and security updates — We applied security-focused updates across almost all of our backend functions that ship with a package.json (Node.js services and Lambdas), including dependency refreshes and hardening. We recommend using the latest API keys.

​

New Features

​

Fetch Text Data API

• New GET /documents/fetch-text-data endpoint - Retrieve text and summary chunks from a text-type namespace via the public API
• List stored chunks - Use for display, export, or RAG; response includes statistics, metadata, and up to 100 items per request
• Text namespaces only - Supports namespaces with type === "text"; authenticate with x-api-key

​

Python SDK v1.3.3

• Direct AI without namespace - Direct AI mode no longer requires a namespace, simplifying general AI generation workflows
• Structured response parameter - Define and enforce structured output formats for more reliable integration with downstream applications (see Structured output)
• Improved validation and tests - Better input validation and test coverage for AI generation requests

​

Other

​

Moorcheh Console

• Security updates - Security improvements applied to the Moorcheh Console.
• View data items in namespaces - The console shows up to 100 data items from your namespace. Open a namespace to see its contents at console.moorcheh.ai/namespaces.

​

New Features

​

Pre-signed Upload URL for Large Files

• New /upload-url endpoint - Get a pre-signed S3 URL to upload files directly, bypassing the 10MB API Gateway limit
• Upload files up to 5GB - Support for large file uploads directly to S3
• Automatic processing - Files uploaded via pre-signed URL are automatically processed through the document pipeline

​

Delete File

• New /delete-file endpoint - Delete one or more files from a namespace in document storage (S3)
• Single or multiple files - Use fileName, fileNames, or query fileName; all listed files are deleted in one request
• Per-file results - Response includes status per file (deleted or error)

​

Removed

​

Upload File API documentation

• Upload File endpoint docs removed - The standalone Upload File (/upload-file) API reference page has been removed. Use Get Upload URL for file uploads (supports up to 5GB).

​

New Features

​

Answer Endpoint: Structured Output

• Structured output in response - The answer endpoint now returns structured output for easier parsing and integration with downstream systems
• Predictable response schema - Consistent JSON structure for answers, citations, and metadata

​

Enterprise Package & Licensing

• Enterprise customer package - New package tailored for enterprise customers with advanced features and support
• Terraform-based licensing - Licensing and provisioning for enterprise deployments managed via Terraform for reproducible, infrastructure-as-code workflows

​

Security & Improvements

​

Answer Endpoint

• Max token security - Updated security controls for maximum tokens on the answer endpoint to prevent abuse and ensure resource fairness

​

Integrations

• n8n: AI models in npm registry - Updated AI models available in the n8n npm registry for the Moorcheh node
• moorcheh-mcp: Security and version - Security improvements and version update for the Moorcheh MCP server
• Chat boilerplate: GitHub repo initialization - Chat boilerplate now also initializes a GitHub repository when you create a new project

​

New Features

​

Python SDK Updates

• Async client support - New AsyncMoorchehClient for asynchronous operations with FastAPI, aiohttp, and other async frameworks
• Automatic batching - Automatic batching for document and vector uploads (100 documents per batch)
• File upload method - Upload files directly via upload_file() method supporting PDF, DOCX, XLSX, JSON, TXT, CSV, and Markdown files (10MB max)

​

Console: Playground Summaries & Memory Management

• Persistent chat memory - Each chat conversation maintains its own memory context
• Long conversation support - Moorcheh internally manages memory to maintain full conversation context
• Maximum output tokens - Increased output limit to maximum for comprehensive responses
• Namespace locking per chat - Each conversation locks to its namespace to ensure relevant context without mixing
• Improved document summaries - Enhanced summaries with full document context for better understanding

​

Improvements

​

Console: Context Management

• Isolated conversation memory - Each chat session maintains separate memory to prevent context mixing
• Full conversation understanding - Moorcheh ensures the entire conversation meaning is preserved

​

Improvements

​

Larger File Uploads in Console

• 100 MB upload limit - Upload files up to 100 MB directly from the console
• Flexible formats - Works with PDF, DOCX, XLSX, JSON, TXT, CSV, and Markdown files
• Automatic ingestion - Files are queued for processing and indexing once uploaded
• Improved upload status - Clearer progress and completion states for large files
• Higher AI output - Generative answers now support up to 8,000 output tokens for richer responses

​

Usage Notes

• Upload time depends on network speed; keep the tab open until the upload completes
• Very large documents may take longer to ingest and index after upload
• Upload status now reflects chunked progress for large files

​

UI & Security Updates

• Accessible console icons - Refreshed icon set with better contrast and clarity
• Framework security patches - Updated Next.js and React across moorcheh.ai, console.moorcheh.ai, and the chat boilerplate to include the latest security fixes

​

New Features

​

Upload Files via Public API

• Direct file upload - Upload files directly to your namespace using the public API endpoint
• Multiple file formats supported - PDF, DOCX, XLSX, JSON, TXT, CSV, and Markdown files
• Multipart form data - Simple file upload using standard multipart/form-data encoding
• Automatic processing - Files are automatically queued for ingestion and indexing

​

Supported File Types

​

Quick Start Example

curl -X POST "https://api.moorcheh.ai/v1/namespaces/{namespace-name}/upload-file" \
  -H "x-api-key: YOUR_API_KEY" \
  -F "file=@path/to/your/document.pdf"

​

Important Notes

​

Processing Time

​

File Size Limit

​

Text Namespace Required

​

New Features

​

Document Summaries

• Automatic summary generation - Uploaded documents now include automatic summaries
• Summaries in search results - Summaries appear in search results to provide context
• Batch summary generation - One summary is generated for every 100 chunks of content

​

Upload Text Directly from Console

• No file needed - Paste or type your text content directly in the namespace console
• Faster processing - Text-only uploads are processed more quickly than file uploads
• Automatic chunking and indexing - Text is automatically chunked and indexed for search
• Optional metadata - Add custom metadata fields to organize your content

​

Improvements

​

Enhanced Search Results

• Summary metadata in results - Top results may include summary_text and summary_chunk_id in metadata for better context
• Automatic summary fetching - Summary information is fetched automatically when available

{
  "id": "doc-123",
  "score": 0.856432,
  "label": "Close Match",
  "text": "...",
  "metadata": {
    "title": "Introduction to ML",
    "summary_text": "This document provides an overview of machine learning...",
    "summary_chunk_id": "summary-123"
  }
}

​

Performance Improvements

​

Faster Search Performance

• 10x faster search queries - Experience dramatically improved search speeds
• Query time reduced from 5-10s to 1s-2s for large datasets (100K+ vectors)

​

New Features

​

Chat History Management for console only

• Automatic conversation tracking for all AI interactions
• Auto-generated conversation titles (first 30 characters)

​

AI Reasoning Transparency

• See how AI models think through problems in real-time
• Reasoning process separated from final answer
• Streaming support for live thought process display
• Works with DeepSeek R1 and OpenAI o1 models

​

AI Model Updates

​

New Models Added

• Claude Sonnet 4.5 - Latest Anthropic model with enhanced capabilities
• OpenAI GPT OSS 120B - Large-scale model with hybrid reasoning
• Qwen 3 32B - Advanced multilingual model

​

Models Removed

• Claude 3.7 Sonnet - Will be deprecated by provider in 2026

​

Security

​

Patches

• Applied security patches across dependencies
• General hardening and minor fixes

​

Dependencies

​

Updates

• Updated project dependencies to latest compatible versions

​

Search Scoring

​

ITS Score Display

• Display of ITS (Information Theoretic Similarity) scores now uses a square-root transformation for improved readability
• Backend label mapping updated; API responses now return the new ITS labels

​

Direct AI Mode

​

Empty Namespace Support

• New Direct AI Mode when passing empty string "" as namespace
• Direct AI model calls without searching your data for general-purpose AI interactions
• Perfect for general AI assistance, creative writing, coding help, and educational content
• Maintains all AI model capabilities while bypassing data search functionality

​

API Flexibility

• Two distinct modes: Search Mode (with namespace) and Direct AI Mode (empty namespace)
• Search Mode: Full data search and retrieval capabilities
• Direct AI Mode: Pure AI model interaction without data constraints
• Seamless switching between modes based on namespace parameter

​

Enhanced Error Handling

​

Field Validation

• Specific error messages for field validation based on namespace mode
• Empty namespace with invalid fields: "Bad Request: namespace, query, temperature, chatHistory, footerPrompt, headerPrompt, aiModel are the only allowed fields with empty namespace."
• Provided namespace with invalid fields: "Bad Request: namespace, query, top_k, threshold, type, kiosk_mode, aiModel, chatHistory, headerPrompt, footerPrompt, temperature are the only allowed fields with provided namespace."
• Improved developer experience with clear guidance on correct field usage

​

Advanced Search Filters

​

Metadata Filters

• New #key:value filter syntax for document properties
• Filter by category, priority, author, status, and other metadata fields
• Case-insensitive exact matching with automatic key validation
• Unknown metadata keys are gracefully ignored without errors

​

Keyword Filters

• New #keyword filter syntax for text content
• Search within document text for specific keywords like #important, #urgent, #critical
• Multiple keyword filters require ALL keywords to be present
• Case-insensitive matching within actual document content

​

Combined Filtering

• Combine semantic search with metadata and keyword filters
• Format: search_text #key:value #keyword
• Filters must be placed at the END of query string
• Use hyphens instead of spaces in filter values

​

Enhanced Search API

• Updated Search API documentation with comprehensive filter examples
• New timing fields: earlyFilter and applyMetadataFilter
• Improved search optimization with early filtering capabilities
• Enhanced response structure for better performance tracking

​

Vector Metadata Enhancement

​

Flexible Metadata Structure

• Any fields beyond id and vector are automatically treated as metadata
• Enhanced vector upload with automatic metadata extraction
• Support for custom fields like category, source, product_id, customer_id
• Metadata is searchable and filterable in vector operations

​

Updated Vector Upload API

• Simplified request structure with automatic metadata handling
• Enhanced examples showing metadata usage in vector uploads
• Improved documentation for vector metadata best practices
• Better error handling for metadata validation

​

Documentation Updates

​

Search API Documentation

• Comprehensive filter usage examples with cURL requests
• Updated response structure documentation
• New timing information for performance monitoring
• Enhanced API call examples for different filter combinations

​

Vector Upload Documentation

• Updated examples showing metadata fields in vector uploads
• Clear explanation of automatic metadata handling
• Enhanced request parameter documentation
• Improved error handling and validation examples

​

Technical Improvements

​

Performance Enhancements

• Early filtering optimization for better search performance
• Enhanced timing tracking for debugging and optimization
• Improved search result relevance with metadata filtering
• Better resource utilization in vector operations

​

Developer Experience

• Simplified filter syntax for easier implementation
• Comprehensive examples for all filter combinations
• Better error messages and validation feedback
• Enhanced API documentation with practical use cases

​

Document Retrieval

​

Get Documents by ID

• New endpoint to retrieve specific documents from text-based namespaces
• Endpoint: POST /namespaces//documents/get
• Request Format: Send array of document IDs in request body
• Flexible Access: Retrieve up to 100 documents in a single API call

​

Local Development

​

Local MCP Support

• Run Moorcheh’s Model Context Protocol locally for development and testing
• Offline Development: Test and develop integrations without internet connectivity
• Local Environment: Full feature parity with cloud environment for testing

​

Document Upload System

​

Features

• Console Upload Interface: Direct document upload through the web console
• Supported File Formats: PDF, DOCX, CSV, JSON, TXT, MD, Excel files
• Real-time Status Tracking: Live document processing status displayed in namespace view

​

Enhanced AI Models

​

New Models

• Claude Sonnet 4 Support: Added new AI model option for improved performance

​

Namespace Improvements

​

New Features

• Document Status Display: View processing status of uploaded documents
• Upload Progress Tracking: Monitor document processing in real-time
• File Management: Better organization and tracking of uploaded content

​

Dashboard

​

Overview

• Quick overview tab for overall usage
• Weekly API usage line graph
• Quick usage metrics
• Recent activity feed

​

Usage Analytics

• Weekly usage graphs for specific APIs
• Platform health dashboard
• Usage limit visualizations with bar graphs
• Daily cost analysis (weekly view) with bar graphs

​

Performance Monitoring

• Quick performance cards
• 24-hour performance matrix with line graphs
• Average execution time tracking for upload, search, and answer operations

​

Namespace Management

​

Text Namespaces

• Namespace name and item count
• Creation time tracking

​

Vector Namespaces

• Namespace name and vector dimensions
• Item count and creation time tracking

​

Playground

​

Chat Interface

• Kiosk mode for focused interactions
• AI model selection

​

Settings & Configuration

• Header and footer prompt customization
• Model parameters (model selection, temperature)
• Top-k and threshold controls
• Metadata toggle options
• JSON configuration export

​

Chat History

• Conversation titles and content
• Chat timing and timestamps
• Local storage persistence

​

API Key Management

​

Key Operations

• Create new API keys
• Revoke existing keys
• Delete API keys

​

Pricing Plans

​

Subscription Tiers

• Community plan
• Professional plan
• Pay-as-you-go options
• Enterprise solutions

​

Documentation

​

Developer Resources

• Complete API reference
• Python SDK documentation