All notable changes to Antfly will be documented in this file.

Roadmap#

Upcoming Features#

  • Multimodal Search Enhancements

    • Support for ImageBind as a multimodal embedder
    • Image-to-image similarity search (search with images as queries)
    • Extended audio file support (MP3, WAV, FLAC)
    • Video file processing and frame extraction
    • Automatic scene detection and indexing for video content

  • Advanced AI Capabilities

    • Custom prompt templates for different content types
    • Model performance comparison tools
    • Automatic model selection based on content type
    • Fine-tuning support for domain-specific embeddings

Releases#

[0.0.13] - 2026-03-03#

Highlights#

  • Sparse Vector Search (SPLADE) — hybrid search combining dense and sparse vectors with weighted fusion for better relevance
  • PostgreSQL CDC Replication — automatically sync data from PostgreSQL into Antfly via logical replication
  • Operator Improvements — PVC lifecycle management, availability zone topology, admission webhooks, and storage resilience
  • Faster Sparse Indexing — up to 3.6x faster sparse index inserts with multi-tier caching

Features#

  • Sparse Vector (SPLADE) Search — hybrid dense+sparse fusion with configurable per-index merge weights
  • PostgreSQL CDC Replication — logical replication with automatic change capture from PostgreSQL tables
  • Chunking and Summarization support in sparse embeddings index
  • Operator: PVC Lifecycle management, availability zone topology, and storage resilience
  • Operator: Admission Webhooks for AntflyCluster resource validation

Performance#

  • Up to 3.6x faster sparse index inserts
  • Multi-tier caching and batched writes for sparse indexes
  • Configurable sync level for embeddings indexes

Bug Fixes#

  • Fix transaction correctness for cross-table OCC commits
  • Fix read availability during leader elections by falling back to healthy replicas
  • Fix full text index version resolution and rebuild during schema migration
  • Fix search result filtering for chunked embedding indexes
  • Fix graph node key decoding in fusion results
  • Fix UTF-8 streaming in classification reasoning parser

Documentation#

  • Add SDK documentation pages for Antfly and Termite
  • Add CDC replication guide with cloud provider setup instructions

Full changelog

[0.0.9] - 2026-02-22#

Highlights#

  • Secrets Management — new API and dashboard page for managing secrets
  • API Key & Bearer Token Auth — authenticate with API keys or bearer tokens
  • AI Provider Timeouts — configurable timeout for AI provider calls
  • PDF Enrichment Pipeline — zip-direct reading, parallel extraction, vision-based categorization, and Florence 2 re-OCR for low-quality pages
  • Omni Edition — renamed install edition with streamlined macOS support

Features#

  • Secrets Management API and Antfarm dashboard page
  • API Key & Bearer Token Authentication
  • AI Provider Timeout configuration
  • PDF Enrichment — direct zip reading, parallel page extraction, page-type categorization with vision support, and Florence 2 re-OCR for low-quality pages
  • Dashboard improvements with reverse proxy support and sidebar redesign

Bug Fixes#

  • Fix transaction intent conflict detection
  • Auto-abort stale transactions to prevent permanent blocking
  • Fix out-of-memory when processing large zip archives
  • Prevent rechunking all documents on every restart in ephemeral mode

Documentation#

  • Rename ONNX edition to Omni
  • Add pod scheduling guide for operator
  • Update quickstart to use built-in providers

Full changelog

[0.0.8] - 2026-02-18#

Highlights#

  • Cross-Table Transactions — optimistic concurrency control (OCC) with read-modify-write support across tables
  • Built-in Embedder & Reranker — bundled INT8 quantized all-MiniLM-L6-v2 embedder and reranker, no external service required
  • Shared Pebble Block Cache — single block cache shared across all DB instances per process for better memory utilization
  • Operator Scheduling Constraints — tolerations, nodeSelector, affinity, and topologySpreadConstraints in AntflyCluster CRD
  • Cluster Hibernation — scale operator replicas to zero while retaining PVCs for cost savings

Features#

  • Cross-Table Transactions with OCC read-modify-write support
  • Built-in Embedder — INT8 quantized all-MiniLM-L6-v2 bundled with Antfly
  • Built-in Reranker — INT8 quantized reranker model bundled with Antfly
  • Shared Pebble Block Cache across all DB instances per process
  • Operator Scheduling Constraints — tolerations, nodeSelector, affinity, and topologySpreadConstraints added to AntflyCluster CRD
  • Cluster Hibernation — allow scaling metadata and data node replicas to zero

Bug Fixes#

  • Fix Raft proposal dropped incorrectly treated as permanent error during leader election
  • Fix case-sensitive matching in transient shard error detection
  • Fix answer agent compatibility layer SSE event mapping and generation enable
  • Fix S3 restore path to use common.ParseS3URL
  • Fix publicAPI.enabled default to false and delete service when disabled

Full changelog

[0.0.7] - 2026-02-13#

Highlights#

  • Retrieval & Generation Agents — new agentic architecture for retrieval-augmented generation
  • MCP & A2A Protocol Support — connect Antfly to AI agents via MCP (/mcp/v1) and Agent-to-Agent protocol
  • Foreign Tables — federated queries against external PostgreSQL databases
  • Named Provider Registry — configure embedders, generators, rerankers, and chunkers by name
  • Audio Transcription — speech-to-text support via Termite
  • Ephemeral Chunks — transient chunk storage with the store_chunks config option

Features#

  • Retrieval Agents — tool-use agentic loop for retrieval and generation, replacing the previous answer endpoint (deprecated /agents/answer still available for backward compatibility)
  • MCP Server at /mcp/v1 for AI agent integration
  • A2A Protocol facade for retrieval and query-builder agents
  • Foreign Tables for federated PostgreSQL queries
  • Named Provider Registry for embedders, generators, chains, rerankers, and chunkers
  • Audio/STT with Termite as speech-to-text provider and media chunking support
  • Ephemeral Chunks mode (store_chunks config option) for transient chunk storage
  • Graph Index — field-based edges, topology constraints, and summarizer
  • Remote Content configuration system for web scraping
  • CLAP & CLIPCLAP model support for audio embeddings
  • Antfly Operator now included in the main repository with docs and install manifests

Bug Fixes#

  • Fix two transaction bugs found via TLA+ model checking
  • Fix shard split finalizing before initialization completes
  • Fix node crashes from closed-channel panics and double-close on shard shutdown
  • Fix S3 endpoint configuration (AWS_ENDPOINT_URL env var now supported)
  • Extensive JPEG2000 decoding improvements for PDF processing
  • Fix MCP schema compatibility issues

Deprecations#

  • /agents/answer endpoint deprecated in favor of /agents/retrieval

Full changelog

[0.0.2] - 2026-01-12#

Highlights#

  • Cross-table join support for queries spanning multiple tables
  • Zero-downtime shard splitting with two-phase split
  • TTS/STT audio library with OpenAI and Google Cloud providers
  • CLIP model support for multimodal image indexing

Features#

  • Cross-Table Joins — query across multiple tables with shard-aware routing
  • Zero-Downtime Shard Splitting — two-phase split for high availability
  • Audio Library — TTS and STT support with OpenAI and Google Cloud providers
  • Dynamic Templates — flexible field mapping with automatic schema inference
  • Aggregations API — range and term aggregations (renamed from Facets)
  • Chat Agent — tool execution, clarification handling, confidence scoring, and multi-turn query builder mode
  • Indexes in Raft Snapshots — faster recovery with pause/resume for index operations
  • ONNX Runtime GenAI bundled for local LLM generation

Bug Fixes#

  • Fix data loss during shard splits with SyncLevelWrite
  • Fix multiple shard split race conditions and state transition issues
  • Fix CoreML memory inflation with full-precision ONNX models
  • Fix memory leak in store shutdown
  • Fix CORS configuration handling

Full changelog

[0.0.1] - 2025-12-20#

First official release of Antfly.

Features#

  • Unified ONNX + XLA build for cross-platform ML inference
  • Termite downloads page with Homebrew support
  • antflycli included in container images
  • Document TTL support

Bug Fixes#

  • Fix shard split data loss and race conditions
  • Fix Raft initialization resilience
  • Improve shard split reliability with two-phase approach

Documentation#

  • Mobile-friendly docs navigation
  • Discord and GitHub community links

Full changelog

[0.0.0-dev13]#

Features#

  • Graph Database — declarative graph query DSL, shortest path queries, edge expiration and TTL
  • Document Transform Operations — MongoDB-style in-place updates, upserts, path-based field manipulation
  • Termite — dynamic model loading, reranker support with ONNX models
  • TOON Binary Encoder for efficient data serialization
  • RAG — result pruning, streaming LLM responses, sync-level enrichment control
  • Dashboard UI — visual Bleve query builder, table filtering, improved results display
  • Encrypted Keystore for secrets management
  • AVX-512 SIMD acceleration for vector operations

Full changelog

[0.0.0-dev3] - 2025-08-XX#

Breaking Changes#

  • Index configuration moved into OpenAPI specification (previous metadata instances are incompatible)
  • API routes namespaced under /api/v1

Features#

  • Dashboard UI — web interface for managing tables and indexes, hosted on metadata server

Full changelog