Skip to main content
All notable changes to the Avala platform.

v0.5.0 — March 2026

New Features

  • datasets.create() organization support — Assign datasets to an organization at creation time via organization_id parameter (Python, TypeScript, REST API)
  • Additional dataset creation options — New optional parameters: gpu_texture_format, metadata, industry, license

v0.4.2 — February 2026

New Features

  • signup() — Create new Avala accounts programmatically via POST /api/v1/signup/ (Python, TypeScript, REST API). No API key required — returns one on success.

v0.4.1 — February 2026

New Features

  • datasets.create() — Create datasets programmatically with optional S3 provider config (Python, TypeScript, MCP, CLI)

v0.4.0 — February 2026

New SDK Resources

  • Organizations — Full CRUD for organizations, members, invitations, and teams (Python, TypeScript)
  • Slices — Create and browse data slices (Python, TypeScript)
  • Dataset Items & Sequences — Browse items and sequences within datasets (Python, TypeScript)

MCP Server Expansion

  • Added 21 new tools: tasks, agents, webhooks, storage, quality targets, consensus, organizations, slices
  • Total: 29 tools (up from 8)

CLI Expansion

  • Added 6 new command groups: agents, webhooks, quality-targets, consensus, inference-providers, auto-label
  • Total: 12 command groups (up from 6)

API Documentation

  • Updated Python SDK reference with all 14 resources (previously only 5 were documented)
  • Added CLI documentation for 6 previously undocumented commands
  • Updated TypeScript SDK reference with organizations, slices, and dataset item resources
  • Expanded MCP setup page with full tool table (29 tools)

February 2026

Webhook Delivery System

New Features
  • Subscribe to webhook events: dataset.created, dataset.updated, dataset.deleted, export.completed, export.failed, task.completed
  • HMAC-SHA256 signed payloads for security verification (X-Avala-Webhook-Signature header)
  • Automatic retry with exponential backoff (up to 5 attempts)
  • Webhook delivery log with status tracking (pending, delivered, failed)
  • Test endpoint to verify webhook connectivity
  • REST API: GET/POST /api/v1/webhooks/, GET/PUT/DELETE /api/v1/webhooks/{uid}/, POST /api/v1/webhooks/{uid}/test/
  • Read-only delivery log: GET /api/v1/webhook-deliveries/

CLI Tool

New Features
  • avala CLI tool available as optional extra: pip install avala[cli]
  • Commands: avala datasets list|get, avala projects list|get, avala exports list|get|create, avala tasks list|get, avala storage-configs list|create|test|delete
  • avala configure for interactive API key setup
  • Rich terminal output with formatted tables
  • Supports AVALA_API_KEY and AVALA_BASE_URL environment variables

SDK Storage Configs & Rate Limit Headers

New Features
  • Python SDK: client.storage_configs.list(), .create(), .test(), .delete()
  • TypeScript SDK: avala.storageConfigs.list(), .create(), .test(), .delete()
  • Rate limit info accessible via client.rate_limit_info (Python) / avala.rateLimitInfo (TypeScript)
  • Extracted from response headers: X-RateLimit-Limit, X-RateLimit-Remaining, X-RateLimit-Reset

Self-Service Cloud Storage

New Features
  • Connect your own S3 or Google Cloud Storage buckets directly from Mission Control settings
  • Test connection button to verify bucket access and credentials
  • Verified/unverified status badges for each storage configuration
  • Support for S3 Transfer Acceleration and bucket prefixes

Work Batch Progress

New Features
  • Customers can now view their organization’s work batch progress in Mission Control
  • Real-time progress stats: total work units, backlog, in-progress, completed, and error counts
  • Progress bar visualization for each batch

API Key Granular Scopes

New Features
  • API keys can now be restricted to specific capabilities (scopes)
  • Available scopes: datasets.read, datasets.write, projects.read, exports.create, tasks.read
  • Empty scopes = full access (backward compatible with existing keys)
  • Scope selection UI when creating API keys in Mission Control
  • Scope badges displayed on each key in the API keys list

Global API Rate Limiting

New Features
  • All API endpoints are now protected by global rate limiting with per-user throttles
  • Standard rate: 100 requests/minute, burst protection at 20 requests/second
  • Dedicated inference throttle: 10 requests/minute for AI inference endpoints
  • Anonymous request throttle: 30 requests/minute
  • Rate limit headers on every response: X-RateLimit-Limit, X-RateLimit-Remaining, X-RateLimit-Reset
  • DRF Retry-After header included on 429 responses

Python SDK v0.3.0

Improvements
  • Default base URL changed from server.avala.ai to api.avala.ai
Previously v0.2.0: Client and AsyncClient, datasets, projects, exports, tasks, storage_configs resources, CursorPage[T] pagination, CLI tool (pip install avala[cli]).

TypeScript SDK v0.3.0

Improvements
  • Default base URL changed from server.avala.ai to api.avala.ai
Previously v0.2.0: storageConfigs resource, rate limit info, comprehensive test suite.

MCP Server v0.3.0

Improvements
  • Default base URL changed from server.avala.ai to api.avala.ai
Previously v0.2.0: Package README, comprehensive test suite, bug fixes.

Agents SDK v0.1.0

New Features
  • avala-agents Python SDK published to PyPI
  • TaskAgent class for building custom annotation workflow agents
  • Event handlers: result.submitted, result.accepted, result.rejected, task.completed
  • Agent actions: approve(), reject(), flag(), skip()
  • Blocking agent.run() and non-blocking agent.run_once() execution modes
  • Public mirror repository: avala-ai/avala-agents-python

Agent Framework API

New Features
  • Register custom automation agents that react to annotation workflow events
  • Supported events: result.submitted, result.accepted, result.rejected, task.completed, plus dataset and export events
  • Agent actions: approve, reject, flag, or skip results
  • Execution logging with detailed status tracking
  • Automatic timeout of stale executions (10-minute limit)
  • HMAC-SHA256 signed webhook callbacks for agent event delivery
  • REST API: GET/POST /api/v1/agents/, GET/PUT/DELETE /api/v1/agents/{uid}/, POST /api/v1/agents/{uid}/action/

Batch Auto-Labeling

New Features
  • Run SAM or YOLO inference across entire datasets to generate draft annotations
  • Progress tracking with real-time counters (processed, successful, failed, skipped items)
  • Configurable confidence threshold and label filtering
  • Dry run mode to preview results without creating annotations
  • Job cancellation with graceful stopping (checks every 50 items)
  • Webhook events on completion (auto_label.completed, auto_label.failed)
  • Maximum 5,000 items per job, one job per project at a time
  • REST API: POST /api/v1/projects/{uid}/auto-label/, GET/DELETE /api/v1/auto-label-jobs/{uid}/

Custom Inference Providers

New Features
  • Connect your own HTTP model servers (PyTorch Serve, Triton, BentoML, Flask, etc.)
  • HTTPS-only with SSRF protection (blocks private/reserved IPs including CGNAT, IPv6 ULA, IPv4-mapped, 6to4)
  • DNS re-validation at invocation time to prevent DNS rebinding attacks
  • Configurable auth (Bearer token or API key), custom headers, and timeout
  • Flexible response parsing (supports predictions key or flat array)
  • Test connection endpoint to verify reachability
  • REST API: GET/POST /api/v1/inference-providers/, GET/PUT/DELETE /api/v1/inference-providers/{uid}/, POST /api/v1/inference-providers/{uid}/test/

Quality Targets & Alerts

New Features
  • Define quality thresholds per project that trigger alerts when breached
  • Supported metrics: issue rate, acceptance rate, rejection rate, mean resolution time, critical issue count
  • Configurable operators (greater than, less than, etc.) and severity levels (warning, critical)
  • Automatic evaluation every 15 minutes via Celery Beat
  • Webhook alerts (quality.alert) when thresholds are breached
  • Breach tracking with count and last-breached timestamp
  • REST API: GET/POST /api/v1/quality-targets/, GET/PUT/DELETE /api/v1/quality-targets/{uid}/, POST /api/v1/quality-targets/{uid}/evaluate/

Consensus Scoring

New Features
  • Compute inter-annotator agreement scores for items labeled by multiple annotators
  • Scoring methods: classification agreement ratio, bounding box IoU with greedy matching
  • Configurable IoU threshold and minimum annotator count per project
  • Bulk upsert with conflict resolution for safe concurrent runs
  • Triggered on-demand via API
  • REST API: GET/POST /api/v1/consensus-configs/, POST /api/v1/consensus/compute/, GET /api/v1/consensus-scores/

SDK Monorepo Integration

Infrastructure
  • Python, TypeScript, and Agents SDKs maintained in the monorepo (sdks/python/, sdks/typescript/, sdks/python-agents/)
  • CI workflows for Python (3.9-3.12), TypeScript (Node 18/20/22), and Agents (Python 3.9-3.12)
  • Automated publishing to PyPI and npm on version tags
  • Mirror workflow syncs to public GitHub repos (avala-ai/avala-sdk-python, avala-ai/avala-sdk-ts, avala-ai/avala-agents-python)

Docs Quick Wins

Improvements
  • Slices API reference rewritten to match real serializer output (response fields, create request body, pagination format)
  • Added llms-full.txt for comprehensive AI tool documentation ingestion

Documentation Overhaul

Improvements
  • All SDK examples updated to use real method signatures
  • Pagination format corrected to match real DRF responses: next/previous/results for cursor-based endpoints, plus count for page-number endpoints
  • Upload flow updated to presigned URL pattern
  • Rate limits page updated with real values and headers
  • Added llms.txt for AI tool discoverability

February 2025

MCP Server v1.0

New Features
  • Launched Avala MCP Server with full support for Claude, Cursor, and VS Code Copilot
  • AI assistants can now query datasets, projects, annotations, and organization data through natural language
  • New SDK documentation pages for both Python (avala) and TypeScript (@avala-ai/sdk)

Integration Guides

New Features
  • Added 6 new integration guides covering popular tools and workflows
  • Step-by-step setup instructions for connecting Avala with external services

Expanded API Reference

New Features
  • New dataset items endpoints for managing individual items within datasets
  • New tasks endpoints for querying and managing annotation tasks
  • New organizations endpoints for member management and invitations
  • New slices endpoints for creating and managing dataset subsets

Documentation Additions

New Features
  • New annotation tools reference covering all supported annotation types
  • Quality control guide with best practices for consensus and review workflows
  • Work batches guide for managing task distribution and annotator assignments

January 2025

Mission Control v2.5.0

New Features
  • Added batch tagging for multiple datasets
  • New keyboard shortcuts for faster navigation
  • Improved 3D point cloud rendering performance
Improvements
  • Search now supports date range filters
  • Reduced memory usage for large datasets
  • Better error messages for upload failures
Bug Fixes
  • Fixed timeline sync issues with multiple cameras
  • Resolved Safari compatibility issues
  • Fixed annotation export including deleted items

December 2024

API v1.3.0

New Features
  • Added webhooks endpoint for event notifications
  • New batch endpoints for bulk operations
  • Support for custom metadata on channels
Improvements
  • Increased rate limits for enterprise accounts
  • Faster response times for list operations
  • Better pagination with cursor support
Deprecations
  • offset pagination deprecated in favor of cursor-based pagination
  • v1/upload endpoint deprecated, use v1/datasets/{id}/upload

Python SDK v0.8.0

New Features
  • Async client support (AsyncClient)
  • Progress callbacks for uploads
  • Automatic retry with exponential backoff
Improvements
  • Full type hints for IDE support
  • Reduced memory usage for large file uploads
  • Better error messages

November 2024

Mission Control v2.4.0

New Features
  • Team-based permissions
  • Project organization
  • Data lifecycle policies
Improvements
  • Faster dataset loading
  • Improved search relevance
  • New dark mode theme

Earlier Releases

For earlier release notes, see our GitHub releases.