Packages

Shell scripting library for Python — provides Pythonic API for running subprocesses, building pipelines, and remote execution. plumbum features: local['cmd'] for local commands, local.path for path manipulation, | operator for pipelines, & for background execution, remote machines via SshMachine, FG/BG/TEE run modes, command composition (ls['-la'] | grep['py']), retcode handling, working directory context manager (local.cwd), environment variable context manager (local.env), CLI application framework (plumbum.cli), and cross-platform path handling. Bridges gap between shell scripts and Python programs.

Full-featured GraphQL server with schema-first resolver pattern, Apollo Studio integration, DataLoader for N+1 prevention, and built-in support for Apollo Federation.

Periodic task scheduler for Celery — runs cron-like scheduled tasks via Celery Beat process. Celery Beat features: @shared_task with beat_schedule in celery.py configuration, crontab() for cron expression schedules (every Monday at 9am), timedelta for interval schedules (every 10 minutes), django-celery-beat for DB-backed schedules editable via Django admin, dynamic schedule modification without restart, PeriodicTask model for programmatic schedule management, clocked() for one-time scheduled tasks, solar() for sunrise/sunset schedules, and timezone support. Separate celery beat -A myapp process runs alongside Celery workers for agent scheduled task execution.

Async Python web crawler optimized for LLM pipelines that renders JavaScript, extracts clean Markdown, and supports structured data extraction via CSS/XPath/LLM strategies.

Cloud API for automated video and image generation from templates. Creatomate provides a template editor (UI or JSON-based) and a REST API to render videos/images with dynamic content — variable text, images, colors, and audio. No need to set up rendering infrastructure (FFmpeg, Chrome, Lambda) — Creatomate handles all rendering server-side. Used for social media automation, personalized video generation, and content workflows.

Cyclr REST API for embedded integration platform as a service (embedded iPaaS). Enables AI agents to manage integration template creation and deployment, handle connector and cycle (workflow) library management, access end-user integration activation and configuration, retrieve integration run status and error monitoring, manage multi-tenant integration deployment across SaaS customers, handle OAuth connection management for third-party services, access connector library with 500+ pre-built integrations, retrieve integration analytics and usage metrics, manage white-label integration marketplace for SaaS products, and embed integration capabilities into SaaS platforms for end-user automation.

Safe, extensible ORM and query builder for Rust. Diesel generates Rust types from your database schema and provides a compile-time verified query DSL — if your query compiles, it's valid SQL. Unlike SQLx (raw SQL with macros), Diesel uses a typed query builder (users.filter(name.eq('Alice')).select(id)) that prevents schema mismatches at compile time. The most mature Rust ORM.

PyTorch-based OCR library supporting 80+ languages with GPU acceleration, returning text with bounding boxes and confidence scores from images.

Google Cloud NLP API for sentiment analysis, entity extraction, syntax analysis, content classification, and entity sentiment on text.

Provides geocoding, reverse geocoding, routing, matrix routing, map tiles, places search, geofencing, and fleet telematics APIs for building location-aware applications and agent workflows.

Fast tokenization library for NLP — Rust-backed tokenizer with Python bindings for training and using tokenizers. HuggingFace Tokenizers features: AutoTokenizer (loads pretrained tokenizer from Hub), encode/decode for single texts, batch encoding, special tokens ([CLS], [SEP], [PAD]), truncation and padding, fast tokenization (10-100x faster than slow tokenizers), offset mapping for token-to-character alignment, custom BPE/WordPiece/Unigram tokenizer training, and integration with HuggingFace Transformers. Used for preparing agent text for LLM inference and fine-tuning.

High-performance ML research framework — NumPy-compatible array library with autograd, JIT compilation, and hardware acceleration. JAX features: jax.grad() for automatic differentiation, jax.jit() for XLA compilation (10-100x speedup), jax.vmap() for vectorization, jax.pmap() for multi-device parallelism, jax.numpy (jnp) NumPy-compatible API, functional programming model (pure functions required), random number via jax.random.PRNGKey (not stateful), and TPU/GPU/CPU backends. Preferred framework for ML research — neural network libraries Flax and Optax build on JAX. Used by Google DeepMind for research.

LangSmith REST API — LLM observability, tracing, and evaluation platform enabling agents to log runs, trace multi-step chains, evaluate outputs with datasets, and monitor production LLM applications for latency, cost, and quality.

Email marketing platform designed specifically for SaaS companies, combining transactional and marketing emails in a single tool with a clean REST API for contact management and event-triggered campaigns.

Multimedia streaming framework for Elixir built on the BEAM. Membrane provides a composable pipeline model for processing audio/video streams — ingest from RTMP/WebRTC/HLS, transcode, mix, and output to multiple targets. Uses the same GenStage-inspired pipeline model as Broadway but for multimedia. Supports WebRTC peer connections, RTMP ingest (for OBS-style streaming), HLS output, MP4/Matroska muxing, Opus/AAC audio, and H264/H265 video. Used by Jellyfish (WebRTC conferencing server) and VideoRoom.

Structured Python agent library with built-in tools, memory, storage, and knowledge bases — now rebranded and superseded by the Agno framework.

Qdrant is a high-performance vector database purpose-built for storing, indexing, and querying high-dimensional embeddings. Its REST API (and parallel gRPC API) supports collection management, vector upsert with arbitrary JSON payloads, approximate nearest neighbor (ANN) search with payload filtering, named vectors for multi-modal data, and sparse vectors for hybrid dense+sparse search. Available as self-hosted (Docker/Kubernetes) or Qdrant Cloud.

Official Salesforce MCP server enabling LLMs to interact with Salesforce orgs for metadata deployment, SOQL queries, code analysis, LWC development, user management, DevOps Center workflows, and mobile development - with 60+ tools across 15+ toolsets.

Spring integration for Apache Kafka — provides KafkaTemplate for producing, @KafkaListener for consuming, and Kafka Streams support. Spring Kafka features: KafkaTemplate.send() for producing records, @KafkaListener(topics='agent-events') for consuming, automatic offset management, consumer group coordination, batch consumption mode, dead-letter topics (DeadLetterPublishingRecoverer), retry with exponential backoff (DefaultErrorHandler), Kafka Streams binding via StreamsBuilderFactoryBean, transaction support (KafkaTransactionManager), embedded Kafka for testing (EmbeddedKafkaBroker), and Spring Boot auto-configuration via spring-boot-starter. Key difference from AMQP: Kafka retains all messages (configurable retention), enables replay and multiple independent consumer groups reading same topic.

Provides hosted map tiles (raster and vector), geocoding/reverse geocoding, turn-by-turn routing, and isochrone analysis built on OpenStreetMap and Valhalla.