Infrastructure for LLM, RAG and memory systems in Python
What is Zerda?
Zerda is an open-source Python library for building LLM applications. It gives you ready-made building blocks — retrieval-augmented generation (RAG), agent memory, query routing, semantic caching, smart chunking, embeddings and observability — all with a clean, Pythonic API. Install it with a single command and go from prototype to production without wiring dozens of tools together.
Explore the modules
Every module has its own quickstart, architecture overview and API reference.
RAG
Retrieval-augmented generation engine plus a library of RAG strategies (agentic, graph, multi-hop, streaming and more).
Memory
Multi-tier memory architecture for conversational agents: buffers, summaries, entities, semantic recall and durable storage.
Router
A hierarchical, hybrid-scoring query router for LLM applications.
Cache
An intelligent, multi-tenant, cost-aware semantic caching module for LLM applications.
Chunks
Production-grade RAG chunking for Arabic, multilingual and English text.
LLM
A dependency-light unified interface for calling LLM providers, with tool calling across all seven backends.
Embeddings
Unified text-embedding interface — with first-class attention to Arabic — across local and hosted backends.
Vector Database
A unified interface over three different vector-search backends.
Context
Retrieval-orchestration and context-engineering for RAG, with support for mixed Arabic/English content.
Document Loaders
Converts files, web pages, directories, knowledge graphs and tables into a normalized LoadedDocument.
Prompt
Builds AI-ready, context-aware prompts for RAG — guardrailed, token-budgeted and provider-ready.
Output Parser
Output parsing, validation, fault-tolerance and retry engine for LLM/RAG/agent pipelines.
Observability
A generic, backend-agnostic tracing system for Zerda.
Plugins
Build, discover, secure and orchestrate plugins that an LLM can call.