Emrea 

Zero-dependency shared types for the Agency AI Toolkit: Dataset and IColumnMetadata used across SQL runners, the RAG formatter, and downstream LLM consumers.

IEmbeddingGenerator interface and shared embedding abstractions for the Agency AI Toolkit. Implementation-agnostic contract consumed by vector stores, SQL runners, and ingestion pipelines.

SqlRunnerBase abstract class for the Agency AI Toolkit: shared OpenTelemetry instrumentation and execution skeleton reused by every concrete SQL runner.

IVectorStore, Query, and SearchHit<T> abstractions for the Agency AI Toolkit. Provider-agnostic contract for nearest-neighbour search across pgvector, SQLite, and future backends.

Core agentic loop for the Agency AI Toolkit: Agent, ChatSession, Context composition, ToolRegistry, and StopConditions for multi-turn LLM agents with tool calling.

IModelProvider interface and shared LLM tool/message types for the Agency AI Toolkit. The single contract that Claude, OpenAI, and any future provider implement.

PostgreSQL raw-SQL runner for the Agency AI Toolkit with a built-in vectorize() macro that inlines pgvector literals from natural-language strings via IEmbeddingGenerator.

OpenAI IModelProvider implementation for the Agency AI Toolkit, built on the official .NET OpenAI SDK with full OpenTelemetry traces and token-usage metrics. Works with Azure OpenAI and OpenAI-compatible... More information

Anthropic Claude IModelProvider implementation for the Agency AI Toolkit, built on the Stainless Anthropic SDK with full OpenTelemetry traces and token-usage metrics.

Renders an Agency Dataset as a Markdown table for injection into LLM context windows. The minimal RAG formatting step between SQL retrieval and chat completion.

Shared configuration layer and ${Section:Key} placeholder resolution for the Agency AI Toolkit.

SQLite raw-SQL runner for the Agency AI Toolkit with a built-in vectorize() macro that inlines embedding literals from natural-language strings via IEmbeddingGenerator.

Ingestion pipeline abstractions and DefaultIngestionPipeline<T> for the Agency AI Toolkit. Compose loaders, chunkers, embedders, and stores into a single load-and-index workflow.

Common types, interfaces, and options for the Agency long-term memory system.

SQLite implementation of IMemoryStore for the Agency long-term memory system. Embedded, zero-server, with an in-process cosine UDF.

Retrieval engine for the Agency long-term memory system. Implements the gated vector search, composite ranking, and context injection for the agent hot path.

PostgreSQL + pgvector implementation of IMemoryStore for the Agency long-term memory system.

IKVStore, Query, SearchHit<T>, and JSON metadata helpers for the Agency AI Toolkit. Provider-agnostic key-value contract used as the metadata layer alongside vector search.

Semantic Kernel TextChunker-based text splitter for the Agency ingestion pipeline. Token-aware chunking with configurable overlap suitable for embedding-and-retrieval workflows.

Distiller background service for the Agency long-term memory system: episode extraction, inactivity timer, and agent tools.