Pipe (Chain) uses Input, Model and Output to make the interaction with the model easier.
Input -> Model -> Output
LLM -> Agent Tools -> Agent Agent -> Executor (Iterator)
(Agents: Static Validator/Limits/Logger)
- Input: Chat messages and prompt
- Models: Ollama and OpenAI
- Parser: String, Json, Markdown, Symbol-Seperator and Map
- Pipe: Simple pipe to easily use input, models and output.
- Embedder: For vector embeddings
- Evaluator: Whitelist or Blacklist output before parsing (Soon in pipe and agent available)
- Basic Agent using the ReAct Pattern
- Static Tools use (Probably add input instruction)
- Agent Executer
- Limits and Logging of Agents and Executor
- Static Agent validator
- Similarity Search (Jaccard and Cosine)
- Vector Database (Qdrant)
- Tools (.txt, pdf, ...)
- Core (Input, Model, Output, Pipe, Embedder)
- Temperature and time Agents
- RAG: Simple RAG with Jaccard
- Vectorstore: Qdrant