PocketFlow - A minimalist flow-based agent framework for Nim

PocketFlow provides a flexible and powerful framework for building LLM-powered workflows, RAG systems, and agent pipelines.

Core Features:

• Node-based workflow orchestration
• Batch and parallel processing
• Retry logic and fallback handling
• Multiple LLM provider support (OpenAI, Anthropic, Google, Ollama)
• Streaming responses
• Caching and cost tracking
• RAG capabilities (chunking, embeddings, retrieval)
• Advanced node types (conditional, loop, timeout)
• Observability (logging, metrics, tracing)
• State persistence and recovery
• Performance benchmarking

Quick Example:

import pocketflow

let ctx = newPfContext()
let llm = newLlmClient(provider = OpenAI, apiKey = "sk-...")

let node = newNode(
  exec = proc(ctx: PfContext, params: JsonNode, prepRes: JsonNode): Future[JsonNode] {.async.} =
    let response = await llm.generate("Tell me a joke")
    return %response
)

let flow = newFlow(node)
waitFor flow.internalRun(ctx)