SimpleAgents

Every agentic SaaS is a config.

Define your AI product as a YAML workflow. Run it in Python or TypeScript. Ship today.

Links

• Docs: https://docs.simpleagents.craftsmanlabs.net/
• Playground: https://yamslam.craftsmanlabs.net/playground

Install

pip install simple-agents-py     # Python
npm install simple-agents-node   # TypeScript / Node

How It Works

1. Define your workflow as YAML -- nodes, edges, structured outputs, routing
2. Run it with 10 lines of Python or TypeScript
3. Ship -- streaming, images, Langfuse/Jaeger observability all work out of the box

Every email classifier, document processor, intake system, interview bot, and support agent is the same pattern: LLM nodes with structured outputs and deterministic routing. SimpleAgents makes that pattern a config file.

Quick Example

workflow.yaml

id: classifier
version: 1.0.0
entry_node: classify

nodes:
  - id: classify
    node_type:
      llm_call:
        model: gpt-4.1-mini
        messages_path: input.messages
        append_prompt_as_user: true
        heal: true
    config:
      output_schema:
        type: object
        properties:
          category:
            type: string
            enum: [billing, support, sales]
        required: [category]
        additionalProperties: false
      prompt: |
        Classify the user message. Return JSON only.

edges: []

run.py

import json, os
from pathlib import Path
from dotenv import load_dotenv
from simple_agents_py import Client
from simple_agents_py.workflow_payload import workflow_execution_request_to_mapping
from simple_agents_py.workflow_request import (
    WorkflowExecutionRequest, WorkflowMessage, WorkflowRole,
)

load_dotenv()
client = Client(os.environ["WORKFLOW_PROVIDER"], api_base=os.environ["WORKFLOW_API_BASE"], api_key=os.environ["WORKFLOW_API_KEY"])

req = WorkflowExecutionRequest(
    workflow_path=str(Path("workflow.yaml").resolve()),
    messages=[WorkflowMessage(role=WorkflowRole.USER, content="I need a refund for order #1234")],
)
result = client.run_workflow(workflow_execution_request_to_mapping(req))
print(json.dumps(result, indent=2))

That's it. Your agentic SaaS is a config.

What You Get

• YAML workflow engine -- classify, route, extract, generate as a graph config
• Python + TypeScript -- pip install / npm install, run with 10 lines
• Streaming -- real-time LLM output streaming
• Images -- multimodal input (text + images) in the same workflow
• JSON healing -- auto-fix truncated/malformed LLM JSON output
• Observability -- Langfuse and Jaeger via OpenTelemetry, one env block
• Custom workers -- plug your own code (DB lookups, APIs) into the workflow graph
• Rust core -- blazing fast engine with Python, TypeScript, and WASM bindings

Documentation

• Start here: docs/WORKFLOW_QUICKSTART.md -- install, create YAML, run in Python/TypeScript
• Examples: docs/EXAMPLES.md
• YAML system guide: docs/YAML_WORKFLOW_SYSTEM.md
• Python binding: docs/BINDINGS_PYTHON.md
• Node/TypeScript binding: docs/BINDINGS_NODE.md
• WASM binding: docs/BINDINGS_WASM.md
• Observability (Langfuse/Jaeger): docs/TRACING_ARCHITECTURE.md
• Rust quick start: docs/QUICKSTART.md
• Rust usage: docs/USAGE.md
• Troubleshooting: docs/TROUBLESHOOTING.md
• Development/Contributing: docs/DEVELOPMENT.md
• Docs map: docs/DOCS_MAP.md

Contributing

• Start with CONTRIBUTING.md and docs/DEVELOPMENT.md.
• Follow task-tracking expectations in TODO.md (and SUBAGENT_TODO.md for larger parallel workstreams).
• Run relevant test/lint/format/parity commands before opening a PR.

License

• Repository license file: LICENSE (Apache License 2.0 text).
• Package metadata in workspace includes MIT OR Apache-2.0 for crates/bindings where declared.

For redistribution/compliance-sensitive usage, verify root license files and per-package metadata.

Star History