Beddel

Agent-Native Workflow Engine — Solution Kits

---

Beddel is a Python SDK for declarative AI workflow execution. Define workflows in YAML, compose them from atomic primitives (llm, chat, guardrail, tool, call-agent, agent-exec), and let the engine handle branching, retry, reflection loops, parallel execution, and multi-provider LLM routing — all without boilerplate code.

Solution Kits extend the slim core with isolated adapters, tools, and integrations. Install only what you need.

Install

pip install "beddel[default]"    # batteries included: litellm + otel + fastapi + httpx

Or start minimal and add kits as needed:

pip install beddel                          # core only (3 deps: pydantic, pyyaml, click)
beddel kit install provider-litellm-kit     # add LLM provider support

Requires Python 3.11+.

Quick Example

# workflow.yaml
steps:
  - id: greet
    primitive: llm
    config:
      model: gemini/gemini-2.0-flash
      prompt: "Say hello and share a fun fact about $input.topic"
      temperature: 0.7

export GEMINI_API_KEY="your-key"
beddel run workflow.yaml -i topic=astronomy

Solution Kits

Each kit is a self-contained package with its own dependencies declared in kit.yaml. The install command resolves sources and handles pip dependencies automatically.

beddel kit install <kit-name>               # from this repository
beddel kit install ./my-custom-kit/         # from a local directory
beddel kit install github:org/repo/kits/x   # from any GitHub repo
beddel kit list                             # list installed kits

🔌 Providers

Kit	Description	Dependencies
provider-litellm-kit	Multi-provider LLM adapter — 100+ providers via LiteLLM (OpenAI, Gemini, Anthropic, Bedrock, Ollama, and more)	litellm

🤖 Agent Adapters

Kit	Description	Dependencies
agent-claude-kit	Claude Agent SDK subprocess-based agent adapter	claude-agent-sdk
agent-codex-kit	Codex Docker-isolated agent adapter	—
agent-kiro-kit	Kiro CLI subprocess-based agent adapter	—
agent-openclaw-kit	OpenClaw Gateway HTTP API agent adapter	httpx

📡 Protocols

Kit	Description	Dependencies
protocol-mcp-kit	Model Context Protocol client (stdio + SSE transports)	mcp, jsonschema

🔐 Auth

Kit	Description	Dependencies
auth-github-kit	GitHub OAuth Device Flow credential management	httpx

👁️ Observability

Kit	Description	Dependencies
observability-otel-kit	OpenTelemetry tracing adapter — workflow, step, and primitive-level spans with token tracking	opentelemetry-api
observability-langfuse-kit	Langfuse tracing adapter — token usage, latency, and cost attribution	langfuse

🌐 Serving

Kit	Description	Dependencies
serve-fastapi-kit	FastAPI handler factory + SSE streaming — expose workflows as HTTP endpoints	fastapi, sse-starlette
serve-mcp-kit	Expose YAML workflows as MCP servers — any MCP-compatible agent can discover and execute them	mcp

🛠️ Tools

Kit	Description	Dependencies
tools-file-kit	File I/O tools (read, write) with path validation	—
tools-shell-kit	Shell command execution via SafeSubprocessRunner	—
tools-http-kit	HTTP request tool via httpx	httpx
tools-gates-kit	Validation gate tools (pytest, ruff, mypy)	—

📦 Development

Kit	Description	Dependencies
software-development-kit	Development workflow tools — validation gates and epic creation	—

Creating Your Own Kit

A kit is a directory with a kit.yaml manifest:

name: my-custom-kit
version: "1.0.0"
description: "What this kit does"
author: "Your Name"

targets:
  python:
    module: "my_custom_kit"
    dependencies:
      - "some-package>=1.0"

Place your Python module under src/ in the kit directory, then install:

beddel kit install ./my-custom-kit/

Links

• Python SDK (beddel-py) — full documentation, quickstart, architecture, examples
• PyPI
• Bug Tracker

Stay in the Loop

We write about agent-native patterns, workflow design, and SDK development on Substack.

License

MIT