Agent Orchestration Framework
Orchflow
A dependency-free Python framework for readable multi-agent pipelines: sequential, parallel, conditional, and resumable flows.
0 deps
zero runtime dependencies in core
Orchflow live demo
live Flow.events() stream
Runtime graph
Ready to run.
Final
No output yet.
Inspector
No event selected.
Watch the complete Orchflow demo
A six-minute walkthrough of the same project. Readable Python steps, parallel branches, live events, retries, traces, and JSON checkpoint resume.
A live look at readable multi-agent workflows
Readable orchestration
The live demo is still a plain Flow([...]) list: plan, a parallel group, synthesis, a condition, a finalizer.
Parallel work you can inspect
All three research agents share a parallel_group_id, so the UI can show they fanned out and came back together.
Resume is visible
The failure mode writes a JSON checkpoint, reloads it, and appends new traces after the resume.
Safe live model demo
You can tweak small inputs and model presets. The backend still enforces allowlisted models and rate limits.
The backend is still plain Python
The UI here is only a viewer. Your step functions stay readable, and the framework handles fan-out, retries, lifecycle events, checkpoints, and resume.
from orchflow import Agent, Flow, StepContext, condition, step@step(name="technical_research", retry=2)async def technical_research(input: dict, context: StepContext):return await researcher.run(prompt, context=context)
flow = Flow([plan,[market_research, technical_research, risk_review],synthesize,condition(when=lambda ctx: ctx.previous["quality_score"] >= 0.8,then=publish_ready,otherwise=revise,),finalize,])
Install from PyPI
$ pip install orchflow$ pip install "orchflow[litellm]"
v0.5.0 · tag-based PyPI releases · core has zero runtime dependencies.