DAG Topology
Define agent workflows as directed acyclic graphs. Sequential, parallel, and hierarchical patterns — all first-class citizens in Diaflow.
Platform / Orchestration
Agent Orchestration
Graph-based multi-agent routing with conditional branching, looping, and parallel execution. Build any agent topology in code, run it reliably in production.
Built for production agent topologies
Retry & Fallback
Configurable retry policies per step. Fallback agents when the primary path fails. Circuit breakers to prevent cascade failures.
Subagent Delegation
Orchestrator agents can spin up and delegate to specialized subagents at runtime. Full trace and cost attribution per subagent run.
Define your topology in Python
The Diaflow SDK makes complex agent graphs readable. Every step is typed, every branch is explicit, and every run is traceable.
- Type-safe agent definitions
- Async-first, built on asyncio
- Automatic trace generation
- LLM-provider agnostic
pipeline.py
from diaflow import Pipeline, Agent
pipeline = Pipeline(name="research-pipeline")
planner = Agent("planner", llm="gpt-4o")
researcher = Agent("researcher", llm="claude-3-5-sonnet")
synthesizer = Agent("synthesizer", llm="gpt-4o-mini")
pipeline.chain(
planner >> researcher >> synthesizer
)
result = pipeline.run(query="market analysis")Build your first agent pipeline
Free tier. No infra setup required.