cascadeflow: model tool for Hermes Agent

In-process runtime that routes agent model calls by cost, latency, quality and policy

What cascadeflow does

cascadeflow sits inside the agent execution loop rather than at the HTTP boundary like an external proxy. It can make a model decision at each step, gate budgets per tool call, and apply four runtime actions: allow, switch_model, deny_tool and stop. It also accumulates information from every model call, tool result and quality score, and records decisions so they can be audited. It is available as a Python package and as TypeScript packages, with integrations for LangChain, OpenAI Agents SDK, CrewAI, PydanticAI, Google ADK, n8n, Vercel AI SDK, OpenClaw and Hermes Agent.

The Hermes Agent integration supports per-skill model cascading, task-complexity cascading, topic-aware subagent cascading, an observe-mode rollout and auditable decisions. According to the README it does not take over provider credentials, base URLs, fallback chains or API modes, so those settings are left alone. Setup details are on the integration page of the cascadeflow documentation.

Key features

Per-step model decisions inside the agent loop
Per-tool-call budget gating and four runtime actions: allow, switch_model, deny_tool and stop
Hermes Agent integration with per-skill, task-complexity and subagent-topic cascading
Observe-mode rollout and auditable decisions
Python and TypeScript packages
Integrations for LangChain, CrewAI, OpenAI Agents SDK, Google ADK, n8n and more

When to use it

Route simple tasks to smaller models and escalate harder ones in a Hermes Agent setup
Try cascading in observe mode before letting it change model choices
Cap spend per tool call and stop a runaway agent loop

Who it is for: Developers who run agents in production and want runtime control over model choice, spend and policy.

How it fits with Hermes Agent

Has a dedicated Hermes Agent integration listed in its README, and supports Hermes among many other agent frameworks.

How to install cascadeflow

These commands are copied from the project's README. Check the repository for the latest steps before you run them.

pip install cascadeflow
npm install @cascadeflow/core

FAQ

What is cascadeflow?

cascadeflow is an in-process intelligence layer for AI agents. It optimizes cost, latency, quality, budget and policy decisions inside the agent loop instead of at the HTTP boundary.

Does cascadeflow work with Hermes Agent?

Yes, it provides a Hermes Agent integration for per-skill, task-complexity and subagent-topic model cascading, with an observe mode and auditable decisions. It does not take over provider credentials, base URLs, fallback chains or API modes.

How do I install cascadeflow?

Run pip install cascadeflow for Python or npm install @cascadeflow/core for TypeScript. The Hermes Agent integration is described on the cascadeflow documentation site.