Enforce Mode — Active Runtime Control

Enforce mode moves from observation to action. Budget caps, tool call limits, and compliance rules become hard constraints that shape agent behavior in real time.

Switch to enforce

import cascadeflow

cascadeflow.init(mode="enforce")  # Changed from "observe" to "enforce"

Add a budget cap

Wrap any block of agent work with cascadeflow.run() and set a budget:

cascadeflow.init(mode="enforce")

with cascadeflow.run(budget=0.50) as session:
    result = await agent.run("Research and summarize this topic")

    summary = session.summary()
    print(f"Cost: ${summary['cost_total']:.4f}")
    print(f"Budget remaining: ${summary['budget_remaining']:.4f}")
    print(f"Steps completed: {summary['steps']}")

If the agent exceeds $0.50, cascadeflow issues a stop action. The agent halts cleanly — no runaway spend.

Combine budget with tool call limits

with cascadeflow.run(budget=1.00, max_tool_calls=5) as session:
    result = await agent.run("Search and analyze this dataset")
    # Stops when either budget OR tool call limit is hit first

Add compliance gating

Restrict which models can process sensitive data:

with cascadeflow.run(budget=1.00, compliance="gdpr") as session:
    result = await agent.run("Process EU customer feedback")
    # Only GDPR-approved models are allowed — non-compliant models are switched

See what happened

The decision trace shows every enforcement action:

for record in session.trace():
    print(f"Step {record['step']}: {record['action']} — {record['reason']}")
    # Step 1: allow — budget ok, compliance passed
    # Step 3: switch_model — model not in GDPR allowlist
    # Step 7: stop — budget exceeded ($0.50/$0.50)

In enforce mode, record['applied'] is True — actions are executed, not just logged.

Gradual Rollout

You do not need to jump from observe to full enforcement. Start with one constraint:

# Week 1: Just budget — see if anything would stop
cascadeflow.init(mode="enforce")
with cascadeflow.run(budget=5.00) as session:  # Generous cap
    ...

# Week 2: Add tool call limits
with cascadeflow.run(budget=2.00, max_tool_calls=20) as session:
    ...

# Week 3: Add compliance
with cascadeflow.run(budget=1.00, max_tool_calls=10, compliance="gdpr") as session:
    ...

TypeScript

import { CascadeAgent } from '@cascadeflow/core';

const agent = new CascadeAgent({
  models: [
    { name: 'gpt-4o-mini', provider: 'openai', cost: 0.000375 },
    { name: 'gpt-4o', provider: 'openai', cost: 0.00625 },
  ],
  quality: {
    threshold: 0.8,
    useSemanticValidation: true,
  },
});

const result = await agent.run('Explain quantum computing');
console.log(`Model: ${result.modelUsed}, Saved: ${result.savingsPercentage}%`);

Run this example: examples/enforcement/basic_enforcement.py | examples/user_budget_tracking.py

Next Step

Ready to attach policy to individual agents? Use the @agent decorator →

Overview

Getting Started

Core Concepts

Harness

Integrations

Guides

Resources

Quickstart: Enforce Mode

Enforce Mode — Active Runtime Control

Gradual Rollout

TypeScript

Next Step

​Enforce Mode — Active Runtime Control

​Gradual Rollout

​TypeScript

​Next Step

Enforce Mode — Active Runtime Control

Gradual Rollout

TypeScript

Next Step