Skip to main content

​
Configure in code

from agent_sentinel import PolicyEngine

PolicyEngine.configure(
    session_budget=5.0,
    run_budget=1.0,
    action_budgets={"call_llm": 0.50},
    denied_actions=["send_money", "delete_database"],
    allowed_actions=None,
    rate_limits={
        "search_web": {"max_count": 20, "window_seconds": 60},
    },
    strict_mode=True,
)

​
Configure via YAML (callguard.yaml)

Agent Sentinel will search for callguard.yaml in:
  • ./callguard.yaml
  • ./.agent-sentinel/callguard.yaml
  • $AGENT_SENTINEL_HOME/callguard.yaml
Example: 
budgets:
  session: 5.0
  run: 1.0
  actions:
    call_llm: 0.50

denied_actions:
  - send_money
  - delete_database

rate_limits:
  search_web:
    max_count: 20
    window_seconds: 60

strict_mode: true
Load it: 
from agent_sentinel import PolicyEngine

PolicyEngine.load_from_yaml()

​
What happens on violation?

Policy checks run before your function executes:
  • Denied action / allowlist mismatch / rate limit → PolicyViolationError
  • Budget exceeded → BudgetExceededError
If you want to handle violations: 
from agent_sentinel import BudgetExceededError, PolicyViolationError

try:
    # call some guarded action
    ...
except BudgetExceededError:
    # degrade gracefully (switch models, stop the run, etc.)
    ...
except PolicyViolationError:
    # treat as security / safety block
    ...

## Tracking current spend

```python
from agent_sentinel import CostTracker

print(CostTracker.get_session_total())
print(CostTracker.get_run_total())

# Reset between agent runs
CostTracker.reset_run()


## Remote policy sync (platform)

You can also download and refresh policies from the platform:

```python
from agent_sentinel import PolicyEngine

PolicyEngine.enable_remote_sync(
    platform_url="https://platform.agentsentinel.dev",
    api_token="as_your_api_key_here",
    agent_id="my-agent",
    run_id="my-run",
    refresh_interval=300,
)