Skip to content

Deployment

When you integrate Sondera Harness into your agent, you need to choose where policy evaluation happens. This page helps you understand your options and pick the right one for your situation.

What you're deploying: Your agent code stays the same either way. You're choosing where your policies live and where evaluation happens:

  • Platform: Policies are managed in Sondera (cloud, self-hosted, or VPC) with built-in tooling
  • Local: Policies are defined in your code and evaluated on the same machine

Both options use the same policy language (Cedar) and the same harness interface. You can start local and move to Platform later without changing your agent code.

Which Should I Choose?

  • Platform

    Choose Platform (SonderaRemoteHarness) when you need:

    • Policy management across multiple agents
    • Pre-built policy packs for common use cases
    • Audit logs and compliance reporting
    • Team collaboration and access control
    • Air-gapped or on-prem deployment options

    Platform setup

  • Local

    Choose local (CedarPolicyHarness) when you want:

    • Quick experimentation without an account
    • A handful of policies you manage yourself
    • Zero external dependencies

    Local setup

Platform (SonderaRemoteHarness)

Platform evaluation sends requests to Sondera (cloud, self-hosted, or VPC). Policies are managed through the web UI or API.

Getting Started

  1. Sign up at sondera.ai with Google or GitHub
  2. Create an API token in Settings → API Tokens
  3. Set your credentials via environment variables:
export SONDERA_HARNESS_ENDPOINT="your-harness.sondera.ai:443"
export SONDERA_API_TOKEN="<YOUR_SONDERA_API_KEY>"

Or create a .env file or ~/.sondera/env:

SONDERA_HARNESS_ENDPOINT=your-harness.sondera.ai:443
SONDERA_API_TOKEN=<YOUR_SONDERA_API_KEY>

Setup

from sondera import SonderaRemoteHarness, Agent

# Create harness (reads SONDERA_API_TOKEN from environment)
harness = SonderaRemoteHarness()

# Define your agent
agent = Agent(
    id="my-agent",
    name="My_Agent",
    description="What this agent does",
)

# Start a trajectory
await harness.initialize(agent=agent)

That's it. Policies are managed in the Platform web UI.

Local (CedarPolicyHarness)

Local evaluation runs entirely on your machine. Policies are defined in your code or loaded from files.

Setup

from sondera.harness import CedarPolicyHarness
from sondera import Agent

# Define Cedar policies
policies = '''
@id("forbid-dangerous-bash")
forbid(
  principal,
  action == Coding_Agent::Action::"Bash",
  resource
)
when {
  context has parameters &&
  (context.parameters.command like "*rm -rf /*" ||
   context.parameters.command like "*mkfs*" ||
   context.parameters.command like "*dd if=/dev/zero*" ||
   context.parameters.command like "*> /dev/sda*")
};
'''

# Generate Cedar schema from the agent
schema = agent_to_cedar_schema(my_agent)

# Create local policy engine
harness = CedarPolicyHarness(
    policy_set=policies,
    schema=schema,
)

await harness.initialize()
# Use same adjudication API as RemoteHarness

For larger policy sets, load from .cedar files:

from pathlib import Path

policies = Path("policies/agent.cedar").read_text()
harness = CedarPolicyHarness(
    policy_set=policies,
    schema=schema,
)

Switching to Platform

Both harnesses share the same interface. When you're ready to move to Platform, it's a one-line change:

# Before
harness = CedarPolicyHarness(policy_set=policies, agent=my_agent)

# After
harness = SonderaRemoteHarness()

Configuration Reference

Environment Variables

Variable Description Default
SONDERA_HARNESS_ENDPOINT Platform gRPC endpoint harness.sondera.ai:443
SONDERA_API_TOKEN Your API token Required for Platform

Platform Options

Parameter Type Description
sondera_harness_endpoint str Platform endpoint (overrides env var)
sondera_api_key str API token (overrides env var)

Local Options

Parameter Type Description
policy_set str Cedar policies as a string
schema str Cedar schema (use agent_to_cedar_schema() to generate)

Next Steps