Boundwr Python SDK
Agent governance that runs anywhere. Register agents, enforce policies, log every action, and export compliance reports — all stored locally as JSON. No server required.
Installation
Up and running in seconds
Install from source
Requires Python 3.10+. Zero external dependencies.
Quick start
Five minutes to governance
Register agents, define policies, log actions, and export compliance reports — all running locally with zero configuration.
from boundwr import Boundwr
# Initialize — all data stored locally in .boundwr/
client = Boundwr(project="my-project")
# 1. Register an agent
agent = client.register_agent(
name="support-bot",
description="Handles tier-1 support tickets",
permissions=["read_tickets", "send_replies"],
owner="cs-team"
)
# 2. Define a policy — deny dangerous actions
client.define_policy("no-delete", rules=[
{
"action":"delete_user",
"deny": True,
"message":"Agents cannot delete users"
}
])
# 3. Check before acting
result = client.check_policy(agent["id"], "delete_user")
if not result["allowed"]:
print("Blocked!", result["violations"])
# 4. Log actions to the audit trail
client.log_action(agent["id"], "send_reply", details={ "ticket": "T-1234" })
# 5. Export compliance report
report = client.export_compliance_report()
print(report["summary"])Features
Everything you need for agent governance
Pure Python. No external dependencies. No server. Just import and go.
Agent Registry
Register agents with names, descriptions, permissions, and ownership. Every agent gets a UUID and timestamp.
Policy Engine
Define deny rules for dangerous actions. Check policies before execution and get detailed violation reports.
Audit Logging
Append-only, immutable log of every action with timestamps, agent IDs, status, and details. ISO 8601 everywhere.
Delegation Tracking
Record agent-to-agent delegations with reasons and permission grants. Full chain of authority.
Compliance Reports
Export JSON or CSV reports with agent inventory, full audit log, policy list, and status breakdowns.
Thread-safe
All file operations are guarded by a threading lock. Safe for multi-threaded applications and concurrent agents.
Storage
Local-first, transparent
All data is stored as plain JSON files in a .boundwr/ directory. No database, no cloud, no lock-in. Inspect and version-control your governance data.
API Reference
Every method, documented
Boundwr(project, api_key=None, storage_dir=".boundwr")Initialize the client. api_key reserved for future cloud mode.
register_agent(name, description, permissions, owner)Register a new agent. Returns dict with generated UUID and timestamps.
list_agents()Return all registered agents as a list of dicts.
log_action(agent_id, action, details=None, status="success")Append an action to the immutable audit log.
log_delegation(from_agent, to_agent, reason, permissions_granted)Record a delegation between agents with permissions granted.
define_policy(name, rules)Define a policy. Each rule needs action (str) and deny (bool).
check_policy(agent_id, action)Check if an action is allowed. Returns {allowed: bool, violations: [...]}.
get_audit_trail(agent_id=None, since=None)Query audit entries with optional agent and time filters.
export_compliance_report(format="json")Generate a full compliance report in JSON or CSV format.
Start governing
your agents today.
Download the Boundwr SDK, register your first agent, and have a compliance report ready in five minutes.