OPEN-SOURCE CONTROL LAYER / PYTHON 3.10+

Govern AI agents before they act.

Define what your AI agents can and cannot do before they act. Lightweight, open-source, and framework-agnostic.

policy trace / 001 runtime ready
Hlinor policy decision trace A policy file flows through validation and an action check before an allowed or blocked result is recorded. source agent.yaml check policy result allowed result blocked declarative pre-dispatch audit signal

One reviewable contract. One decision before side effects.

599

automated tests

3.10–3.13

Python CI matrix

Apache 2.0

open-source license

A policy is a file.
A decision is a line.

Keep governance close to the code, readable in a pull request, and enforceable at runtime.

my_agent.yaml YAML / from hlinor-registry init
id: my-agent
type: agent
name: My First Agent
department: engineering

allowed_actions:
  - read_database
  - read_database:reports/*
  - refund_payment:order/*

blocked_actions:
  - send_external_email

policies:
  - refund-needs-approval

enforcement_mode: strict
runtime / policy decision pre-dispatch
$ hlinor-registry compile --manifest registry.yaml --output bundle.json
Successfully compiled 2 entries to bundle.json
Agent 'my-agent' enforces: refund-needs-approval
$ hlinor-registry check --bundle bundle.json --agent my-agent --action read_database
ALLOWEDEXPLICITLY_ALLOWED
action is on the agent allowlist
$ hlinor-registry check --bundle bundle.json --agent my-agent --action send_external_email
DENIEDACTION_BLOCKLISTED
blocked always wins over allowed
receipt every check emits a Decision ID and timestamp

Governance that answers the hard questions early.

01 / PRIVACY

Prevent PII leaks

Block external email, record mutation, or other exfiltration paths for agents handling sensitive data.

02 / COST

Control API costs

Keep search counts, premium calls, budget checks, and rate limits visible beside the agent definition.

03 / EVIDENCE

Ensure compliance

Build an audit trail around approvals, execution receipts, policy decisions, and evidence-bound claims.

A clear path from intent to action.

Registry files become a shared contract between the people who design agent capabilities and the systems that execute them.

Hlinor registry architecture Developer writes a YAML registry, the policy engine validates an action, and the agent receives an allowed or blocked result. 01 Developer intent 02 YAML registry contract 03 Policy engine decision 04 Agent action side effect blocked + audit record Hlinor registry mobile architecture A vertical flow from developer intent through the YAML registry and policy engine to an allowed or blocked agent action. 01 / INTENT Developer 02 / CONTRACT YAML registry 03 / DECISION Policy engine 04 / SIDE EFFECT Agent action blocked + audit

Make your next agent easier to trust.

Start with a YAML contract, add a policy check before the first side effect, and invite your security team into the review.

OPEN SOURCE / EARLY ACCESS

HlinorAI works with teams designing internal agent platforms, compliance workflows, and production action boundaries.

Open an issue or start a discussion ↗