Intent-Aware Governance for Autonomous AI Agents

Synentra is an open-source governance gateway designed to control how autonomous AI agents interact with enterprise APIs.

It sits between AI agents and downstream services, evaluates the intent and risk behind each request, and makes an explicit governance decision before the action reaches the target API.

Allow. Review. Deny.

Traditional API gateways understand things like:

• HTTP methods
• Routes
• Authentication
• Rate limits
• Network policies

But autonomous agents introduce a different question:

What is the agent actually trying to do?

Two requests can use the same endpoint, HTTP method, and credentials while representing very different intentions.

Synentra adds an intent-aware governance layer specifically for this problem.

• Intent-Aware Governance: Classify what an AI agent is attempting to accomplish rather than relying only on HTTP-level attributes.
• Risk & Trust Evaluation: Combine request risk, agent behavior, historical signals, and semantic intent before making an authorization decision.
• Policy Enforcement: Define deterministic governance policies using Synentra policies or Open Policy Agent (OPA).
• Human-in-the-Loop: Route high-impact or uncertain actions to authorized reviewers instead of automatically allowing or denying them.
• Agent Identity: Register agents, issue credentials, and integrate with external identity providers and JWT-based authentication.
• Auditable Decisions: Maintain a traceable decision pipeline from the original agent request through intent classification, policy evaluation, approval, and final execution.
• Local Intent Classification: Run intent classification locally using DistilBERT + ONNX Runtime, without requiring every request to be sent to an external LLM.

Run Synentra with Docker:

docker run --rm -p 7080:7080 ghcr.io/synentra/synentra:latest

Then check the gateway:

curl http://localhost:7080/health

Explore the documentation at synentra.io/dcos.

The core intent-aware governance gateway. Built with .NET.

Command-line tooling for working with Synentra deployments.

Client libraries for integrating applications and autonomous agents with Synentra.

Models and tooling for intent classification using DistilBERT, ONNX Runtime, and Synentra's intent taxonomy.

A typical request passes through:

flowchart TD
    A["Agent Identity"] --> B["Intent Classification"]
    B --> C["Risk & Trust Evaluation"]
    C --> D["Policy Evaluation"]

    D --> E["ALLOW"]
    D --> F["REVIEW"]
    D --> G["DENY"]

    E --> H["API"]
    F --> I["Human Approval"]
    I --> H

The goal is not to replace existing API gateways, IAM systems, or WAFs.

Synentra complements them by adding an agent-aware governance layer focused on intent, risk, trust, and accountable execution.

Synentra is designed for teams building:

• Autonomous AI agents
• Agentic workflows
• Enterprise AI platforms
• MCP-based integrations
• AI-powered infrastructure automation
• Security-sensitive API automation
• Multi-agent systems
• Internal AI developer platforms

Synentra is open source under the Apache License 2.0.

We welcome:

• Bug reports
• Feature requests
• Documentation improvements
• Architecture discussions
• Integrations
• Pull requests
• Security research
• Community feedback

If you're interested in AI-agent governance, we'd love to have you involved.

• 🌐 Website: synentra.io
• 📖 Documentation: synentra.io/docs
• 💻 GitHub: github.com/synentra
• 💬 Discussions: GitHub Discussions
• 🐛 Issues: GitHub Issues

Contributions are welcome.

If you're new to the project, a good place to start is:

1. Explore the documentation.
2. Run Synentra locally.
3. Check open issues and discussions.
4. Share feedback about your AI-agent governance use case.
5. Submit an issue or pull request.

See the contribution guidelines in the main repository for details.