Agyn is an open-source, Kubernetes-native AI agent management platform that lets organizations deploy and govern AI agents safely in production — with security, budgets, and access controls built in.
What is Agyn?
Agyn is a management layer for AI agents that takes them off employee laptops and runs them safely across the whole organization. It accepts any agent (Claude, Codex, custom builds) and any model as input, and produces deployed, governed agents with isolated sandboxes, least-privilege policies, and per-team spend caps. The platform is self-hosted (Kubernetes) or cloud, and is built by Agyn (agyn.io). It achieves 72.2% issue resolution on SWE-bench Verified — #1 among GPT-5-based systems (read the post).
Key Features
- Multi-environment deployment — Deploy agents into private networks (VPN, VPC) behind firewalls in minutes; supports instant rollback and private network access to internal services.
- Least-privilege policy enforcement — Every agent runs under static policies with a policy agent that inspects every tool call before execution; secrets stay hidden from the model, defending against prompt injection and sensitive data leaks.
- Per-agent cost tracking and budgets — Track spend across agents, teams, and workflows; set monthly budget limits, get usage alerts, and attribute costs to individual agents.
- Role-based access and audit logs — Give the right employees access to the right agents, share agents safely across teams, and keep a complete audit trail of who used what.
- GitOps agent and policy management — Declare agents, sandboxes, MCPs, skills, and prompts in code (Terraform-style
.tf files); agents run consistently across every team and environment via git clone and apply.sh.
Who is it for?
- IT and Security teams — Deploy agents into corporate networks behind VPNs and firewalls, enforce static policies, and prevent shadow AI by keeping every action audited and scoped.
- Engineering managers — Define agent configurations in Git, manage role-based access, and maintain oversight of agent behavior across teams.
- Finance and operations — Set per-team budget limits, monitor daily spend, and receive alerts when budgets are exceeded (e.g., Customer Success team over its $600 monthly limit).
- Non-technical team members — Use deployed agents (Code Reviewer, Data Analyst, Support Agent) without needing to configure infrastructure — agents are shared via access control and ready to use.
What can you do with Agyn?
- Deploy a Data Analyst agent — Give a Claude Opus agent read-only access to your production database, set a $2,000 monthly budget for the Data Science team, and let analysts ask for Q1 sales reports without exposing credentials.
- Set up a Code Reviewer — Deploy a Codex GPT-5 agent that reads repos, comments on PRs, and cannot email external addresses (policy drops that action); monitor real-time policy gate decisions (247 allowed, 18 blocked today).
- Roll out a Support Agent — Run a Claude Haiku agent with access to tickets, repos, and email (internal domains only), enforce a $600/month budget per team, and get alerts when spend exceeds limit — then increase the limit from the dashboard.
How does Agyn work?
- Clone the bootstrap repo (
git clone https://github.com/agynio/bootstrap) and run ./apply.sh -y to deploy the platform on Kubernetes.
- Define agents, sandboxes, MCPs, and policies in Terraform-style
.tf files (e.g., agents.tf, policy.tf) stored in Git.
- Apply the configuration — resources (agents, policies, MCPs) are created automatically. Agents run in isolated sandboxes inside your VPC or VPN. Every tool call is checked by the policy agent before execution.
