The Package Registry & Control Plane for AI Agents
Stop copying and pasting prompts between laptops. AgentHub publishes, cryptographically signs, versions, and deploys AI agents, skills, and tools across your entire enterprise engineering organization.
Enterprise Control for Modern AI Engineering
Built for engineering leaders who need disciplined version control, cryptographic supply chain security, and compliance across all AI workflows.
Publish AI prompts, agent personas, skill toolsets, and slash-commands as immutable, content-addressed bundles. What npm and crates.io do for code, AgentHub delivers for AI agents.
Every asset bundle is signed with enterprise cryptographic keys. Consumers and CI pipelines verify publisher identity and SHA-256 digests before anything touches developer machines.
Pin exact asset versions, manifest digests, and tool configurations in version-controlled lockfiles to ensure byte-identical agent behavior across developer laptops and CI/CD runners.
Install published skills and agents into any agent harness with a single command: Claude Code, OpenCode, Cursor, Codex CLI, and GitHub Copilot each get the assets formatted for their native conventions.
Granular namespace isolation, permission manifests, and complete audit logging of who published, approved, or downloaded each agent asset, ensuring full compliance with AI governance frameworks.
Federate between private internal hubs and export self-contained archive bundles across secure air gaps without leaking prompts or proprietary tools to public cloud registries.
One Unified CLI: `agenthub`
Define the skills and agent personas your project requires in `agenthub.json`. The CLI resolves exact dependencies, verifies cryptographic signatures, and generates the target layout for your team's coding agent of choice.
- Immutable bundle downloads verified before writing to disk
- Synchronizes prompt updates across entire engineering fleets
- Connect API for automated CI/CD publishing and validation
How AgentHub Compares
Compare AgentHub against ad-hoc AI prompt management and unvetted tool sprawl.
| Feature | AgentHub (PiLAB) | Ad-Hoc / Shadow AI Tooling |
|---|---|---|
| Asset Management | Centralized package registry with content-addressing & immutable versions | Prompts & skills copy-pasted in Slack, wiki docs, or untracked local folders |
| Security & Provenance | Cryptographic signatures, permission manifests, SHA-256 digest checks | Zero provenance: impossible to tell who wrote or modified a running prompt |
| Reproducibility | Deterministic lockfiles (`agenthub.lock.json`) guarantee byte-identical execution | Silent drift across team members running different unversioned prompt files |
| Multi-Tool Support | One command installs across Claude Code, OpenCode, Cursor, and Copilot | Manual reformatting required for each IDE and AI CLI tool layout |
| Governance (EU AI Act / NIS2) | Immutable audit trail, role-based publish permissions, air-gap ready | Zero compliance visibility; shadow AI tool usage unchecked across organization |
| Hosting & Sovereignty | 100% Self-hosted Go service (`agenthub.pilab.hu`), on-premise or private cloud | Vendor lock-in on proprietary SaaS platforms with third-party prompt ingestion |
Govern Your Enterprise AI Assets with AgentHub
Bring disciplined software engineering practices to your AI agent adoption. Deploy AgentHub internally, eliminate prompt drift, and maintain full compliance.