AgentSystems - Discover and run self-hosted AI agents like apps
by•
Run third-party AI agents on your infrastructure without sending data to third parties.
AgentSystems is a self-hosted app store (Apache-2.0) for discovering and running agents built by others. Works on private cloud, on-premises, or local hardware with container isolation and federated Git-based discovery.
No gatekeepers. Fully forkable. Works with Ollama (local) or cloud APIs (Anthropic, OpenAI, Bedrock).
Replies
Aloha Product Hunt!
We spent the last year building AgentSystems: a self-hosted app store for AI agents that run on YOUR infrastructure (private cloud, on-premises, or local).
Why We Built This
You don't build your own email client or maps app — you download one. AI agents should work the same way.
But there's been no good way to run third-party agents without sending your data to third-party servers.
Most organizations either build every agent in-house (expensive, slow) or send data to cloud services (privacy/compliance nightmare).
1️⃣ Federated Discovery
Instead of a centralized marketplace, agent metadata lives in a Git-based index. Agent builders can publish, organizations can fork and run their own indexes. No gatekeepers.
2️⃣ Isolation Patterns
Container isolation (each agent in separate container)
You configure which external URLs each agent can access via egress proxy (default: none)
Your credentials are injected at runtime (not shared with agent builders)
Hash-chained audit logs
3️⃣ Model Abstraction
Agent builders declare model IDs ("gpt-oss:20b", "claude-sonnet-4.5", "llama3.3:70b"); you pick providers (Ollama, Bedrock, Anthropic, OpenAI).
Current Status
Pre-release with active development. Functional for testing.
What I'm Looking For
Early users interested in testing
Feedback on the federated index approach
Security researchers to review the architecture
Agent builders to publish to the index
Resources:
GitHub: https://github.com/agentsystems/agentsystems
Docs: https://docs.agentsystems.ai
Happy to answer questions! 🚀