Brandon Bennett

Self-hosted, Docker-native AI control plane: zero-egress, hash-chained audits. Apache-2.0; optional commercial support.

[Brandon Bennett @agentsystems]

@kempsterrrr @tomismeta Thanks @kempsterrrr / @tomismeta - just announced the @LangChain integration here: https://t.co/MW2ye4z1Ka

[Brandon Bennett @agentsystems]

Today's AI systems lack accountability.

In finance, we have independent auditors.
In legal, we have chain of custody.
In AI, we have "just trust us."

Introducing AgentSystems Notary - independently verifiable logging for AI.

Now a @LangChain integration.

[Brandon Bennett @agentsystems]

@chamath @chamath - Most AI companies are doomed. They are building for the wrong future. This inversion is inevitable: 1) local models/hardware keep improving and 2) building AI is becoming trivial - The challenge will become discovery and trust. I'm fixing that. https://t.co/1LgxeUeQpi

[Brandon Bennett @agentsystems]

AgentSystems: Self-host third-party AI agents. No gatekeepers. https://t.co/mLJTtRelY0 via @producthunt

[Brandon Bennett @agentsystems]

10/ The infrastructure we build now determines the future.

Self-hosted infrastructure for AI agents exists now. The agent index is mostly empty, but someone has to build the rails so the trains can run.

If you care about data sovereignty and open infrastructure, this is for you.

[END]

[Brandon Bennett @agentsystems]

1/ You wouldn't build your own email client or maps app - you'd download one.

AI agents should work the same way.

I spent a year building a self-hosted app store for AI agents. Here's why it matters:

[Brandon Bennett @agentsystems]

9/ Looking for:
• Agent builders to publish to the index
• Security researchers to review the architecture
• Organizations that need self-hosted AI infrastructure
• People who give a shit about data sovereignty

GitHub: https://t.co/1LgxeUeQpi

[Brandon Bennett @agentsystems]

8/ Apache-2.0 open source.

Pre-release but functional.

Deploy on your infrastructure (not someone else's cloud). The infrastructure we build now determines whether we're locked into centralized platforms or not.

[Brandon Bennett @agentsystems]

7/ Built for organizations that need AI for sensitive workflows but can't:
- Send data to third parties (compliance, privacy)
- Build everything in-house (resource constraints)
- Trust janky workflows built in no-code tools (technical debt)

[Brandon Bennett @agentsystems]

6/ Works with:
- Ollama (fully local, no cloud APIs)
- Cloud providers (Anthropic, OpenAI, Bedrock)
- Hybrid (sensitive data local, complex tasks cloud)

Same agent code, your choice of backend.

[Brandon Bennett @agentsystems]

5/ Architecture highlights:
• Federated Git-based index (fork = your own marketplace, no gatekeepers)
• Per-agent egress allowlists (you configure which URLs each agent can access)
• Model abstraction (agents specify "gpt-oss:20b", you configure Ollama/Bedrock/etc)
• Thread-centric storage (isolated artifacts per invocation)

[Brandon Bennett @agentsystems]

4/ The platform works. The agent index is mostly empty. But someone has to build the rails so the trains can run. That's what we did.

[Brandon Bennett @agentsystems]

3/ AgentSystems: self-hosted app store for AI agents Works on laptop, private cloud (AWS/Azure/GCP), or air-gapped. Key features:

📦 Container isolation (each agent sandboxed)
🚦 Egress proxy (default-deny networking)
🔐 Credential injection (keys configured on host)
📋 Hash-chained audit logs
🌐 Federated Git-based discovery

[Brandon Bennett @agentsystems]

2/ The problem: most organizations either build every agent in-house or send their data to third-party servers.

Neither scales.

We need infrastructure that lets you discover agents built by others and run them on YOUR infrastructure (private cloud, on-premises, or local).

[Brandon Bennett @agentsystems]

Watch the demo & jump in

This is Day One of an open agent ecosystem. Build agents from the template, file issues, and help shape the roadmap.
🎥 Video: https://t.co/ZbQOh39ytL
📚 Docs: https://t.co/sfM8zcuTHp
#AI #Agents #OpenSource #SelfHosted #Privacy

[Brandon Bennett @agentsystems]

Try to break it

Unapproved network call? Blocked by policy.
Audit log edited? Chain check shows tampering.
Bring your toughest tests - we'd rather find edge cases together.

#AppSec #RedTeam

[Brandon Bennett @agentsystems]

See it work

UI discovers agents → invoke with inputs/files → outputs land in thread-scoped artifacts. Observe and repeat. Github: https://t.co/fn7uh8RZiM

#LLM #MLOps

[Brandon Bennett @agentsystems]

Install in one command

Spin it up fast: gateway + UI + discovery of containerized agents. Start testing in minutes, not days. Learn more at our YouTube channel: https://t.co/lGysFxYjwJ
#DX #BuildInPublic

[Brandon Bennett @agentsystems]

Zero-trust, in practice

Four pillars:
1. Credential injection at runtime (agents don’t store your keys)
2. Egress allowlist (block unapproved URLs)
3. Cryptographic audit chain (tamper-evident logs)
4. Container isolation (each agent sandboxed)
#Security #DataSovereignty

[Brandon Bennett @agentsystems]

Run it anywhere...

Homelab PC, 4090 desktop, or a small VM in your cloud - AgentSystems is portable and self-hosted. No built-in telemetry. Apache-2.0. https://t.co/x5BUCfEVIR
#Homelab #DevOps