- Coming soon · Powered by Railtown AI
The agent you need is already running somewhere.
The Agent App Store is a catalog of AI agents already proven in real businesses. Install one in minutes: vetted, and yours to run on your own keys. Or build on it and list your own.
Minutes, not sprints
Browse and pick an agent, connect your accounts, go live. Nothing to rebuild from scratch.
Reviewed, not just listed
A Railtown engineer reviews every agent before it goes live. Nothing gets listed automatically.
Your keys, your tenancy
You connect your own credentials when you install. Nothing is shared from the agent’s creator.
HOW IT WORKS
Browse, install and make it yours.
The store is open to anyone, not just partners. Sign up, take any agent from the catalog, and use it as is or as a starting point for your own.
Browse
Find an agent that already does the job.
Install
Run it with your own credentials, in minutes.
Make it yours
Customize it, improve it, or build something new from it.
Looking to personalize and publish an agent from the store? See how publishing works.
What powers the store
Every agent in the store runs on Railtown AI’s agent observability and evaluations platform. It records each step an agent takes, so you can see how it reached an answer, which tools and functions it used, how long each step took, and what it cost in tokens. That record is what lets us review agents, connect them together, and give each team the right view of what’s happening.
COMPOSITION
Install one agent. Then wire it to the next.
Agents in the catalog are composable. One agent’s output becomes another’s input, so a pair does what neither could alone. Deploy them independently, then connect them on the Railtown AI platform.
Each agent still runs on its own secrets and its own footprint. Connecting them is configuration, not a rebuild.
ENTERPRISE VISIBILITY
Two views of the same agent. Neither sees the other's secrets.
A shared agent has two audiences with different questions. The Railtown AI platform splits visibility along that line. Developers get execution detail, enterprises get governance detail, and neither view reveals what belongs to the other side.
Developer view
The team that built and published the agent
SEES
- Execution health: runs, failures, retries, step-level outcomes
- Performance and latency across calls and tool invocations
- Errors and stack context, traced to the failing step
- Evaluations and regressions against prior versions
NEVER SEES
- Enterprise prompts and model responses
- Customer documents and business content
- Any tenant data flowing through the agent
Enterprise view
The organization running the agent on its own data
SEES
- Application: what's installed, where, at what version
- Usage, models invoked, and data classifications touched
- Actions taken and any external processing or egress
- Reliability and policy events, with an auditable trail
NEVER SEES
- Proprietary developer prompts
- Orchestration logic and agent internals
- Implementation detail behind the published behaviour
Both views read from the same data the agent already reports to the platform. The difference is what each role is allowed to see, not what gets collected.
- COMING SOON
Choose where your agents run
Sovereign cloud support is on the roadmap. You’ll be able to install a catalog agent into the region or environment your data residency rules require. It will be the same reviewed agent, deployed under your jurisdiction.
- Live
Managed by Railtown
Run your agents in the region of your choice. Available today.
- Live
Your own tenancy
Run the agent in your own cloud account, under your own controls.
PERSONALIZE & PUBLISH
Make an agent of your own then share it.
Take any agent from the store and customize it to fit your work. When it’s ready, submit it for expert review. Once approved, it’s listed in the store for others to install.
ELIGBILITY
To be eligible, an agent must be a self-contained container image built on the Railtown AI platform, ship with a deployment template, and run entirely on the installer's own stack and secrets. Each installer deploys and monitors it through their own Railtown AI platform account.
What the review checks
A Railtown engineer reads the agent’s source and checks it against a fixed checklist before it can be listed.
- Scope matches behaviour
- Secrets by reference
- Data stays in bounds
- Dependecies pinned
- Runs for someone else
- Secrets stay yours. Review runs in an isolated sandbox with no network egress and no access to your environment. Recipients supply their own credentials at install.
Be first to browse, and to publish.
The store opens to existing Railtown AI platform users first. Anyone can join the waitlist.
