Early access · AI agent control plane

Deploy AI agents to your own cloud.

Connect a GitHub repository, validate beanshost.yaml and deploy to Vercel, Render, AWS or Azure. Every agent gets its own API, and BeansHost keeps each release, run, secret and approval in one control plane.

  • Runs in your cloud account
  • Pinned to exact commits
  • Prompts never stored
app.beanshost.com
AAcme Inc.
Build
Ship
Operate
Govern
Deployment

support-triage

production · Amazon ECS Express Mode

Active
Regioneu-west-1
Versionv1 · a41c9e2
Health/health
Deployment logsLive
  1. Environment variables synced
  2. Image built with AWS CodeBuild
  3. Image pushed to Amazon ECR
  4. Health check passed
  5. Deployment active
Deploys into the account you already own
  • VercelVercel Edge Network
  • RenderRender managed services
  • AWSAmazon ECS Express Mode
  • AzureAzure Container Apps
Why teams choose BeansHost

Ship agents faster, without giving up control.

Your team writes the agent. BeansHost takes care of everything between the repository and production, so you spend your time on what the agent does instead of how it gets deployed.

Faster launches

Go from repository to a live agent in one session.

No deployment scripts, Dockerfiles or manifests to write by hand. BeansHost generates what is missing, builds the agent and only lets it go live once it is healthy.

  • beanshost.yaml generated
  • Dockerfile generated
Lower, clearer costs

Keep your cloud account, your credits and your bill.

Agents run in the Vercel, Render, AWS or Azure account you already pay for, so existing discounts, credits and billing alerts still apply. BeansHost is a flat plan on top.

  • No markup on compute
  • From $5 per month
No lock-in

Describe the agent once and deploy it anywhere.

One manifest works across four providers. Moving an agent from Render to AWS is a deployment choice, not a rewrite, and your release process stays the same.

  • Vercel
  • Render
  • AWS
  • Azure
Instant integration

Turn every agent into an API your products can call.

Each agent gets an HTTPS endpoint, keys limited to that agent, and ready-made code in curl, JavaScript and Python. Hand it to another team without sharing cloud access.

  • OpenAPI file
  • Execution IDs
Fewer bad releases

Know exactly what is live, and recover in one step.

Every release is pinned to a commit and checksum and checked against its health path before it takes traffic. On Team, roll back to the last healthy deployment.

  • Health-gated
  • Rollback on Team
Confidence at scale

See every run and every decision in one place.

Success rate, p95 duration, failures and usage for every agent, plus roles, an append-only audit log and production approvals. Prompts and responses are never stored.

  • Audit log
  • Approvals on Enterprise
Doing it yourself
  • Deployment scripts for each cloud provider
  • Secrets pasted into several dashboards
  • No record of which commit is live
  • Every caller talks to a different provider URL
  • Runs and failures scattered across provider logs
  • Anyone who can deploy can do everything
With BeansHost
  • One guided flow for Vercel, Render, AWS and Azure
  • Secrets encrypted once and synced for you
  • Every release pinned to a commit and checksum
  • One API per agent, with scoped keys
  • Runs, success rate and usage in one view
  • Seven roles, an audit log and approvals
How it fits your stack

Your code. Your cloud. One control plane.

BeansHost sits beside your infrastructure, not in place of it. It deploys from your repository into your own cloud account and gives your applications one stable way to call every agent.

Source
acme/support-triage
beanshost.yaml · main · a41c9e2
Your applications
Web app, backend, workflows
POST /agents/{id}/invoke
BeansHost control planeOrganization scoped
  • Agent registry
  • Deployments
  • Secrets
  • Agent API and keys
  • Runs and usage
  • Roles
  • Audit log
  • Approvals

Decides what runs, where, and keeps the record. Never stores prompts or responses.

Your cloud account
VercelVercel Edge Networkagent
RenderRender managed servicesagent
AWSAmazon ECS Express Modeagent
AzureAzure Container Appsagent
4cloud providers from one manifest
1API endpoint and key set per agent
15guided steps from sign-up to production
0prompts or responses stored
How it works

Six steps from repository to governed agent.

The console follows the same path: Organize, Connect, Define, Deploy, Operate and Govern. Each agent version keeps its commit, specification, deployment and runs connected.

  1. 01

    Organize

    Create an organization and a project. A development environment is added for you.

  2. 02

    Connect

    Sign in to GitHub, pick a repository and branch, and link the cloud account you deploy into.

  3. 03

    Define

    Validate beanshost.yaml, or let BeansHost AI write one, then register the agent at an exact commit.

  4. 04

    Deploy

    Add environment variables, review the summary and deploy while the build logs stream in.

  5. 05

    Operate

    Call the agent through its own API, then follow runs, failures and monthly usage.

  6. 06

    Govern

    Assign roles, review the audit log and approve production releases before they go live.

Walk through every screen, from sign up to the first production run.

The full workflow
Every agent gets an API

Call any agent from any application.

Choose API on an agent in the console and you get its HTTPS endpoint for Development and Production, keys limited to that one agent, and code ready to paste in curl, JavaScript or Python.

  • Send any JSON body up to 1 MiB, passed to the agent unchanged
  • Get the agent's answer with an execution ID and correlation ID
  • Download an OpenAPI 3.1 file for Postman, gateways and AI tool calling
  • Connect from Postman, API gateways, workflow tools or partner systems with one key
  • Works the same on Vercel, Render, AWS and Azure
Read the agent API reference
curl -X POST "https://app.beanshost.com/api/v1/runtime/agents/$AGENT_ID/invoke?environment=production" \
  -H "Authorization: Bearer $BEANSHOST_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"message": "Hello"}'
200 OK · response
{
  "executionId": "0b6c2f7e-6d2a-4a0e-9d55-2f4f1c9e8a10",
  "correlationId": "req_8f2d1c",
  "status": "succeeded",
  "data": { "reply": "The agent's own response" }
}
Inside the console

Everything after the deploy button.

Deployment is one step. BeansHost keeps the operating context around every agent, so your team can answer what changed, when, and why.

Executions

Every run, whoever made the call.

Runs through BeansHost and runs your agent reports itself, with status, duration, provider and error category. Never what was sent or answered.

Usage

Runs against your plan.

Runs, failures, run time and data by day, agent and project, with a warning at 80 percent.

Secrets

Set once, synced to your provider.

Environment variables per project and environment, encrypted with AES-256-GCM and never shown again.

Roles

Seven built-in roles.

Invite teammates by email and give each one exactly the access they need.

API keys

Keys shown once, stored as a hash.

Scope keys to calling agents or reporting runs, limit them to one agent, and set an expiry.

Audit log

An append-only record of every privileged action.

Members, keys, secrets, connections, deployments, governance and billing, including refused attempts. Filter, search and export as CSV.

Enterprise

Production waits for a person.

Releases pause in Awaiting approval until an owner or admin approves or denies them with a reason. Requesters can be stopped from approving their own release.

Deployment states, rollback and governance in detail.

Explore operations
Deploy targets

Four providers. One deployment history.

Each provider has guided setup, a deployment summary and build logs. Deployment history shows every agent across all four, so switching providers never means switching tools.

Vercel

Vercel Edge Network

Connect your Vercel account with OAuth. BeansHost creates the Vercel project and deploys the registered commit.

Render

Render managed services

Connect with a Render API key. Choose the runtime, plan, region and the build and start commands for each service.

AWS

Amazon ECS Express Mode

Connect a cross-account IAM role with an external ID, or access keys. BeansHost builds the image with CodeBuild, pushes it to Amazon ECR and runs it on ECS.

Azure

Azure Container Apps

Sign in with Microsoft Entra and choose a subscription. BeansHost prepares the resource group, registry and Container Apps environment, then builds with ACR Tasks.

Regions, build paths and setup for each provider.

Compare deploy targets
Built for the people who run agents

One workspace for builders, operators and admins.

Developers

From repository to a callable agent in one session.

Connect GitHub, generate or validate beanshost.yaml, deploy, then paste the ready-made API call into your app. No deployment scripts to write.

Developer reference
Platform and operations

Every deployment, run and secret in one place.

Watch deployment history across providers, follow runs and failures, manage environment variables and roll back to the last healthy release.

Operations
Organization admins

Control who can do what, and keep the record.

Invite members with roles, manage the plan through Stripe, export the audit log and approve production releases.

Security and access
Pricing

Simple plans. Your cloud bill stays yours.

Builder is $5 per month and covers everything you need to connect, deploy and call agents. Team adds production, rollback and members. Enterprise adds approvals.

Start operating

Builder

$5per month

For individual builders shipping their first agents.

1,000 runs per month, 60 calls per minute
  • GitHub and cloud account connections
  • Agent registry with beanshost.yaml validation and generation
  • Deploy to Vercel, Render, AWS and Azure
  • Development environment with health-checked releases
  • A dedicated API endpoint and API keys for every agent
  • Executions, usage and deployment history
Good forAgent developers, development workloads and first deployments
Choose Builder
Govern at scale

Enterprise

Customcustom billing

For organizations that need approvals and longer evidence trails.

Unlimited runs, 3,000 calls per minute
  • Everything in Team
  • Approval before production releases
  • Policy by risk level, with optional separation of duties
  • Evidence timeline for every decision
  • Audit log with 365 days of history
  • Architecture and support options
Good forLarger organizations and regulated production environments
Contact sales
Prices in USD. Creating an account is free. Your cloud provider bills infrastructure separately.Compare every limit
Straight answers

Know exactly where BeansHost fits.

What it does, where your agents run, what it records and what it costs.

Talk to the team

What is BeansHost?

BeansHost is a control plane for AI agents. You connect a GitHub repository, describe the agent in a beanshost.yaml file and deploy it to your own Vercel, Render, AWS or Azure account. BeansHost gives every agent an API endpoint and keeps deployments, runs, secrets, members, usage, approvals and audit events for your organization in one place.

Where do my agents run?

In your own cloud account. BeansHost deploys to Vercel, Render, Amazon ECS Express Mode or Azure Container Apps using the connection you authorize, so the infrastructure and its bill stay with you. BeansHost does not run your agent logic inside its control plane.

Which languages and frameworks can I deploy?

Any agent that answers HTTP requests and exposes a health path. JavaScript workers built on Web-standard APIs are packaged directly. Everything else runs as a container: if your repository has a Dockerfile, BeansHost uses it, and if not, BeansHost generates one for Node.js 22 or Python 3.12 with FastAPI, Flask or a plain script. No particular agent framework is required.

What is beanshost.yaml, and do I have to write it?

beanshost.yaml is the agent specification. It sets the runtime, entrypoint, health check, triggers, secrets, resilience, governance and observability settings. If your repository does not have one, BeansHost AI analyzes the repository, writes a starting file and commits it to the branch you chose so you can review and edit it like any other file.

How does my application call a deployed agent?

Every agent has its own HTTPS endpoint with Development and Production tabs in the console. Create an API key for that agent, then POST any JSON body to the endpoint. BeansHost applies your plan's rate limits, forwards the body unchanged and returns the agent's answer with an execution ID. The console gives you ready-made curl, JavaScript and Python snippets and an OpenAPI file.

Can I connect my agent from a third-party platform?

Yes. Any platform that can send an HTTPS request with a header can call your agent: your own backend, API clients such as Postman, API gateways, workflow and automation tools, partner systems or another agent. Create a key limited to that agent in the console, store it as a secret on the platform, and POST JSON to the agent's endpoint. You can also import the agent's OpenAPI 3.1 file instead of configuring the call by hand. Keep the key on a server, since browser requests from other websites are blocked.

What does BeansHost record about each run?

Metadata only: status, duration, provider, error category, byte counts and correlation IDs. Prompt and response content is never stored. Executions shows success rate, p95 duration and the most common failure, and Usage shows runs against your plan for the month.

Is there a free plan?

Creating an account and an organization is free. Connecting a repository or cloud account, registering agents, storing secrets, calling agents and deploying need the Builder plan at $5 per month or higher. Your cloud provider bills the infrastructure your agents use separately.

Which source control providers are supported?

GitHub is supported today, including private repositories. GitLab, Bitbucket and Azure DevOps are shown as coming soon in the console. Deployments are started from the console or the API; deploy on push is not available yet.

How are production releases governed?

Every member has one of seven roles, and privileged actions are written to an append-only audit log that you can filter and export as CSV. On Enterprise, production releases wait for approval from an owner or admin, who approves or denies with a reason. You can require that the person who requested a release cannot approve it.

Can I use a custom domain?

Not yet. Each agent is served on its provider URL and called through its BeansHost endpoint, and custom domains are not available in BeansHost today. You can still put your own gateway or domain in front of the BeansHost endpoint.