An AI app platform your IT can self-host, govern and audit
Business teams build apps and AI agents in Flightroom. It runs in your cloud or on-prem, signs people in with your identity provider, writes code to your Git, reviews it with your rules, uses your AI models and deploys through your Argo CD.
IT sets the guardrails once. Teams build inside them.
IT connects it once
Teams build inside it
Engineers keep review
Your infrastructure
Run Flightroom where your other systems run: on-prem or in any cloud. The hosted console is there to evaluate it, free to start.
Helm chart
Install on your own Kubernetes. Point it at your MongoDB and Redis: the chart does not bundle them.
Installation guideDocker Compose
Flightroom, MongoDB and Redis on one host, for a pilot or a small team.
Installation guideAll-in-one container
One image, lory1990/flightroom:all-in-one, with everything persistent in a single volume.
Installation guideTerraform and OpenTofu
Modules that run Flightroom, MongoDB and Redis on a Docker host, kept with the rest of your infrastructure code.
Installation guide- The console never holds credentials for your cluster. It commits to Git and talks to the Argo CD API, and Argo CD syncs the cluster.
- Argo CD is reached over HTTPS. Plain HTTP has to be switched on explicitly, for a local cluster.
- Every environment of a project is a namespace. With exposure on, it also gets its own host and TLS certificate.
Your identity and roles
People sign in with Microsoft Entra ID, Google or Auth0, or with local accounts. Then two layers of roles decide what they can reach and what they can change.
Organization roles
- Owner and Admin
- Administer the organization: AI models, members, and every project in it.
- Member
- Reaches only the projects they were invited to.
Project roles
- Owner and Member
- Build: create repositories, ask the AI for changes, deploy.
- Viewer
- Cannot ask for or apply AI changes, create or rename repositories, change storages or databases, deploy through GitOps, or use the write tools of the MCP server and the console assistant.
- A project membership can be limited to some functional domains, so a team sees and changes only its own apps.
- Scoped API keys with an expiry date let your CI register versions and deploy without a personal login.
- On the MCP server and in the console assistant, viewers get read tools only.
Your code and your process
Your repositories
Your review rules
Git as the audit trail
Your AI models
Organization admins add the models Flightroom may use, with your own keys. Run Claude on Azure AI Foundry or AWS Bedrock, or GPT on Azure OpenAI, and prompts and code go through the cloud contract you already have.
| Model | Through |
|---|---|
| Claude | Anthropic, Azure AI Foundry or AWS Bedrock |
| GPT | Azure OpenAI |
| Gemini | Gemini API |
- Restrict each project to some of the organization's models, or give it none and AI is off for that project.
- Mark one model as the default, and cap the reasoning effort a model may be asked for, to keep costs predictable.
- Token usage is recorded per user, per project and per model.
Guardrails on every AI change
- Every AI change lands on its own branch. The base branch stays untouched until someone merges or applies it.
- The agent cannot edit CI files or lockfiles, and cannot read credential files such as .env or keys.
- Each request is capped: 20 files, 500 KB and 40 model rounds.
- The AI does not run your code. Your CI builds and tests the change, as it would any other.
- In the console assistant, a deploy, a rollback or a delete waits for a confirmation card.
The full picture of how people build with AI is on Build with AI.
Audit trail
Assistant and MCP actions
AI usage
Deploys
Audit lines go to the server log as structured JSON, ready for the log platform you already run. There is no audit screen in the console.
Security, stated plainly
Stored credentials
Set SECRETS_ENCRYPTION_KEY and repository tokens, storage keys, database passwords, the Argo CD token, AI model keys and service secret values are encrypted in the database with AES-256-GCM, and the API stops returning them. Without the key they are stored in the clear, so set it on day one.
Encryption at restKubernetes Secrets in Git
Per-environment secrets are committed to the configuration repository as plain Kubernetes Secret manifests, base64 encoded, not encrypted. Keep that repository private and limit who can read it.
MCP server
Off by default on self-hosted installs. When you turn it on, clients sign in with OAuth 2.1 and PKCE, the user's role is checked again on every call, and each grant or API key is rate limited.
Telemetry and error reporting
A self-hosted install sends one anonymous daily ping: a random installation id, the version and a few aggregate counts. TELEMETRY_DISABLED=true or DO_NOT_TRACK=1 turns it off. Error reports go to Sentry only if you set a DSN.
Requests are rate limited per IP, with a lower ceiling for assistant questions. Run self-hosted installs on a private network or behind your identity-aware proxy.
Open source, no lock-in
Flightroom is open source under Apache-2.0, and the templates under MIT. Read the code your security review asks about on GitHub.
- The code of every app is in your Git provider.
- The manifests and Helm values of every environment are in your Git provider too.
- Argo CD is yours. If you stop using Flightroom, it keeps syncing from those repositories and your apps keep running.
What you need for a rollout
- Kubernetes and Argo CD 2.8+
- Argo CD reachable over HTTPS with an API token. Environments become namespaces in the cluster it manages.
- A Git provider
- GitHub, GitLab or Azure DevOps, for the app repositories and the GitOps repositories.
- An AI model subscription
- Anthropic, Azure AI Foundry, AWS Bedrock, Azure OpenAI or Gemini. One is enough to start.
- Exposure
- Traefik v3 with its Gateway API provider, the Gateway API CRDs, and cert-manager with a ClusterIssuer for TLS.
- Single sign-on
- Microsoft Entra ID, Google or Auth0. Optional: local accounts work too.
- To self-host
- MongoDB, Redis and an SMTP server for invitations and password resets, plus your own JWT_SECRET and SECRETS_ENCRYPTION_KEY.
Flightroom does not create clusters or databases, scale workloads or show pod logs, and it has no SAML or SCIM provisioning today. People are invited to organizations and projects from the console.