Documentation
You bring a git repository containing agents and one elan.yaml. The platform builds it, provisions what it needs, deploys it into an isolated namespace and gives you an address to invoke it at. You do not write a Dockerfile, a Kubernetes manifest, a Terraform module or a CI pipeline.
Start here
- Platform overview — The four nouns everything else is derived from: a tenant, a project, a workload and a target. What each one owns, and which of them a caller ever sees.
- Quickstart — Install the CLI, scaffold a repository, give the platform your model key, deploy, and invoke. About fifteen minutes, most of it waiting for the first build.
- Try it in the console — What the console shows once you have deployed: the deploy and its logs, the Playground for talking to an agent, runs and their traces, and what each one cost.
- Call it from your code — Create a service user, put its id and secret in your environment, install the SDK, and run your agent from a backend — in Python or TypeScript.
Everything else
ConceptsHow the platform thinks about agents, sessions, runs, policy and money. Read these before you build much.GuidesDoing one thing well: an agent's tools and limits, a workflow's graph, secrets, approvals, plugins, and charging your own users.APIEvery route, with its request, its response and the refusals it can answer with.SDKThe Python and TypeScript clients: what each method takes, what it returns, and the ordering it handles for you.CLI and configEvery command and flag, and every field of the deploy contract.ReleasesWhat changed in each published package, and when.
When the docs do not answer it
Start with Troubleshooting, which covers the failures people actually meet. Otherwise email support@openelan.ai. Include the invocation id if a run is involved — it is on the run's page in the console and in every event the stream emits.