your-app.korve.app URL and
optional databases, storage, queues, crons, caches, observability, recovery,
custom domains, and spend controls — plus a deployment agent that investigates
failures and proposes approval-gated actions.
The shortest path from repository to live application is korve init: it
detects the Git source, writes a reviewable manifest, shows the plan, applies
it, links the directory, deploys, and verifies the terminal deployment state.
It is built agents-first. Public operator capabilities are declared once and
projected onto four surfaces with the same permissions and behavior:
- API —
https://api.korve.dev, described by a public spec atGET /v1/spec. See the API reference. - CLI — the
korvebinary. Every public API operation a key can call maps to exactly one command (a tested parity guarantee). See the CLI reference. - MCP —
https://mcp.korve.dev/mcp, two tools:searchthe API andexecutecode against it. See MCP setup. - Dashboard — the convenience layer at korve.dev for the things humans prefer to click.
How Korve is built
- Paid-only, usage-based. No free tier, no seats. You pay for what you run, metered against a public rate card to the cent, with a $5/month platform minimum credited against usage.
- One spec, one enforcement point. Auth, roles, and billing gates are declared per operation and enforced once in the API layer. The dashboard, CLI, and MCP inherit identical behavior — there is no surface with secret extra powers.
- Declarative when you want it. Describe projects, environment, and domains in a manifest and apply it. The platform reconciles and never deletes anything you didn’t ask it to.
- Assisted operations. Failed deploys can be diagnosed by the deployment agent, which persists the conversation and requires approval before any mutation.
Runtime classes
Every project picks one of two runtime classes at creation:
See Deploys for what each class expects from your
code.
Next steps
- Quickstart — account to live app in a few minutes.
- Install the CLI.
- Connect an agent over MCP.