Build reefpods
A reefpod is a reusable, versioned AI-agent package — a prompt directive,
model, inputs, budget, and a definition of “done,” bundled in a
pod.toml manifest. konareef gives you the tools to author,
validate, publish, and cryptographically verify one.
Everything in the Quickstart — scaffold, edit, and validate a real
pod — runs locally against the vendored schema, with no account and no
server. Publishing, execution and custody proofs need a
reef-core deployment: reef-core is
open source and self-hostable
(github.com/digitsu/reef-core,
docker compose up), and a hosted instance is available to
the invite-only private beta.
What is a reefpod?
A prompt tells an AI what to say. A script tells a computer what to do.
A reefpod declares an AI workflow: which runtime and model
it uses, what inputs it takes, what it may spend, what counts as success,
and what proof it should leave behind. It is a directory anchored by a
pod.toml manifest plus its prompt files.
The 15-minute builder path
Author and validate your first reefpod in about fifteen minutes:
- Install konareef — build the CLI from source (Go).
- Scaffold a pod —
konareef pod init hello-world. - Validate it —
konareef pod validate ./hello-world/pod.toml. Your first success moment. - Shape it — edit the directive prompt and add an
[inputs]block, then re-validate. - Publish & verify — create a publisher identity, publish, then install and offline-verify it as a commissioner.
Explore the docs
Start the Quickstart
Scaffold and validate a Hello World reefpod in 15 minutes.
Install konareef
Build the CLI and confirm your environment works.
pod.toml reference
Every table in the v0.1 pod manifest, from runtime to marketplace.
What is a reefpod?
The mental model: runtime, directive, budget, and custody proof.
Security & trust
Secrets, tool allowlists, sat budgets, and verifiable proofs.
Browse examples
Copy working reefpods and adapt them to your own workflows.