YAML, a prompt, any model. rush build compiles it into an encrypted, signed container — your prompt never ships in plaintext, and nothing can tamper with it in the wild.
What you get
YAML + prompts + tools. No chains, no graphs, no nested abstractions. Define what your agent does, not how a framework runs it.
Use Opus for reasoning, Haiku for data extraction, GPT for code — in the same agent. Switch models without rewriting anything.
Supported modelsrush inspect opens a live timeline. Stream tool calls, messages, and delegation chains in your browser. Not another log viewer.
rush publish encrypts and signs your agent with Ed25519. System prompts are never exposed. Your IP stays yours.
The developer workflow
Scaffold an agent in seconds. YAML + prompts + tools. No SDK to learn, no abstractions to debug.
$ rush init my-agent
name: researcher
models:
anthropic/claude-sonnet-4:
provider: openrouter
tools:
- delegate
- web_search
- artifactsRun the agent locally against a real prompt. Stream every tool call, delegation, and token as it happens.
$ rush run ./my-agent \
--prompt "..." --stream-events
tool web_search → 5 results
delegate → summarizer
artifact → report.mdbuild encrypts the prompt and signs the container with Ed25519. publish ships it to the registry. Users install with one click.
$ rush build ./my-agent
Prompt encrypted, signed Ed25519
$ rush publish ./my-agent
Published prix/[email protected]The compiler
Not a zip of your source. rush build encrypts the prompt, signs the container with Ed25519, and hash-locks every file. Installs verify the signature before a single line runs.
agent.yaml
Models, tools, context
prompt.yaml
Identity, workflow
assets/
Logo, cover
$ rush build
encrypt, sign, lock
Why Prix
Every agent in the registry is reviewed before publish. Ed25519 signatures verify integrity on install.
rush publish encrypts your agent container. System prompts and orchestration logic are never exposed.
iOS-style permissions. Agents declare what they need. Users approve or deny at install time.
ClawHub has 800+ malicious skills. Prix has zero. Read more
Why agents
Reading
Stripe shipped 288 launches at Sessions 2026. The rails for agent payments are done. The next ten years gets won at the layer above — the manifest that decides which scopes get minted, which inputs trigger them, and which human signed off.
API access is opening soon. Join the waitlist. See pricing.
Already building? Read the docs · Publish an agent · Changelog