Runtime for PI

PI support is coming soon. When it lands, each agent will run in its own cloud environment, with a scoped credential for every provider PI supports, RPC and print/JSON modes mapped onto a stable endpoint, somewhere durable for extensions to live, and state for long sessions. Today, OpenClaw and Hermes already run on this model.

Coming Soon

What PI needs from a runtime

PI intentionally ships without MCP, sub-agents, permission popups, or background execution. The runtime around it has to carry that weight.

Credentials for every provider

PI reaches Anthropic, OpenAI, Google, Azure, Bedrock, Mistral, Groq, Cerebras, xAI, Hugging Face, Kimi, DeepSeek, GLM, OpenAI OSS routes, MiniMax, OpenRouter, and Ollama. The work is to project a scoped credential per provider so the harness can switch models without env-file juggling.

Print and RPC modes on the API

PI's print/JSON and RPC modes pair naturally with a stable HTTPS endpoint. The contract has to map cleanly onto `/exec` so application code can pipe output downstream.

Somewhere for extensions to live

PI extensions are TypeScript modules that assume a filesystem, installed packages, and persistent state. All three need a clear contract so extensions don't rebuild from scratch.

Long sessions that hold their context

PI's tree-structured session history and auto-summarization need state handling that matches those expectations across long conversations.

The Spinup Angle

The harness stays lean. The runtime carries the rest.

PI's design philosophy is explicit: provide extension primitives, not opinionated features. That choice pushes responsibility for state, secrets, and lifecycle out of the harness and onto the environment around it.

You can't run PI on Spinup yet. OpenClaw and Hermes are the supported harnesses today, on the same isolated microVM, projected secrets, and persistent state model. PI's credentials, modes, extensions, and state land on top once each has a real shape.

PI fits the runtime model well. The work ahead is the harness-specific glue, not the architecture underneath it.

FAQ

Common questions about running PI on Spinup

Why does a minimal harness like PI benefit from a cloud runtime?+

PI is intentionally minimal: MCP, sub-agents, permissions, and background execution all live in extensions. Those extensions need a stable environment around them. That's exactly what Spinup provides, which is why PI is on the roadmap. Today, OpenClaw and Hermes are the supported harnesses.

Does Spinup manage API keys for every provider PI supports?+

Not yet. When PI lands, credentials for Anthropic, OpenAI, Google, Azure, Bedrock, Mistral, Groq, Cerebras, xAI, Hugging Face, Kimi, DeepSeek, GLM, OpenAI OSS routes, MiniMax, OpenRouter, and Ollama get projected under the names PI expects. Today, OpenClaw and Hermes are the supported harnesses.

Can I call PI's RPC or print/JSON modes through a Spinup agent endpoint?+

Not yet. PI's print/JSON and RPC modes map naturally to a Spinup endpoint, but the contract gets defined when PI lands. Today, OpenClaw and Hermes are the supported harnesses through `/exec`.

What happens if I compare PI against another harness later?+

Today the supported swap is between OpenClaw and Hermes. PI joins the same portable runtime model when it lands: same agent, same environment, same secrets, only the harness layer changes.

Related

The runtime stays bigger than any one harness

Early access

PI is on the way. Try OpenClaw or Hermes on the same runtime today.

Request access if this is the runtime shape your team has been missing.

We reply with a booking link and review the fit before inviting your workspace.