Skrun

Skrun’s core idea is deliberately simple: turn any agent skill into a callable API with one endpoint, POST /run, rather than requiring integration with a specific vendor’s managed agent platform. It’s positioned explicitly as an open-source alternative to Anthropic’s Claude Managed Agents (CMA) and Google’s Gemini Enterprise Agent Platform (GEAP), aimed at teams that want the managed-agent-API experience without being locked into one model provider’s hosted infrastructure.

The runtime is multi-model by design — it works with any LLM rather than being wired specifically to Claude or Gemini — and self-hostable on any infrastructure, distributed as an npm CLI (@skrun-dev/cli) and SDK (@skrun-dev/sdk) alongside the runtime itself. Streaming and tool-calling are supported as core runtime features rather than bolted on per model provider.

MIT licensed, Skrun’s monorepo separates api, cli, runtime, schema, sdk, and web concerns, reflecting a genuine platform architecture behind the single-endpoint developer experience.

What You Get

• A single POST /run API endpoint for deploying any agent skill, instead of integrating with a vendor-specific managed agent platform
• Multi-model support that works with any LLM rather than being tied to one provider
• Self-hostable deployment on your own infrastructure via an npm CLI and SDK
• Built-in streaming and tool-calling support as core runtime capabilities

Common Use Cases

• Deploying an agent skill as a production API endpoint without adopting a specific vendor’s managed agent platform
• Running the same agent runtime across multiple LLM providers instead of rewriting integration code per model
• Self-hosting agent infrastructure for compliance or cost reasons instead of using a hosted managed-agent service
• Building internal tools that call agents via a simple HTTP API rather than a vendor SDK

Under The Hood

Architecture Skrun’s monorepo separates runtime (the core agent execution engine), api (the HTTP layer exposing POST /run), cli and sdk (developer-facing packages published to npm), schema (shared type/validation definitions), and web (likely a dashboard or docs site) — a structure that keeps the model-agnostic runtime decoupled from both the HTTP API surface and the client tooling built on top of it.

Tech Stack TypeScript and Node.js throughout, distributed as @skrun-dev/cli and @skrun-dev/sdk npm packages, with streaming and tool-calling implemented as first-class runtime features rather than provider-specific add-ons.

Code Quality The project runs CI on every change (per its GitHub Actions badge) and ships versioned npm packages for both CLI and SDK, indicating a genuine release process rather than a single-script tool; as a young project, broader maturity signals (large test suites, long track record) are still developing.

What Makes It Unique Rather than building yet another agent framework, Skrun specifically targets the deployment layer — positioning itself against vendor-managed agent platforms (Claude Managed Agents, Gemini Enterprise Agent Platform) by offering the same “call an agent as an API” experience without requiring you to run on that vendor’s infrastructure or model.

Self-Hosting

Licensing Model MIT licensed — the runtime, API, CLI, and SDK are all open source with no license key.

Self-Hosting Restrictions None found; self-hosting on your own infrastructure is the primary intended use case, explicitly positioned against vendor-managed alternatives.

License Key Required No.