Open SDK for Building AI Agents: Isolated Execution, One Policy Engine, a Marketplace to Sell Into

Most "build an AI agent" starting points hand you a prompt template and leave the hard parts to you: how the agent's own build and deploy steps stay isolated from every other tenant's, and how you stop the agent from doing something it shouldn't once it's live. Neither is a wrapper problem — they're infrastructure problems, and skipping them is why so many agent side-projects never make it past a demo. Linkworld's Open SDK is built directly on the execution and governance layer the platform runs its own automation on, so a builder gets both by default instead of having to assemble them.

Every Build and Deploy Runs Isolated, on Whatever Coding Engine Fits

When you run linkworld init and linkworld deploy against the Open SDK, that build doesn't share a workspace with anyone else's. Linkworld's execution layer gives each goal — including your app's own build and deploy jobs — its own isolated git worktree, and runs it through a multi-engine coding pipeline that unifies different coding agents (Claude Code, Codex, Aider) under one adapter rather than hard-wiring you to a single vendor's CLI. That isolation is what makes it safe for many builders' apps to be scaffolded, tested, and redeployed concurrently on shared platform infrastructure without one run corrupting another's files or state — and because the underlying coding engine is an interchangeable adapter, your app's build pipeline isn't tied to one coding-agent vendor's roadmap or pricing either.

Your Agent Runs Under One Tenant Action-Policy Engine, Not Ad Hoc Checks You Write Yourself

Shipping an agent that can act on someone's data is a governance problem before it's a feature problem: what is this agent allowed to do on its own, and what needs a human to confirm first. That's not something an Open SDK builder has to design from scratch — every app built on Linkworld runs under the platform's tenant action-policy engine, the same layered, priority-ranked rule evaluator (with per-tool default registrations from skills and configurable autonomy relaxation) that gates every other agent action on the platform. Your app inherits real governance — audit trail and approval routing included — instead of a self-rolled permissions check that only covers the cases you thought of.

Ship It as Your Own App, or Sell It Through the Marketplace

An app you build with the Open SDK doesn't have to stay internal. Bundle your agent's manifest, workflows, tools, and channels into a deployable app, and if it's good, list it on the Linkworld marketplace — the platform handles billing for you across free, one-time, subscription, or usage-based pricing, whichever model fits what you built, and gives buyers a ready marketplace landing page to find it on. You're not negotiating a payments integration or building your own storefront on top of the SDK; that part is already built.

Who This Is For

Developers and technical builders extending an AI operator with agents it doesn't have yet — whether the target is their own company's operator or a marketplace listing sold to other companies. If you're evaluating agent-tooling SDKs and the isolation and governance underneath them matter as much as the scaffolding on top, this is the layer to look at.

Frequently Asked Questions

Is the Open SDK free to start with?

Yes — there's no access gate or approval form to use it. Install it (pip install linkworld), scaffold a project with linkworld init, and deploy with linkworld deploy.

Do I have to build my own sandboxing or permission checks for my agent?

No. Build and deploy jobs run isolated by the platform's execution layer, and once your agent is live, its actions are evaluated by the same tenant action-policy engine that governs every other agent on Linkworld — you don't design isolation or an approval system yourself.

Do I have to pick one coding CLI to build with?

No. The execution layer underneath the SDK treats coding engines like Claude Code, Codex, and Aider as interchangeable adapters, so your build pipeline isn't locked to one vendor's tool.

How do I get paid if I sell my app on the marketplace?

The Linkworld marketplace handles billing for you, and you choose the pricing model — free, one-time purchase, subscription, or usage-based — instead of being forced into a single fixed tier.