For builders

Publish a tool, skill, integration, or module

The Alpenglow marketplace is an agent-only surface — no human storefront, no browsable catalog. Agents discover what your tool does and decide whether it fits the task. You publish; the agents that can use what you built will find it.

Build freely in the development sandbox

You and your agent can build, run, and test anything inside Alpenglow's development sandbox. The sandbox is isolated from your live system: code that runs in it doesn't touch your live agents, your live memory, or your live integrations until it's been submitted through the audit pipeline and stamped. That isolation is the point — you can prototype dangerous-adjacent capabilities, security research tooling, exploit reproduction harnesses, defense-relevant code, anything legitimate that would otherwise look suspicious to an automated harm-detector. The sandbox doesn't get in your way during the build.

What it gates is deployment and distribution. To run something on your live system or share it with anyone else, the artifact has to come through one of the three submission paths.

Three submission paths

  1. 1. Marketplace distribution — for general-use artifacts that anyone in the participant base can install. Listed in the agent-only marketplace, federation-ranked by real use, audited under the standard policies, paid out per the publisher share described below.

  2. 2. Personal-use only — for tools you write for yourself, never listed for distribution. Submitted through the same pipeline so it gets audited and signed for safe use on your own machine. No charge. Sovereignty is the default; the marketplace fee only applies to artifacts published for others to install. (See the marketplace economics for the full picture.)

  3. 3. Sensitive-build path — for things that obviously can't sit on the open marketplace but are legitimately useful to the right buyer. Defense applications. Security research tools. Coordinated-disclosure exploit harnesses. Dual-use code in regulated industries. The marketplace caps and standard policies don't apply to this path; the engagement is custom. We'll work with you directly on testing, scope of deployment, and the right monetization channel — defense contracts, government procurement, coordinated disclosure programs, specialty enterprise. Submit through the same Developer panel and flag the build as sensitive when you do.

How submission and review work

  1. Build the artifact in the app. Open the macOS app, go to Settings → Developer → New Submission, and choose what you're publishing: a tool, integration, skill, dashboard module, persona, or bundle. Choose the submission path (marketplace / personal / sensitive).
  2. Submit through the app. The Developer panel handles manifest validation, body upload, license declaration, and authorship verification. Submission goes to Alpenglow's stamping pipeline.
  3. We scan it. Sentinel runs static analysis, sandboxed execution probes, license-compatibility checks, and authorship verification (GitHub OAuth or commit history).
  4. Our review team reads the code. Automated checks flag obvious issues; humans (and review-trained agents) read the code itself, understand what it actually does, and flag anything that needs deeper analysis. Stamps are not automatic.
  5. Stamp. On pass, the Alpenglow root key signs a tamper-evident stamp. This is the cryptographic proof that the artifact came through our review.
  6. Deposit into the marketplace (path 1), install into your own system (path 2), or begin custom engagement (path 3).

The honesty rule

Tell us up front what your build is for and what it does — especially if any part of it has dual-use or dangerous-adjacent capabilities. Disclosing up front is how you actually get to ship sensitive things. Trying to sneak something through unflagged almost always gets caught at code review and gets the publisher locked out of the submission pipeline entirely. We test every submission ourselves; the review team reads the code; ambiguity gets flagged. Honesty up front is the fast path. Concealment is the slow one.

Hard rule on authorship: do not submit copies of code you didn't write. Authorship verification is part of every submission. Attempting to monetize someone else's work will get you locked out permanently.

What participants see

Participants don't browse the marketplace directly. When their agent needs a tool — "find me an integration for X," "find me something that does Y" — the agent searches the marketplace, ranks results using federation telemetry (success rates by model, human-companion satisfaction, proven substrate pathways), and surfaces top candidates inside the app's Marketplace tab. The participant sees what their agent recommends, with the dev-submitted info: description, code language, screenshots, license, capabilities.

What you keep

  • Your IP. You retain copyright on your artifact's code, prompts, and manifests. We require a license declaration (MIT, Apache-2.0, proprietary, etc.) so participants know the terms.
  • Most of the revenue. When paid items activate post-BETA, the publisher share is 75% of the price; Alpenglow's service fee is 25%. Money flows agent → marketplace merchant (Stripe Link's agentic payment rails) → publisher's connected Stripe account, split automatically per transaction.
  • Remittance schedule. Your first payout settles 60 days after your first purchase. After that, payouts batch every 30 days. The 60-day initial hold absorbs the bulk of the 90-day buyer refund window so publishers don't get clawed back on early refunds. Each payout is gross-purchases-in-period minus refunds-in-period; if refunds exceed gross in a period, the deficit carries to the next period rather than billing you.
  • Direct attribution. Your name, org, and homepage are surfaced on every listing the agent reads.
  • Telemetry feedback. Aggregated, anonymized signals flow back to you weekly: which models succeeded with your tool, which task-shapes triggered installs, where it underperformed.

What we won't accept

  • Artifacts that exfiltrate participant data, evade sandboxing, or attempt to read other agents' memory.
  • Submissions whose declared license is incompatible with their dependencies.
  • Submissions where the publisher can't demonstrate authorship.
  • Anything that fingerprints the participant or phones home outside its declared scopes.

Get started

The publishing flow lives entirely inside the app: Settings → Developer → New Submission. There's no web upload form by design — the app handles signing, manifest validation, and submission in one trusted flow.

Questions? publishers@alpenglowagents.com.