How to publish an AI agent: a readiness checklist
Updated · By Algen AI
The publishing checklist
A version is a promise. Everything you submit should describe this exact release.
- Declare the agent in a manifestDescribe identity, capabilities, license, ownership, data classes, outbound hosts, secrets and delivery using the algen-marketplace-manifest/v1 contract.
- Package exactly one versionBuild a package or container for a single version. Remove secrets, personal data and customer content.
- Add evidence that matches that versionInclude deterministic tests, an SBOM, a threat model, limitations and a support contact.
- Sign in and create the listingUse a passwordless email link, then create the listing with its display details and delivery path.
- Upload for inspectionThe package is quarantined while archive safety, secrets, dependencies and evidence are checked.
- Respond to review findingsFix issues and resubmit. Administrators then apply gated ASCEND controls and a separate risk overlay.
- Publish and maintainThe release is published as an immutable version. Any material change ships as a new version.
What makes ASCEND assessment faster
- A complete manifest, so M0 declarations are already explicit.
- Observability: correlated run identity and bounded, privacy-aware logs and traces (M1).
- Dependency evidence, a threat model, evaluations and documented least privilege (M2).
- Enforced budgets, scopes, approvals and egress controls at runtime (M3).
Why publish on Agent Hub
- Buyers see your evidence next to your listing, so trust is part of discovery.
- Framework- and cloud-neutral: Algen Runtime, LangGraph, CrewAI, OpenAI Agents SDK, containers and hosted endpoints.
- Verified ratings come only from users who acquired your agent.
- Shareable Published and ASCEND badges that link back to the exact version.
- Free to publish in the first phase.
Browse versioned agents with visible ASCEND levels, permissions and delivery paths.
How to publish an AI agent: a readiness checklist: common questions
What do I need to publish an AI agent on Agent Hub?
A package or container for one exact version, a manifest declaring ownership, license, runtime, data use, network access and delivery, plus limitations and a support contact. Deterministic tests, an SBOM, a threat model and evidence that matches the version make assessment faster.
Which agent frameworks are supported?
Agent Hub is framework- and cloud-neutral. Algen Runtime, LangGraph, CrewAI, OpenAI Agents SDK, custom containers and hosted endpoints use the same review boundary.
What happens after I upload my agent?
The upload is quarantined while archive safety, secrets, dependencies and evidence are checked. Administrators then apply gated ASCEND controls and a separate risk overlay before publication. Submitted code is never executed in the web or API process.
Do I need a new version when my agent changes?
Yes. Material code, manifest, permission, delivery or evidence changes require a new immutable version. Earlier acquisitions and reviews keep their version attribution.