FOR AI SAAS TEAMS & AGENCIES
Run your AI product with your own infrastructure and action controls. [5]
Build on a managed platform with in-app agents and memory. [4]
Agents · Memory · Knowledge · Tools [5]
Self-hosted Runtime · Local models · Air-gap [1]
App integration and Runtime configuration required. Air-gap needs local models and dependencies. Commercial Runtime licensing applies. [11]
WHAT YOUR PRODUCT NEEDS
Read the scope and plan alongside each capability. These descriptions are not scores or a claim that every integration is automatic.
Build your product’s UI and domain logic around the Runtime APIs or SDK. Agree acceptance for the complete app, its authentication and its data services. [5]
Continue application development in your repository and editor; evaluate requirement changes against existing behavior. [7]
Two-way GitHub synchronization requires Builder or higher. Only the app owner can connect or reconnect the repository; workspace connection policy affects collaborator access. [8]
Use agreed application journeys, grounding checks and action-denial tests for acceptance. Runtime evidence and a test of the full application answer different questions. [9] [3]
Testing Agent runs browser scenarios, including on the Free plan. Review findings, request fixes and rerun the relevant tests. [10]
Runtime self-hosting and air-gap are documented. Configure local models, embeddings and tool dependencies; validate the actual workload. [1] [9]
Local development is documented, with in-memory entity data and some AI/OAuth/integration calls forwarded to the deployed app. Those docs do not establish a standalone production or air-gapped backend. [6]
GOVERNANCE HAS A SCOPE
Look at roles, model access, sensitive actions and the resulting audit evidence.
In-app agents have configurable tool/data access. Business and Enterprise plans document workspace publishing controls; model-access policies require Enterprise. These are meaningful capabilities to include in your evaluation. [4] [12]
The Runtime action ledger records approval and execution outcomes. Desk adds review queues and per-tool policies. Configure and test the application's path through these controls; external execution needs separate enforcement. [13] [14] [3]
TRY THE SAME CUSTOMER TASK
Use this evaluation brief with either platform. Agree the required behavior and the test environment before you build.
Sign in, retrieve permitted product knowledge, retain the right user context and draft a reply. Require human approval for the agreed sensitive action.
Add an escalation rule. Repeat the original journeys and check that access boundaries and saved data still behave correctly.
Build from the delivered source, run in the agreed environment and verify backup and restore. Record model services, licenses and operating effort.
A working preview is one checkpoint. Acceptance also requires the critical journeys, permission checks and required data persistence to pass. Compare cost per accepted delivery, with human effort and ongoing usage recorded separately.
A local development server, an exported app and an independently operated production backend are different deliverables. Ask each provider to demonstrate the one your contract requires, with data restoration and required services included.
Published by LibraOS. Official documentation reviewed . Selection guidance, not a performance or cost benchmark. Plans and versions can change.
™