How it works
From sanctioned workspace to governed production.
Calliope AI gives each part of private AI a clear home: Calliope AI Workbench for building, Astrolift AI for deployment and runtime, and Zentinelle AI for agent observability and policy. Start with one product or operate the full lifecycle as one stack.
Start with the boundary
Put the operating model where the work has to happen.
The hard part of enterprise AI is not opening a model chat window. It is giving teams useful tools while keeping identity, infrastructure, data access, runtime behavior, and evidence under deliberate control.
Calliope AI starts with your boundary. Workbench and workloads run in the environment you choose. Your IAM, network rules, storage, and logging remain part of the architecture instead of becoming integrations after the fact.
Model traffic is explicit. In BYOC, Calliope Labs Inc does not receive or proxy prompts, code, model responses, or provider credentials. Requests go directly to the endpoints you approve. Use local or in-boundary models when zero egress is required.
Start with one problem. Add the rest when it earns its place.
The topology follows your security and operating requirements.
A team can begin with a Workbench deployment, an Astrolift AI runtime, or Zentinelle AI policy enforcement. The products share a lifecycle model without requiring an all-or-nothing purchase.
See the three-product architecture →Choose the starting constraint
Give builders a sanctioned workspace, standardize deployment, or gain control over AI traffic. Start with the problem creating risk now.
Compare the products → 02Place the workload
Select BYOC, managed, on-premises, or isolated deployment based on data classification, operations, and access requirements.
Review topologies → 03Approve model endpoints
Bring provider keys for approved external models, or keep inference local for zero-egress workloads.
Review the data flow → 04Deploy through one runtime
Astrolift AI takes workloads from CI to a repeatable runtime contract with health, logs, and operational settings.
Explore Astrolift AI → 05Observe and enforce
Zentinelle AI captures agent events, evaluates policy, and allows or blocks actions at the enforcement point.
Explore Zentinelle AI → 06Preserve the evidence
Route access records, runtime telemetry, and policy decisions into the systems your operators and reviewers already use.
Open the Trust Center →Clear responsibilities
Three products, without three disconnected stories.
Calliope AI separates the jobs that buyers may need independently, then makes their boundaries deliberate when they run together.
- Calliope AI Workbench gives technical teams a governed place to use IDEs, notebooks, chat, data tools, and agents. It covers Build and development-time Run.
- Astrolift AI packages, deploys, and operates AI workloads. It covers Build in CI, Run, runtime Observe and Control, and supply-chain Secure.
- Zentinelle AI observes AI and agent traffic, evaluates policy, enforces decisions, and records evidence. It covers Observe, Control, and Secure.
This division matters commercially too. Workbench is licensed per seat. Astrolift AI and Zentinelle AI have MIT-licensed cores, with support, implementation, and engineering services available separately.
A better answer than the three bad defaults
Instead of consumer SaaS
Place the Workbench and workloads inside the environment your security team can approve, with model destinations chosen explicitly.
Review the trust model →Instead of building the stack
Use a reference architecture and supported products instead of assigning a platform team to invent every integration and maintain it indefinitely.
Open the blueprint →Instead of tool sprawl
Give development, deployment, telemetry, policy, and evidence one lifecycle model without forcing every capability into one license.
See the platform →Instead of a one-way door
Keep model choice, deployment portability, and MIT-licensed operating cores between your organization and permanent lock-in.
See the open-core model →Where we are today
A realistic path to production
Fast provisioning is the beginning, not the whole deployment.
Automated infrastructure can provision the baseline in under an hour. A production rollout also includes identity, networking, model access, logging, security review, and the workflows your team intends to operate. Assisted onboarding is normally measured in days.
AWS is the supported cloud path today. The architecture is portable by design, with additional cloud backends and runbooks in progress. On-premises and air-gapped deployments are scoped to the environment rather than advertised as a universal one-click install.
The result is a sanctioned operating path. Builders get useful tools, platform teams get a repeatable runtime, and security teams get explicit data flows and evidence.
