Installing a customer-hosted AI agent platform
Install a customer-hosted AI agent platform from the agreed release package and its matching operator instructions. Confirm infrastructure, model serving, identity, network routes and backups first. A successful container start is the beginning of acceptance, not proof of secure, recoverable or disconnected operation.
Evaluating this for your team? Discuss your use case, deployment boundary and success criteria in a technical briefing.
Book a technical briefingConfirm the deployment package
Customer-hosted deployments use packaged deployment configuration. An offline deployment requires an accepted bundle, local model and disconnected installation tests.
Record the release identifier, package source, authenticity procedure and matching installation instructions. Agree the connected or disconnected licensing path before installation.
Prepare the environment
- Infrastructure capacity based on host count and model serving.
- Local DNS, approved TLS and identity configuration.
- Explicit outbound routes for the connected design.
- Backup location, recovery owner and maintenance window.
Connect the approved model and hosts
Bring your own model — OpenAI-compatible endpoints, Ollama, Anthropic, or Cloudflare AI Gateway / Workers AI — with keys envelope-encrypted at rest.
A host joins the fleet with one install command and an enrollment token; revoking the host unpairs it.
Read-only execution policy is the default on every enrolled host; the full policy is an explicit, per-host choice.
Complete acceptance before expanding
Verify sign-in, host enrolment, one supported investigation, evidence and policy rejection. Then test backup restoration and the release procedure. Record what was actually tested and any customer-specific dependency.
Questions buyers ask
Which install command should we use?
Use the command supplied with the accepted release and deployment package. Avoid copying an older example into a different release.
Customer-hosted deployments use packaged deployment configuration. An offline deployment requires an accepted bundle, local model and disconnected installation tests.
Can we install without external services?
That needs a disconnected design and acceptance test covering images, model, identity, licensing and recovery.
What model does it use?
Identify the approved endpoint and whether its processing is local or hosted.
Bring your own model — OpenAI-compatible endpoints, Ollama, Anthropic, or Cloudflare AI Gateway / Workers AI — with keys envelope-encrypted at rest.
Where to go next
All guidesBring one question of your own.