Skip to content
Docs

Integrations

Lifeboat is designed to require no application rewrites. It presents the same API surface applications already use, so integration is normally a configuration change rather than a code change.

How do applications connect to Lifeboat?

Section titled “How do applications connect to Lifeboat?”

By pointing an existing SDK at Lifeboat’s base URL with a Lifeboat-issued API key. Because the inference surface is OpenAI-compatible, anything that speaks that format works unchanged:

  • Official SDKs — the OpenAI Python and Node SDKs, and Anthropic-compatible clients for supported endpoints.
  • Agent and RAG frameworks — LangChain, LlamaIndex, and any framework with a configurable OpenAI-compatible base URL.
  • Chat and developer front-ends — any tool that accepts a custom OpenAI-compatible endpoint.
  • Direct HTTP — plain REST calls, including streaming, with no SDK at all.

How does Lifeboat fit with the rest of the Iterate.ai suite?

Section titled “How does Lifeboat fit with the rest of the Iterate.ai suite?”
Product How it connects
Interplay Point Interplay’s LLM and agent nodes at Lifeboat so agentic workflows run against private models on your own GPUs.
Generate Generate consumes private inference endpoints alongside commercial providers, so end-user assistants can be served entirely in-house.
AgentWatch Put AgentWatch in front of Lifeboat to add data-loss prevention, guardrails, per-team budgets, and a unified audit trail across private and commercial models — the governance layer Lifeboat deliberately does not duplicate.

What monitoring and operations tools integrate?

Section titled “What monitoring and operations tools integrate?”
  • Prometheus — Lifeboat exposes a standard metrics endpoint that an existing Prometheus server or compatible agent scrapes directly; from there Grafana and any Prometheus-compatible alerting work normally.
  • Log platforms — container logs and the mountable log directory are collected by whatever agent the organization already runs (Splunk, ELK, Loki, Datadog, or similar).
  • Reverse proxies and ingress — NGINX, Envoy, HAProxy, and Kubernetes ingress controllers front the control plane for TLS and access control.
  • Container registries — images can be mirrored into any OCI-compliant private registry (Harbor, Artifactory, ECR, ACR, GCR, Quay) for air-gapped installs.

Where do models come from?

Section titled “Where do models come from?”

Either the Hugging Face Hub — including gated and private repositories via the encrypted token vault — or a local models directory mounted into the deployment, which is the standard approach for air-gapped sites and for organizations that mirror weights internally. See Model management.

Can Lifeboat front inference engines it did not launch?

Section titled “Can Lifeboat front inference engines it did not launch?”

Yes. External inference endpoints can be registered and managed through Lifeboat, so an engine already running in the environment can be consolidated behind the same control plane, API surface, and load balancer instead of being migrated first.

Sources and references

Section titled “Sources and references”