Deployment

Deployment

Every cove1 deployment is delivered by our team: we provision the environment, connect your channels, and tune the agent with you until it fits. This page explains the deployment options, what the moving parts are, and how a rollout works.

Two ways to run

The product is the same either way; the difference is where the infrastructure sits and what constraints it satisfies.

• Managed cloud. Jincove runs the stack for you. Fastest to start — no environment to provision, no services to operate, updates handled for you. This is the default for most teams.
• Dedicated deployment. For enterprises with residency, compliance, or integration constraints, we deliver a dedicated environment — including in-region hosting, or an environment you designate — as part of a custom rollout. More to agree up front, but the deployment fits your security and network controls instead of asking you to bridge to an outside service.

Many teams start on managed cloud to evaluate the product, then move to a dedicated deployment once support becomes business-critical. Because it is the same underlying stack, that transition does not mean rebuilding your agents.

The architecture

A cove1 deployment is a small set of cooperating services. At a high level:

• Conversation engine. Handles channels, inboxes, message history, and the human agent workspace. This is where conversations physically live and where the console reads from.
• Agent runtime and knowledge base. Runs the canvas logic and retrieval: it takes an incoming message, retrieves relevant knowledge, reasons over it, and decides whether to answer or hand off.
• Console. The single workspace your team uses to watch conversations, jump in, and manage agents and channels. It sits on top of the engine and the runtime and presents them as one product.
• Model provider. Any OpenAI-compatible endpoint, with DeepSeek as the default. This can be a hosted API or an in-region endpoint chosen for your deployment.

The flow of a message: a customer sends a message on a channel, the conversation engine receives it, the agent runtime retrieves knowledge and calls the model provider to reason, and the result is either an answer sent back through the channel or a handoff surfaced to a human in the console.

Where your data lives

The design goal is simple: your conversation data is processed to run your agents, and the boundaries are explicit.

• Conversations, message history, and your imported knowledge live inside your deployment's environment — in a dedicated deployment, that can be a specific region or an environment you designate.
• The one boundary to decide on is the model provider. When a reply is generated, the prompt is sent to the configured endpoint. For strict residency requirements, we configure an in-region provider so that traffic stays put too.
• Your data is never sold, shared, or used to train foundation models.

How a rollout works

1. Scope. We map your channels, knowledge sources, tone, and escalation rules together.
2. Provision. We stand up your environment — managed cloud, or the dedicated setup we agreed on — and configure the model provider.
3. Connect and import. Your channels come online and your knowledge is imported and indexed.
4. Tune. We shape the agent's flows on the canvas with you, and you review real conversations in the console until the behavior fits.
5. Go live. Publish, watch the first conversations, and adjust. From here you are in the Quickstart loop: refine knowledge, tighten handoff, connect the next channel.

Questions about a deployment? Reach out at gan@jincove.com or book a demo.