What is Endpoint AI?

Endpoint AI is Harriet's provisioning product: you configure skills, connectors, and model access once, and every enrolled device in your company receives that configuration automatically.

Configure once, provision everywhere

Without central provisioning, desktop AI spreads one laptop at a time. Each person pastes their own prompts, configures their own MCP servers, manages their own API keys, and nobody can say what tools the AI on a given machine can reach. The setup drifts, and the company has no record of what the AI did.

Endpoint AI replaces that per-user setup with one configuration step. An admin bundles skills and settings into a profile and assigns it to a team. Every enrolled device belonging to that team then receives the same skills, the same connectors, and the same model access. Devices check in every five minutes and apply whatever changed, so updating a profile updates the whole team.

Change flows in the other direction are governed too: skills built by employees go through the review queue before they ship to anyone else's device, and people can ask for capabilities they don't have with a skill request.

The desktop surfaces

The default surface is Harriet Desktop, the workspace app your whole company signs into. Two Claude Desktop modes exist for teams that want them, and all three are managed the same way:

• Harriet Desktop is the company workspace: chat, skills, and every leading model in one app, signed in with a Harriet account, syncing its configuration while it runs. See Harriet Desktop.
• Claude Desktop with a Claude account is for teams that prefer Anthropic's app on their existing Claude subscriptions; Harriet provisions your company's skills and MCP servers into it. See Claude Desktop, your subscription.
• Claude Desktop (no Claude account) is Claude Desktop without individual Anthropic subscriptions; Harriet supplies the models through its gateway, including non-Anthropic and EU-resident options. See Claude Desktop, no subscription.

One smaller piece keeps Claude Desktop in sync: the background agent, a terminal install for macOS and Windows that also covers MDM deployments. The desktop overview compares the surfaces.

The gateway and the audit trail

Tool calls from provisioned connectors pass through Harriet's gateway rather than going straight from the laptop to the integration. That is what makes the audit log possible: each call records who made it, from which device, with which skill, and whether it succeeded. When you route model traffic through the LLM proxy, usage and cost are recorded per device and per person as well, and budgets can cap managed spend.

When to use it

Endpoint AI earns its place when:

• You are rolling AI out to many machines, or to specific teams that need the same configuration.
• You need an audit trail and governance over which tools and skills appear on company computers.
• You want one owner for endpoint policy, with a review workflow for new skills and MCP connectors.

A small pilot where a handful of people share one machine's worth of experiments may not need it yet. The moment a second team wants the same setup, provisioning beats repeating the setup by hand.