Meet Archie: Your AI Personal Assistant That Actually Respects Your Privacy

Every HiveClaw customer eventually asks the same question: "Can the Swarm help me manage my personal workload too?" They are already watching AI agents build their product. They see the daily status updates, the phase transitions, the budget summaries. And they think: if agents can manage a software project this well, why can't one manage my calendar, my inbox, and my to-do list?

Today, we are introducing HivePA and its resident agent, Archie — an AI personal assistant built on the same infrastructure that powers HiveClaw. Archie is a standalone product in the HiveClaw family, alongside HiveVault. You do not need a HiveClaw project to use it. But if you have one, Archie becomes significantly more useful.

Why another AI assistant?

The honest answer: because every existing one gets privacy wrong. Most AI assistants treat your personal data the same way they treat any other data — it goes into a model, gets processed, and lives somewhere in a cloud you do not control. When the assistant manages both your work and your personal life, the boundaries between those domains blur. Your calendar events, your health notes, your contact list — it all becomes training data or, at minimum, data that sits on servers managed by people who are not you.

We did not want to build another assistant that asks you to "just trust us." We wanted to build one where the architecture makes privacy violations structurally impossible, not just contractually prohibited. That led us to the Two-Vault architecture and the five-tier privacy system.

Two modes, one agent

Archie operates in two modes, and understanding the distinction is important because it drives everything about how the product works.

Mode A: Project assistant

If you have an active HiveClaw project on Pro or Dedicated tier, Archie is automatically available as your project coordinator — included at no extra cost. In Mode A, Archie handles the operational surface of your project relationship:

• On-demand status updates so you do not have to wait for the next phase report.
• Approval reminders that escalate if you have not responded. The Swarm does not stall silently — Archie nudges you.
• Budget burn rate monitoring with natural-language summaries. "You have spent 42% of your Implementation budget with 60% of features complete. You are tracking slightly under budget."
• Document retrieval from the project workspace. Instead of digging through Notion or the dashboard, ask Archie to find the latest architecture decision record.
• Multi-project unified view if you are running more than one HiveClaw project simultaneously.

Mode A uses only project data — status, budgets, deliverables, approvals. It never touches your personal information. It communicates through the same chat interface you use with Alfred, and the message router automatically detects whether a message is project-related or PA-related.

Mode B: Personal life assistant

Mode B is the standalone product. It is available to anyone — you do not need a HiveClaw project. This is the full personal assistant: calendar management, email triage, reminders, a personal CRM, life admin tracking, travel planning, research, and morning briefs that combine your personal schedule with project updates (if you have any).

Mode B is accessed through a dedicated interface and requires its own subscription. The three tiers — Essentials at $19/month, Complete at $39/month, and Private at $49/month plus a dedicated VPS — gate which features are available. Essentials covers calendar, reminders, and knowledge base. Complete adds email, CRM, life admin, and travel. Private adds health and finance tracking with the maximum privacy tier.

The five-tier privacy system

Privacy in Archie is not a toggle. It is a classification system applied to every piece of data, with different protections at each level.

• T0 — Public. Your name, timezone, and communication preferences. Not encrypted, visible to HiveClaw agents if you have a project. This is the minimum information needed for the system to function.
• T1 — Project. Project status, budgets, deliverables, approval states. Not encrypted, visible to HiveClaw agents. This is standard project data that the Swarm needs to do its job.
• T2 — Professional. Business contacts, meeting notes, research files. Encrypted at rest. Not visible to HiveClaw agents. Logging is redacted — the system records that a T2 access occurred, but not the content.
• T3 — Personal. Calendar events, reminders, to-dos, travel plans, personal CRM entries. Encrypted at rest. Not visible to HiveClaw agents. Logging is redacted.
• T4 — Sensitive. Health data, financial records, personal relationships. Encrypted at rest with customer-keyed encryption. Not visible to HiveClaw agents. Never logged. Requires the Private plan and a dedicated VPS — this data never touches shared infrastructure.

The critical design decision: data only moves from a lower tier to a higher tier's security level, never the other way. Project data can inform your personal calendar ("you have an approval pending, want me to block time for it?"), but your personal calendar never flows back to the project agents. The bridge between Mode A and Mode B is one-way, and every crossing is logged in a dedicated audit table.

Customer-keyed encryption

For T4 data on the Private plan, we use customer-keyed encryption. Your vault key is generated and given to you. We store the fingerprint — a hash that lets us verify you hold the right key — but never the key itself. If you lose it, we cannot recover your data. This is by design.

This means that even if our servers were compromised, your T4 data would be useless to an attacker without your key. It also means we genuinely cannot read your health records or financial data, even under legal compulsion. We do not have the capability. The architecture makes it impossible, not just against policy.

The dedicated VPS option

The Private plan runs on its own Hetzner VPS. Not a container on shared infrastructure — an actual dedicated server. It accepts outbound connections only. HiveClaw cannot SSH into it. The PA process runs in isolation, and the only data that leaves the VPS is what you explicitly allow through the bridge.

If you cancel the Private plan, you can transfer the VPS to your own Hetzner account. Your data stays on hardware you control. We do not hold it hostage and we do not keep copies. This is how we would want our own personal data handled, so it is how we handle yours.

Model selection per task

Archie does not use a single model for everything. Different tasks have different intelligence requirements, and using a frontier model to set a reminder is wasteful. Our model selection strategy matches the task:

• Reminders, to-do management, and status summaries use a fast, lightweight model. These are structured, predictable tasks.
• Calendar management, email triage, and meeting prep use a mid-tier model. These require understanding context and making judgment calls.
• Complex research requests use a frontier model. When you ask Archie to compare three competing products or analyze a market segment, it needs the full reasoning capability.
• Crisis detection and escalation drafting use a mid-tier model with specific safety constraints.

This tiering keeps costs predictable and response times fast. The majority of PA interactions resolve in under two seconds because they hit the lightweight model path.

The bridge: where personal meets project

The most interesting design challenge was the bridge between Mode A and Mode B. When a user has both an active HiveClaw project and a HivePA subscription, Archie can connect the contexts in useful ways:

"You have an approval pending for Project X. You also have a dentist appointment at 2pm. Want me to review the approval before your appointment?"

This kind of contextual awareness requires data from both domains — project state and personal calendar. The bridge handles this carefully: project data flows into the personal context (so Archie knows about the pending approval), but personal data never flows back to the project agents (the CTO agent does not learn about your dentist appointment).

Every bridge crossing is logged in the bridgeAuditLog table. You can review exactly what project data was shared with your personal context, when, and for what purpose. If you want to disable the bridge entirely, one toggle turns it off. Mode A and Mode B then operate in complete isolation.

Why standalone

We considered making Archie exclusively a HiveClaw add-on. It would have been simpler architecturally and easier to market. But the personal assistant use case does not require a HiveClaw project. Someone who wants AI-managed calendar, email, and to-dos should not have to start a software project to get it.

HivePA as a standalone product also lets us build a relationship with users before they need HiveClaw. Someone who trusts Archie with their calendar and email is more likely to trust HiveClaw with their product build. And when they do start a HiveClaw project, the Mode A integration makes the transition seamless — Archie already knows their communication preferences, their schedule, and their working patterns.

Getting started

HivePA is available now. If you are an existing HiveClaw customer on Pro or Dedicated, Mode A is already active in your dashboard — just start messaging Archie through the chat interface. New users can get started by submitting a project through our intake flow.

We built Archie because we think the AI personal assistant category has a trust problem. Everyone wants one. Nobody trusts the existing options with their actual personal data. The five-tier privacy system, customer-keyed encryption, and dedicated VPS option are our answer to that trust gap. Your data is yours. The architecture guarantees it.