Product philosophy

Open-source, local-first personal AI. Chat with Ada, delegate tasks, and work with files and projects.

xopc is personal AI for the one-person company. It runs in your environment, remembers your goals and context, and helps move work forward across conversations, tools, and time. Instead of starting over every time, xopc picks up where you left off—so it becomes more useful the longer you work together.

Working independently should not mean working alone. xopc brings memory, context, and intelligence into everything you do while keeping direction, judgment, creativity, and final decisions with you.

It is not an AI employee waiting for a job description, a coding agent confined to one repository, or a task manager that expects you to organize everything first. Models, Agents, Tasks, Projects, Workflows, and Automations are supporting capabilities. The product is the long-term relationship they make possible.

Ada is the personal Agent included in xopc: a conversation for goals, preferences, delegation, and returned results. The workspace is the direct path for files and projects. Both support independent work in the same installation.

Let AI carry the execution. Keep the direction human.

xopc is not meant to live life on a person's behalf or turn them into an operator who merely supervises automation. It lets AI carry repeatable, time-consuming, verifiable execution while keeping goals, judgment, creativity, and final decisions with the person. People should not have to surrender agency to gain leverage, nor personally perform every step to stay in control.

This relationship follows four principles:

1. AI should take on execution such as organizing, researching, following up, using tools, and verifying results.
2. The person decides what is worth doing and retains consequential tradeoffs, creative judgment, and final decisions.
3. Automation stays bounded by explicit authority, inspectable processes, and correctable outcomes.
4. Time saved is part of the product's value. It should return to life, relationships, and what genuinely matters.

The 80/20 brand loop expresses the direction in which this relationship can develop: as understanding, capability, and trust grow, AI can carry more execution, so people spend less time on mechanical work and preserve more attention for judgment and life. It is not a fixed promise for every task, nor does it reduce the human role to 20 percent. Human intent and decisions continue to define the entire loop.

The relationship, not just the interface

xopc should not pretend to know someone on the day it is installed. Understanding is earned through an explicit progression:

Meet
→ explore authorized sources
→ form an initial understanding
→ help with real work
→ earn trust
→ become gradually proactive

The assistant begins with conversation and safe defaults. The user may then connect a work folder or another source, see what xopc examined, review what it inferred, and correct it. Each useful result and correction makes the next interaction better grounded.

The desired first-value moment is not “xopc can access my files.” It is:

xopc noticed what I am actually trying to accomplish, showed why it reached that conclusion, and helped me take a credible next step.

Four product capabilities

Understand a person

xopc maintains structured, governed understanding rather than treating every piece of data as permanent memory. Useful understanding may include:

• goals, preferences, and working habits;
• important people and relationships;
• active projects and longer-running directions;
• decisions and the reasons behind them;
• commitments that still need attention;
• recent pressures, blockers, and current focus.

Ordinary understanding is maintained automatically and can be inspected, corrected, or deleted. Uncertain and stale items are quietly withheld rather than queued for user approval. Direct statements and explicit rules take priority over inference. See User understanding.

Hold messy, long-running work

People rarely produce perfectly organized tasks. A useful assistant must be able to receive an incomplete thought, file, link, voice note, message, or piece of evidence and help turn it into something actionable:

Capture
→ connect to a goal or Project
→ add context
→ define the next action
→ identify dependencies and blockers
→ resurface it at a useful time
→ preserve decisions and evidence
→ retain reusable learning

Conversation is the natural entry point. A Task becomes the durable commitment when work needs to continue across time. A Project provides related context without creating a competing task state.

Solve problems with the user

xopc should understand why a result matters, clarify ambiguity, investigate authorized context, make a plan, use tools, verify the result, and leave an inspectable record. Complex work can span multiple conversations, devices, and execution methods without requiring the user to reconstruct the story each time.

Earn initiative

Initiative is not notification volume. It is the ability to notice a meaningful stalled commitment or new signal, propose a useful response, and stay quiet when interruption would not help.

Authority grows in stages:

Observe only
→ remind
→ propose an action
→ act after confirmation
→ automatically handle explicitly authorized, low-risk work

The proactive foundation is evidence-first and suggestion-first. External writes and consequential actions require explicit policy or approval. See Automations.

Trust is a product surface

Exploration must remain visible and controlled. Product behavior should follow these rules:

1. Authorize each source separately and allow access to be revoked.
2. Show what is being examined, within what scope, and for what purpose.
3. Preserve evidence lineage for conclusions and recommendations.
4. Distinguish observed facts from inference.
5. Allow confirmation, correction, rejection, deletion, and forgetting.
6. Keep durable state local by default and disclose when selected context will be sent to a cloud model.
7. Require separate authority for sending, deleting, purchasing, or other high-impact actions.
8. Prefer silence over low-value interruption.

Local-first describes ownership and architecture; it does not automatically mean that every model request stays on-device. Use a local model when context must not be sent to an external provider.

What exists now

The current product includes the foundations of this experience:

Product need	Current foundation
Local ownership	Desktop app, self-hosted Gateway, local state, bring-your-own keys, cloud and local models
Automatic understanding	Source-backed personalization, explicit collaboration rules, user edits and deletion, quiet stale/conflict handling
Visible work discovery	Bounded, read-only analysis of explicitly selected folders with evidence-backed next steps; optional native context on macOS
Durable work	Conversations, Tasks, Projects, Notes, workspaces, evidence, and run history
Execution and follow-up	Tools, Skills, Workflows, Automations, and proactive suggestion infrastructure
Access surfaces	Desktop, web, CLI, TUI, mobile, Telegram, Weixin, Feishu/Lark, and APIs

On macOS, the experimental Work Discovery onboarding can request separate operating-system access to Apple Notes, Calendar, and Reminders. It uses bounded, read-only scans and does not store raw native-source content in the xopc database. Availability and exact behavior may vary by release.

Start with the work you have

Ask your assistant to follow a topic, keep Task results and acceptance criteria, or schedule repeatable work. Authorize external sources through Connectors.

Describe recurring work in chat and use Tasks or Automations to manage it. The separate Scenes feature has been removed.

Product measure

Message volume and automatic task count are not the north-star measure. The more important question is:

After seven days, does the user feel that xopc understands them better than it did on day one—and that it genuinely helped advance something important?

Equally important: did xopc reduce the repeatable work the person had to carry without weakening their sense of control, and return meaningful time and attention to them?

One person. One AI. A new kind of company.

Start from the current experience

Meet Ada · Delegate and receive results · Work with files and projects · Task acceptance.