EXXETA/exxperts

What it solves

Exxperts provides a local-first AI agent system where memory is governed and user-approved. It prevents the "black box" memory of typical AI assistants by ensuring that nothing is remembered without explicit user consent, and every saved fact is traceable back to the original conversation.

How it works

The system is organized into "rooms," which act as durable workspaces. Each room replaces a standard chat thread and can be configured with specific tools, folders, and skills.

Memory growth happens in three stages:

1. Work: The agent performs tasks, searches the web, and creates documents.
2. Remember: At the end of a session, the agent proposes specific facts to keep. The user must edit or approve these proposals before they are saved.
3. Memorize: Over time, the system consolidates these notes into long-term memory, again subject to user review.

All data—including conversations, credentials, and memory—is stored as plain files on the user's local machine, allowing users to switch between different AI providers (e.g., Claude, ChatGPT, local models) without losing the agent's learned knowledge.

Who it’s for

Users who want AI agents that learn and remember their specific work context but require strict control over privacy, data sovereignty, and the auditability of the AI's memory.

Highlights

• Approval-gated memory: No information is automatically saved; every memory proposal must be reviewed and approved by the user.
• Provenance receipts: Every remembered fact links directly back to the conversation from which it was learned.
• Local-first storage: All data stays on the user's machine as plain files, eliminating cloud lock-in and telemetry.
• Provider agnostic: Compatible with multiple AI providers and OpenAI-compatible endpoints, including local models.
• Built-in tools: Includes web search (via DuckDuckGo), MCP connectors, and the ability to produce artifacts like documents and decks.