Agent Col: AI Collaborator with Controlled Continuity

Описание и применение

Agent Col – AI‑ассистент, сохраняющий контекст между сессиями, но с контролем пользователя над тем, что запоминается, решая проблему «контекстного рота».

Задача

AI assistants lose context and become unhelpful over long projects, leading to context rot and continuity issues.

Результат

Agent Col demonstrates that an AI collaborator can maintain useful context across sessions while keeping the user in control of what is remembered, solving context rot and continuity problems.

Рабочий процесс

• 1. User starts a conversation with Agent Col.
• 2. Agent Col captures immediate context and proposes relevant workspace notes or memories.
• 3. User approves or rejects proposals, governing what becomes durable memory.
• 4. Agent Col stores approved data in Firestore under appropriate scopes (user or workspace).
• 5. In subsequent sessions, Agent Col retrieves relevant memories and notes to inform responses.
• 6. Agent Col can ask clarifying questions, suggest next steps, and propose new notes, all subject to governance.

Доказательства

Agent Col uses a layered, multi-tiered memory architecture built around Google Cloud Firestore, with explicit governance controlling what information can become durable memory.

Источник

Long-term user memory is governed by contracts defining what Agent Col is allowed to retain, and proposed memories require explicit user approval before becoming authoritative.

Источник

Agent Col also maintains a temporary internal working context during conversation.

Источник