AFK

Overview

AFK is 何嘉琛’s enterprise-Agent company, presented in the source as building a continuously learning agentic system for organization-wide work.

Current Profile

AFK’s claimed product form is one highly concurrent enterprise agent that can operate across channels and tools while maintaining separate user memories and runtimes. Its proposed advantage is not merely task execution but shared learning: reuse prior context, identify effective employee practices, distill recurring trajectories, and connect AI cost to business outcomes. These are founder-reported architecture and strategy claims rather than independently measured product results.

Key Characteristics

• Enterprise-wide shared Agent rather than isolated personal assistants.
• Individualized memory and runtime separation within a common organizational system.
• Cross-environment operation through chat, collaboration tools, phone, browser, CLI, MCP, and operating systems.
• Continuous learning from work traces, tool calls, results, and preferences.
• Small-model strategy for routing repeated tasks and serving enterprise-specific work.

Evidence

• Product architecture: 复盘 2C 和 2B Agent 发展脉络｜对谈 Afk.surf 创始人嘉琛 describes AFK as always online, proactive, highly concurrent, and accessible beyond one interface.
• Learning strategy: 复盘 2C 和 2B Agent 发展脉络｜对谈 Afk.surf 创始人嘉琛 says company-wide experience can be shared while retaining person-specific memory and execution context.
• Model strategy: 复盘 2C 和 2B Agent 发展脉络｜对谈 Afk.surf 创始人嘉琛 describes a task-routing model and a roughly 20B–40B internal work model.

Qualifications

The source provides no audited customer metrics, security design, benchmark results, model-training details, cost curves, or deployment case studies. Product scope, model sizes, and benefits are first-party descriptions.

What Changed

• Created the entity from the founder interview.

Relationships

• 何嘉琛 - founder and source of the company’s product thesis.
• AFFiNE - predecessor experience that informed the move away from fixed workspace interfaces.
• Organizational Agent Learning / 组织级智能体学习 - proposed organization-level learning advantage.
• Enterprise Agent Memory - company and person-specific memory layer the product claims to combine.
• Model-Distilled Task Memory / 模型蒸馏式任务记忆 - proposed route for converting successful work into cheaper recurrence.