Proposal: Persistent Custom Assistants (Agent Profiles) like Custom GPTs / Gemini Gems

[Muhammad313a]

I would like to propose adding persistent, user-customizable AI assistants to the Qwen ecosystem, similar in concept to Custom GPTs (ChatGPT) and Gems (Google Gemini).

Problem

Today, users must restate system constraints, role definitions, and behavioral expectations in every new session. This leads to:

• inconsistent behavior across conversations,
• repeated prompting overhead,
• weaker outcomes for long-term professional or educational workflows.

Proposal (what I’m asking for)

Support assistant profiles / agent profiles that can persist across sessions, including:

• persistent system-level instructions (behavior/policy),
• a stable role (e.g., strict tutor, code reviewer, formal analyst),
• optional attached documents / knowledge sources (RAG / knowledge base),
• consistent behavior across sessions.

Why this matters for Qwen

Qwen is widely used for serious technical and enterprise-style workflows. Persistent assistants would enable:

• reusable and repeatable workflows,
• structured learning and training setups,
• more predictable outputs and reduced hallucination via constrained behavior.

This is not model fine-tuning. It’s a persistence + configuration layer around the model (role/instructions/optional knowledge attachments).

Design questions (for discussion)

• Should profiles be scoped per user, per workspace/project, or both?
• How should document attachments be handled (built-in knowledge base vs external integration)?
• What policy/safety constraints would be needed for user-defined system instructions?

Thanks for considering this idea. I’d be happy to share concrete examples or use cases if useful.

[Muhammad313a]

Thanks! Clarification: this request is about the Qwen platform / agent experience (persistent custom assistants like Custom GPTs / Gemini Gems), not about changing Qwen3 model weights.

If Qwen3 repo isn’t the best place for product/agent feature requests, could you please point me to the correct repository or channel? I can repost it there (e.g., Qwen-Agent if that’s appropriate).

[jklj077]

Hi, thanks for the suggestion! In Qwen Chat, you can customize certain aspects of the conversation style on a per-user basis by visiting https://chat.qwen.ai/settings/personalization. For features not currently supported, ping @dxdong1234 to assess potential enhancements.