FAQ

What is the setup wizard? Is it required? It is optional but recommended for first use. Missing required settings automatically enables /config/ and prints a session token. The eight steps save .env, import chats and generate the persona. See manual setup for the alternative.

Must I use Alibaba Cloud Qwen3-Embedding? No. Use any compatible /embeddings service, including a local one, and set EMBEDDING_API_URL, EMBEDDING_MODEL and EMBEDDING_DIM. Changing models after importing requires rebuilding the vector database; see Troubleshooting.

Must I use OpenAI? No. The main model supports compatible Chat Completions and Responses upstreams, selected by CHAT_API_PROTOCOL=chat_completions|responses. DeepSeek, Moonshot, Qwen and Ollama commonly provide Chat Completions. Verify your provider's actual endpoints.

Can I turn off PII redaction? Yes: ENABLE_PII_REDACTION=false. Alternatively, load custom rules with PII_RULES_PATH.

Why are QQ IDs and URLs not redacted? QQ IDs are needed to associate senders; shared URLs carry conversational context. Built-in redaction targets phone numbers, email, identity documents, bank cards and IP addresses. Check the rules against your own data rather than assuming every identifying detail is removed.

Can I import WeChat, Telegram or Discord? Yes. WeChat supports WeFlow arkme-json and ChatLab JSONL; Telegram supports Desktop HTML exports; Discord supports DiscordChatExporter JSON. WhatsApp TXT/Markdown and QQChatExporter are also supported. Other platforms can use an ingestion plugin or Afterglow Chat v1.

Will the companion drift as I chat? Turns are asynchronously written to live_messages with lower trust than human history. AI messages are excluded from cross-conversation semantic retrieval by default. You can pause writeback in frontend settings. Enabling long-term AI retrieval may increase drift.

How do I delete a memory? Use DELETE /memory/friend_messages/{chunk_id} or DELETE /memory/response_pairs/{pair_id}. These perform soft deletion.

Can everything run locally? Yes. Use local chat and embedding services, configure every enabled auxiliary model locally, and disable web access. See offline configuration.

Why is output in a different language from the UI? The page language and backend language are separate. Set XUWEN_LANGUAGE=en|zh-CN and restart for startup/proactive output; use CLI --language or HTTP Accept-Language for those scopes. Chat history and existing stored reports are not automatically translated.