Enter to send · Shift+Enter new line · ⚡ templates · / skills · screenshots read with OCR
Enter to send · Shift+Enter new line · ⚡ templates · / skills · screenshots read with OCR
Leave blank to call /api/* on this host (your panel proxy forwards to AiProxy). Or set a full host like https://aiproxy.th3va.com to call it directly.
Sent as x-api-key. Leave blank to use the key your proxy serves from its .env.
Your model's context size — drives the usage meter under the composer.
How hard GLM 5.2 reasons before answering. Higher settings think more thoroughly — best for coding and complex problems.
Extra language data downloads once from the CDN on first use, then caches.
Standing facts about your infrastructure and conventions. Sent with every message in every mode, so answers fit your world by default.
Each chat runs in a work mode (Build · Architect · Research · Review) with its own expert prompt. Anything here extends or overrides that prompt in all modes.
Layered under the mode prompt on every turn and stored with this conversation. Your global stack profile lives in Connection settings.
Injected into the system prompt of every conversation and refreshed in the background after each reply. Edits here stick — the updater merges around them.
Reusable coding workflows. Type / in the composer to invoke one — its full instructions are sent with that message only, so they never waste context when idle. The model always sees a one-line index of what exists.
Files pinned to this chat. Their current versions are sent to the model on every turn, and when it replies with a file=path code block you can apply the update back here in one tap.