deytenit
b3638f4d63
🤖 PR generated for branch `fix/cyrillic-fonts-dev-certs-and-page-head`.
### Changes Summary
- **Fonts**: Add Cyrillic glyph subsets for Lato, EB Garamond, and Fira
Code.
- **Dev**: Support remote Root CA and encrypted dev CA key in
`.env.dev`.
- **UI & Kits**: Add size variant to `Breadcrumbs` component and enlarge
`PageHead` breadcrumbs with tuned collapse scroll threshold.
- **ESLint**: Add `no-hanging-particles` i18n & markdown typography rule
with auto-fixer.
- **Content & i18n**: Apply `ru-text` editorial and typography polish
(em dashes, number ranges, «ё» restoration, non-breaking spaces) across
`services/www` and `services/docs`.
1.5 KiB
1.5 KiB
title, weight
| title | weight |
|---|---|
| Conversational AI Integration | 4 |
Goal: Integrate Large Language Models (LLMs) while managing context and Discord's timeout constraints.
1. Context Management Strategy
Conversations require history to be coherent.
- Scoping:
Map<GuildID, Map<UserID, Message[]>>- Guild Key: Prevents cross-server leaks.
- User Key: Prevents cross-user confusion in public channels.
- Lifecycle: Implement a rolling window (e.g., last 50 messages) to prevent context bloat.
2. Response Streaming
LLMs are slow; Discord is fast (3s timeout).
{{% steps %}}
Step 1: Defer Interaction
Immediately call interaction.deferReply(). This buys you 15 minutes.
Step 2: Invoke LLM
Call your AI service asynchronously.
Step 3: Edit Response
When data arrives, use interaction.editReply(content).
{{% /steps %}}
3. Implementation Sketch
// commands/chat.ts
export async function execute(interaction, controller) {
// 1. Defer
await interaction.deferReply();
// 2. Process
const history = controller.getHistory(interaction.guildId, interaction.user.id);
const response = await aiService.generate(interaction.options.getString('prompt'), history);
// 3. Respond
await interaction.editReply(response);
controller.updateHistory(interaction.guildId, interaction.user.id, response);
}
Warning
API Costs: AI APIs are often billed by token. Implement strict user-based rate limits in your Controller to prevent abuse.