القائمة
Use when adding LLM features to a yaebal bot with @yaebal/ai — ctx.ai streaming replies, model adapters, conversation memory, and rate limits.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Use when building multi-step dialogs in a yaebal bot — choosing between scenes, conversation, and prompt, and wiring their persistence.
Use when writing or modifying a Telegram bot built on the yaebal framework — bot setup, handlers, filter queries, commands, and context typing.
Use when writing a custom yaebal plugin — the Plugin<In, Out> contract, derive vs decorate, and typed plugin dependencies.
Use when a yaebal bot fails to compile, start, or respond — install-order type errors, ESM specifier errors, Telegram 400/409/429 responses.
Use when running a yaebal bot in production — long polling with @yaebal/runner vs webhooks with @yaebal/web, graceful shutdown, and error handling.
Use when adding inline/reply keyboards or handling button presses in a yaebal bot — @yaebal/keyboard builders and @yaebal/callback-data typed payloads.
| name | yaebal-ai-features |
| description | Use when adding LLM features to a yaebal bot with @yaebal/ai — ctx.ai streaming replies, model adapters, conversation memory, and rate limits. |
bot.install(ai({ model })) adds ctx.ai to every context: model calls flow through optional
per-user limits, conversation memory and a system prompt, and stream into the chat using
telegram's native draft animation where available.
import { createBot } from "yaebal";
import { ai, openaiCompatible, AiLimitError } from "@yaebal/ai";
const bot = createBot(process.env.BOT_TOKEN!).install(
ai({
model: openaiCompatible({ model: "gpt-4o-mini", apiKey: process.env.OPENAI_API_KEY }),
system: "you are a concise, friendly assistant.",
limits: { perUser: "20/h" },
parseMode: "MarkdownV2", // finalized messages only — see behavior below
}),
);
bot.on("message:text", async (ctx) => {
try {
await ctx.ai.replyStream(ctx.text);
} catch (error) {
if (error instanceof AiLimitError)
return ctx.reply(`slow down — try again in ${Math.ceil(error.retryAfterMs / 1000)}s`);
throw error;
}
});
bot.command("reset", async (ctx) => {
await ctx.ai.reset();
await ctx.send("forgotten.");
});
ctx.ai.replyStream(prompt, options?) — stream into the chat: in private chats via
sendMessageDraft (telegram's own "Thinking…" placeholder + animated draft, finalized as a
real message); in groups/channels via throttled editMessageText with a ▍ cursor. answers
past 4096 chars split at word boundaries and finalize message-by-message. returns
{ text, messages, mode: "draft" | "edit", ticks, aborted }.ctx.ai.reply(prompt, options?) — generate fully, then send (typing action kept alive
meanwhile); returns { text, usage?, messages }.ctx.ai.generate(prompt, options?) — run the model, return { text, usage }, send nothing.
does not write memory by default (opt in with { memory: true }).ctx.ai.stream(prompt, options?) — raw AsyncIterable<string> for custom rendering; reads
memory but never writes it.ctx.ai.history() / ctx.ai.reset() — read / forget this conversation's memory.ctx.ai.model — the resolved adapter (.id for logs).prompt is a string or an AiMessage[]; per-call options can override system, memory,
temperature, maxTokens, and pass an AbortSignal as signal.
openaiCompatible({ model: "llama3.2", baseUrl: "http://localhost:11434/v1" }); // any /chat/completions api
anthropicModel({ model: "claude-sonnet-5", apiKey: process.env.ANTHROPIC_API_KEY! });
ai({ model: anthropic("claude-sonnet-5") }); // a vercel ai sdk LanguageModel, auto-detected
// (or wrap explicitly: aiSdk(anthropic("...")))
customModel(async function* ({ messages }) { // anything that yields strings
yield "hello ";
yield "world";
});
openaiCompatible and anthropicModel are zero-dependency fetch adapters; the ai sdk bridge
lazy-loads the ai package only when used.
**bold can never 400 mid-stream. if telegram rejects the finished entities,
the text is resent plain instead of being lost. don't try to apply markup to previews.@yaebal/sklad StorageAdapter via memory: { storage } to persist across restarts, or
memory: false for stateless calls.limits.perUser budgets throw AiLimitError (with retryAfterMs) — catch it and
answer politely; don't let it hit the global error handler.streaming: intervalMs (500 draft / 1200 edit), drafts: false
to opt out of drafts in private chats, cursor ("" disables), maxLength.typing: false disables the typing action for non-streamed replies.