Agents run on the official pi SDK (agent-host/), not the old built-in loop. One durable runtime. Text events stream over SSE with cursor reconnect.Prompt templates live on the Go backend. Dynamic context and the JSON contract are injected server-side and cannot be overwritten by a user template. Chapter extraction pulls characters, scenes, and props into pending assets.

Media is handled locally first. H.265 originals can be stored on the canvas without a browser decode step. Playback prefers the source; if it fails, an H.264 preview is generated on demand, single-threaded, resolution-capped, and reused. Windows builds ship the video tools — no separate FFmpeg install. Hover preview is muted, first 3 seconds, one global decoder. Download and model reference always keep the original file.

We're so glad to launch our new product: BeefTV.

Six more models on the global price list. Gemini 3.7 Flash, Gemini 3.8 Flash, MiniMax M3, Qwen3.8 Max, Qwen3.8 27B, Kimi K2.7 Code. 27 → 33. Same key, USD credit, each request still itemized in Usage. global.beefapi.com/pricing

Thanks for all Founding Testers first. You didn’t wait for a finished product. you tested it, broke it, and helped us ship the next version. That’s why this isn’t a promo code. Your account keeps a permanent 20% recharge discount. For every top-up after this. Seats are limited. When they close, they close. global.beefapi.com/console/t…

Pix is live at checkout for Brazilian user!🇧🇷 If you open billing from a Brazilian IP and pick the $10 credit, Pix shows up as a payment option. Same prepaid USD credit, same Usage receipt after the request. We added it because people asked. If the option doesn’t appear, say where you’re checking out from.

GPT-6 Astra vs Fable 5.1 Which one performing the best?👇

GPT-6 #Astra is live now on BeefAPI! Same key. Same OpenAI-compatible client. About 70% less than running it through the usual checkout — public USD price before you send, a receipt after. People who’ve been running it keep landing on the same feeling: Less like switching models, more like tuning one that actually listens. Use it now: global.beefapi.com/pricing

Usage rows now open. Click a request and you get the bill for that call: input, cache read, cache write, output, per-call or tool use. Each line is usage × price = amount. The last line is Total charged — same number as Cost in the list. If long-context pricing applied, the receipt says so in one sentence. If we couldn’t observe usage, it says the charge was estimated — we don’t invent a unit price. That’s the product. One request, one receipt you can read.

If you already speak OpenAI, you should not have to learn a new SDK. 📌OpenAI-compatible tools: global.beefapi.com/v1 📌Claude Code and other Anthropic-style tools: global.beefapi.com Create a key, keep it on your server, send a call. Codex CLI is documented the same way. The setup is supposed to feel slightly boring. That is the point.

One key. Prepaid USD credit. Claude, GPT, Grok, DeepSeek, GLM on the same account. You see the price before a request leaves, and the receipt when it comes back.