Pollo Assistance
The main AI chat: smart, multimodal, and brand-friendly.
| Type | Main module |
|---|---|
| Availability | FREE (with credits) / Premium (more credits) |
| Tiers | Palos · Luces · Summum |
| AI pool | 20+ models across Meta, OpenAI, Mistral, Google, Alibaba, Moonshot, DeepSeek |
What it does
Pollo Assistance is the central AI chat: ask anything, get responses with text, images, voice or web context. Different from a single-model chatbot, Pollo Assistance has a brain selector: for each message, an orchestrator AI picks the best model from the pool of the active tier (Palos / Luces / Summum) based on the type of task.
Key capabilities
Multimodal chat
Text, images and code in the same message. The brain routes to a vision-capable AI if you attach images.
Voice in / voice out
Speech-to-text input with real-time transcription. Listen to responses with native browser TTS.
Image generation
The brain decides when to generate an image with Pollinations and embeds it inline in the response.
Web search
If the question requires recent info, the brain fires a web search and feeds the snippets to the specialist.
Public chat sharing
Share any conversation with a public link (read-only) Google Docs-style. Recipient does not need an account.
Auto-compression
Long chats compress old turns into a summary so the AI keeps the thread without breaking context.
How the brain works
Every time you send a message, the brain (a separate fast AI from the specialist) decides:
- Which AI will respond (a code task → llama-70b, Mistral or Qwen; an analysis → GPT-OSS-120B or DeepSeek R1; a casual chat → Llama 8B; etc.).
- Whether to search the web for current information.
- Whether to generate an image.
- A brief briefing for the specialist with tone, length and "must do/avoid".
If the main brain fails (out of credits, downtime), there is a fallback chain: secondary brain → tertiary → built-in heuristic. The chat never breaks.
Auto-disabling broken models
If an AI fails twice in 10 minutes (or once with a 404 / out-of-credits error), Pollo Assistance automatically marks it as broken for the rest of the session. The brain stops suggesting it. After 24 hours the model returns to the pool (in case the issue is temporary).
Tiers
See the full breakdown on AI Models. Brief summary:
