Meet the Specialist Agents: Free, Ready-to-Use Tools Powered by NVIDIA
Most BotTrottr agents are generalists: you give them a role, a model, and some connected tools, and they work out the rest. Sometimes you don’t want that. Sometimes you want to drop in a noisy recording and get back a clean one, or paste a paragraph and hear it in Japanese.
That’s what Specialist Agents are for. They’re a new section of the dashboard (and the demo) — ready-to-use agents, each built around a single NVIDIA model that’s very good at one job. They don’t need a swarm, a system prompt, or any setup. Open the page and use them.
What’s in the section
🔇 Noise Agent — NVIDIA Background Noise Removal
Hand it a WAV recording and it strips out the background noise — fans, traffic, keyboard clatter — and returns a cleaned clip. The card shows a before/after player so you can hear exactly what changed, and an intensity setting (0–100%) lets you decide how aggressive to be. Output comes back at 48 kHz.
Good for: meeting recordings, voice memos, podcast intros, and cleaning audio before you transcribe it.
🗣️ Multilingual Text-to-Speech Agent — NVIDIA Magpie-TTS
Type or paste text and get natural speech back in 12 languages: English, Spanish, French, German, Italian, Vietnamese, Mandarin Chinese, Hindi, Japanese, Arabic, Portuguese (Brazil), and Korean. Each language has several voices (and some have moods — calm, happy, neutral), and you can preview a voice before committing. Language is auto-detected from your text, or you can specify it.
Good for: narration, accessibility, IVR prompts, or reading an agent’s report out loud. (This is the same engine behind the new speaker icon on every agent card.)
🌍 Translation Agent — NVIDIA Riva-Translate
Translate between English and 36 languages. Send the text with to: Spanish (or just write “Translate to Spanish: …”). The source language is auto-detected, and non-English text with no target is translated into English. For pairs that don’t include English, the agent pivots through English automatically.
Good for: localizing support replies, understanding foreign-language messages, and feeding translated text into other agents.
📄 Document Parser — NVIDIA Nemotron-Parse 2.0
Also available as a model you can pick for any agent: give it an image or a PDF and it returns the document as structured markdown, with layout awareness — headings, tables, and where each element sits on the page. PDFs are rasterized page by page so multi-page documents work too.
Good for: invoices, forms, scanned reports, anything where the text is trapped in pixels.
Why they’re a separate section
Specialists are tools, not teammates. A generalist agent might spend a few turns reasoning about what to do; a specialist does one transformation reliably — audio in, clean audio out; text in, speech out. Keeping them in their own section means:
- They don’t clutter your swarms. Specialist cards sit below your swarm grid, clearly labeled and powered by NVIDIA.
- They’re included in the free plan. They run on BotTrottr’s NVIDIA access, so there’s no key to bring.
- They compose. Because every specialist is still a normal BotTrottr agent underneath, you can chain them: Document Parser → Translation → Text-to-Speech turns a scanned invoice into a spoken summary in another language. Noise Agent → transcription cleans a call before it’s transcribed.
How we built them
Under the hood, each specialist is a regular BotTrottr agent whose “model” is an NVIDIA service instead of a language model:
- A model slug (for example
nvidia-bnrornvidia-magpie-tts) tells BotTrottr to route the run to a purpose-built handler instead of a chat loop. - The handler talks to NVIDIA’s hosted APIs — gRPC streaming for noise removal, HTTP for speech and translation, vision for document parsing — and knows each service’s quirks: audio is chunked to fit gateway limits, rate limits are retried with backoff, and flaky first calls are retried automatically.
- Generated audio is stored and served through signed, expiring links, so a clip can be played in the dashboard or an email without exposing anything else.
- The dashboard recognizes these agents by model and renders a purpose-built studio panel for each — an uploader and before/after player for noise removal, a voice picker for speech, a language pair picker for translation — rather than a generic chat box.
The result: the same agent you can schedule, email, and call over the API, with an interface shaped to the job.
Try them
Open the BotTrottr demo, scroll to Specialist Agents (or click it in the left menu), and try one. More specialists are on the way as we add more NVIDIA models.