Twenty-six capabilities. One platform.

ReACT-loop agents that read, edit, and execute code with vision-grounded clicks.

Hybrid BM25 + vector retrieval over your documents with LlamaIndex.

Lesson pearls, thumbs-up distillation, and an autonomous code engine.

Orchestrate multiple agents in parallel with shared memory and role specialization.

Click, type, and screenshot trainers that teach the agent your apps. Lesson pearls bake the wins.

Diffusers + LoRA. Bulk CSV/XML pipelines. Style packs. No upload required.

Wan2.2 and CogVideoX pipelines on your GPU. Batch queues. ComfyUI integration.

Whisper.cpp STT + Piper TTS, streaming, with full conversation memory.

Generate, edit, stitch, and overlay audio. Narration, transcripts, podcast cuts.

Local 4× upscaling via ESRGAN/Real-ESRGAN. Batch queues + watch folders.

Trim, stitch, transition. ffmpeg-backed scrubbing for cleaning generated clips.

Captions, lower-thirds, title cards. Auto-transcribe + style + render.

Bulk content pipelines for CSV, XML, and structured outputs at scale.

First-class conversation surface. Model switching, system prompts, token counting.

Upload, parse, index, search. Auto-indexed into LlamaIndex; usable everywhere.

Workspaces with scoped chats, tasks, files, memory. Clean context switching.

Browsable archive of every generated output. Filterable, tagable, re-usable.

Persistent canvas. Searchable, taggable, RAG-indexed. Your scratchpad.

Running jobs, GPU status, model load, recent activity. The home you land on.

Monaco editor wired to the same agent that runs your tasks.

Pull, generate at scale, push back. Multi-site bulk pipelines.

Drafted comments and value-add replies on Reddit, Discord, and forums.

Detect, schedule, and balance multiple GPUs and models.

Sync state and outputs across multiple Guaardvark instances.

Drop-in folder with manifest. Loaded at startup. Extends without forking.

System prompts and policies, versioned and assignable to projects, agents, features.