onPanda
Steer LLMs and agents at the token level
Last verified:
What is onPanda?
onPanda (on-Policy Alignment Data Annotator) is a web-based interface for interacting with and annotating LLM outputs at the token level. It visualizes token probabilities, allows editing or rewriting response chunks, and supports creating alignment/annotation data for LLMs and agents.
onPanda pricing
Pricing model: Freemium
onPanda pros
- Token-level probability visualization with color-coded chunks (green = high, red = low)
- Interactive editing: candidate continuations on hover, double-click to edit, select & rewrite spans
- Multimodal input support including images, audio, and video
- Rich examples and built-in tooltips for onboarding
onPanda cons
- Image inputs only supported by specific models
- Requires following specific annotation workflows (e.g., deleting non-annotation history before saving)
- Limited model/endpoint options visible in the UI
Frequently asked questions about onPanda
What is onPanda used for?
It's a data annotator tool for LLMs and agents that provides token-level supervision and probability visualization for creating alignment data.
Can I input images or video?
Yes, you can paste images, audio, and video directly into the input box, though only specific models support image inputs.
How do I edit model responses?
Double-click a chunk to modify it, or select a span of text to rewrite it; you can also click candidate continuations shown on hover.