Models & Types

Every model on Pawan.Krd declares one or more types — these decide which API endpoint(s) accept it and which parameters are valid. A single model may expose more than one type (for example, an image model that also supports editing and upscaling).

Supported model types

chat/v1/chat/completions

Conversational LLMs. Multi-turn messages, vision input, tool calling, streaming, reasoning.

text/v1/completions

Base / instruct text completion models for prompt-style generation.

proofreading/v1/proofreadings

Grammarly-style spelling, grammar, punctuation, tone, pure-language and rewrite corrections.

embedding/v1/embeddings

Convert text into vectors for semantic search, clustering and RAG.

reranker/v1/rerank

Cross-encoder rerankers that score documents for a query.

moderation/v1/moderations

Safety classifiers that flag unsafe content.

speech-to-text/v1/audio/transcriptions, /v1/audio/translations

ASR — turn audio into text, optionally translated to English.

text-to-speech/v1/audio/speech

TTS — synthesize speech from text with selectable voice.

image-generation/v1/images/generations

Text-to-image generation.

image-editing/v1/images/edits

Mask-based and instruction-based image editing.

image-upscaling/v1/images/upscale

Super-resolution / upscaling of input images.

video-generation/v1/videos/generations

Text-to-video and image-to-video generation.

video-editing/v1/videos/edits

Video inpainting, restyling, and editing.

Capability flags

In addition to a type, each model exposes capability flags in its config that tell you what it can do. Common flags:

vision

boolean

Model accepts image inputs in chat messages.

thinking

boolean

Model exposes a hidden chain-of-thought / reasoning trace.

toolCalling

boolean

Model supports OpenAI-style function/tool calling.

internetAccess

boolean

Model can browse the web during a request.

ragSupport

boolean

Model honours your knowledge base / RAG preferences.

blockedWords

boolean

Model honours your blocked-words policy via logit bias.

translation

boolean

STT model supports the /v1/audio/translations endpoint.

wordTimestamps

boolean

STT model supports word-level timestamp_granularities.

matryoshka

boolean

Embedding model supports truncation to smaller dimensions.

inpainting

boolean

Image model supports mask-based edits.

imageToVideo

boolean

Video model accepts a still image as conditioning.

Parameters and ranges

Each model also publishes a parameters object describing the allowed range or enum for every tunable knob (temperature, top_p, max_tokens, dimensions, voice, aspect ratio, etc.) along with a recommended default. SDKs and the playground use these ranges automatically; you can read them from the model object returned by List Models.

Listing models

You can list every model your account can access:

💻curl

curl https://api.pawan.krd/v1/models \
  -H "Authorization: Bearer pk-your_key"

✓

Browse visually

The Models page in the dashboard groups models by type, shows capability badges, and includes a live playground for every model.

Quick Start

Base URLs & Routing