Kestrel

High-performance inference engine for the Moondream vision-language model.

Kestrel is the inference engine behind Photon, Moondream's on-device deployment option. Most users should install via pip install moondream — this repo is the internal engine for those who need direct access.

Kestrel provides async, micro-batched inference with streaming support, paged KV caching, and optimized CUDA and Metal kernels. It's designed for production deployments where throughput and latency matter.

Features

• Async micro-batching — Cooperative scheduler batches heterogeneous requests without compromising per-request latency
• Streaming — Real-time token streaming for query and caption tasks
• Multi-task — Visual Q&A, captioning, point detection, object detection, and segmentation
• Paged KV cache — Efficient memory management for high concurrency
• Prefix caching — Radix tree-based caching for repeated prompts and images
• LoRA adapters — Parameter-efficient fine-tuning support with automatic cloud loading

Requirements

• Python 3.10–3.14.
• One of:
  • NVIDIA GPU on Linux x86_64 / aarch64 or Windows x86_64. Optimized kernels for SM80 (A100), SM86 (A10, RTX 30-series), SM87 (Jetson Orin), SM89 (L4, L40S, RTX 4090), SM90 (H100, H200, GH200), SM100 (B200), SM110 (Jetson Thor), SM120 (RTX PRO 6000). Other CUDA GPUs may work but have not been tested.
  • Apple Silicon Mac (M-series) on macOS 13 (Ventura) or later, with native Metal kernels.
• MOONDREAM_API_KEY environment variable or api_key parameter (get a key from moondream.ai)

Installation

pip install kestrel

For Jetson Orin (JetPack 6) or Jetson Thor (JetPack 7), see the Jetson setup guide.

Model Access

Kestrel supports both Moondream 3 and Moondream 2:

Model	Repository	Notes
Moondream 2	vikhyatk/moondream2	Public, no approval needed
Moondream 3	moondream/moondream3-preview	Requires access approval

For Moondream 3, request access (automatically granted) then authenticate with huggingface-cli login or set HF_TOKEN.

Quick Start

import asyncio

from kestrel.config import RuntimeConfig
from kestrel.engine import InferenceEngine


async def main():
    # Weights are automatically downloaded from HuggingFace on first run.
    # Use model="moondream2" or model="moondream3-preview".
    cfg = RuntimeConfig(model="moondream2")

    # Create the engine (loads model and warms up)
    engine = await InferenceEngine.create(cfg, api_key="your-key-here")

    # Load an image (JPEG, PNG, or WebP bytes)
    image = open("photo.jpg", "rb").read()

    # Visual question answering
    result = await engine.query(
        image=image,
        question="What's in this image?",
        settings={"temperature": 0.2, "max_tokens": 512},
    )
    print(result.output["answer"])

    # Clean up
    await engine.shutdown()


asyncio.run(main())

Tasks

Kestrel supports several vision-language tasks through dedicated methods on the engine.

Query (Visual Q&A)

Ask questions about an image:

result = await engine.query(
    image=image,
    question="How many people are in this photo?",
    settings={
        "temperature": 0.2,  # Lower = more deterministic
        "top_p": 0.9,
        "max_tokens": 512,
    },
)
print(result.output["answer"])

Caption

Generate image descriptions:

result = await engine.caption(
    image,
    length="normal",  # "short", "normal", or "long"
    settings={"temperature": 0.2, "max_tokens": 512},
)
print(result.output["caption"])

Point

Locate objects as normalized (x, y) coordinates:

result = await engine.point(image, "person")
print(result.output["points"])
# [{"x": 0.5, "y": 0.3}, {"x": 0.8, "y": 0.4}]

Coordinates are normalized to [0, 1] where (0, 0) is top-left.

Detect

Detect objects as bounding boxes:

result = await engine.detect(
    image,
    "car",
    settings={"max_objects": 10},
)
print(result.output["objects"])
# [{"x_min": 0.1, "y_min": 0.2, "x_max": 0.5, "y_max": 0.6}, ...]

Bounding box coordinates are normalized to [0, 1].

Segment

Generate a segmentation mask (Moondream 3 only):

result = await engine.segment(image, "dog")
seg = result.output["segments"][0]
print(seg["svg_path"])  # SVG path data for the mask
print(seg["bbox"])      # {"x_min": ..., "y_min": ..., "x_max": ..., "y_max": ...}

Note: Segmentation requires Moondream 3 and separate model weights. Contact moondream.ai for access.

Streaming

For longer responses, you can stream tokens as they're generated:

image = open("photo.jpg", "rb").read()

stream = await engine.query(
    image=image,
    question="Describe this scene in detail.",
    stream=True,
    settings={"max_tokens": 1024},
)

# Print tokens as they arrive
async for chunk in stream:
    print(chunk.text, end="", flush=True)

# Get the final result with metrics
result = await stream.result()
print(f"\n\nGenerated {result.metrics.output_tokens} tokens")

Streaming is supported for query and caption methods.

Response Format

All methods return an EngineResult with these fields:

result.output          # Dict with task-specific output ("answer", "caption", "points", etc.)
result.finish_reason   # "stop" (natural end) or "length" (hit max_tokens)
result.metrics         # Timing and token counts

The metrics object contains:

result.metrics.input_tokens     # Number of input tokens (including image)
result.metrics.output_tokens    # Number of generated tokens
result.metrics.prefill_time_ms  # Time to process input
result.metrics.decode_time_ms   # Time to generate output
result.metrics.ttft_ms          # Time to first token

Using Finetunes

If you've created a finetuned model through the Moondream API, you can use it by passing the adapter ID:

result = await engine.query(
    image=image,
    question="What's in this image?",
    settings={"adapter": "01J5Z3NDEKTSV4RRFFQ69G5FAV@1000"},
)

The adapter ID format is {finetune_id}@{step} where:

• finetune_id is the ID of your finetune job
• step is the training step/checkpoint to use

Adapters are automatically downloaded and cached on first use.

Configuration

RuntimeConfig

RuntimeConfig(
    model="moondream3-preview",  # or "moondream2"
    max_batch_size=4,            # Max concurrent requests
)

Environment Variables

Variable	Description
MOONDREAM_API_KEY	Required. Get this from moondream.ai.
HF_HOME	Override HuggingFace cache directory for downloaded weights (default: ~/.cache/huggingface).
HF_TOKEN	HuggingFace token for gated models like Moondream 3. Alternatively, run huggingface-cli login.

Triton Inference Server

Kestrel can be deployed as a Triton Inference Server backend. See the Triton setup guide.

Benchmarks

Throughput and latency for the query skill are tracked in PERFORMANCE.md, with results broken out by GPU.

License

Kestrel requires a Moondream API key for billing. See moondream.ai/pricing for plans.