# gizai

Tensors on the CPU and GPU, gradients, GPU kernels, and language, image and video models. Import a module by name, for example `import * as Store from "gizdb:store";`. Each page lists the module's public declarations exactly as an installed toolchain provides them.

| Module | Purpose |
| --- | --- |
| [gizai:a2a](/docs/reference/gizai/a2a) | the Agent2Agent protocol (0.3, JSON-RPC over HTTP) for agents. |
| [gizai:agent](/docs/reference/gizai/agent) | an agent is a model, its instructions, typed tools and the conversation so far. |
| [gizai:ai](/docs/reference/gizai/ai) | AI models on this computer, by name. |
| [gizai:cluster](/docs/reference/gizai/cluster) | clustering with a predict step: k-means (k-means++ and Lloyd), mini-batch k-means, BIRCH (CF subclusters, then Ward agglomeration by the nearest-neighbor chain), affinity propagation and mean shift. |
| [gizai:diffusion](/docs/reference/gizai/diffusion) | video and image generation models (diffusion transformers), fast on any hardware.  A model family is a recipe: its config.json read as an Arch, and the names of its tensors. |
| [gizai:grad](/docs/reference/gizai/grad) | derivatives of tensor programs. |
| [gizai:grammar](/docs/reference/gizai/grammar) | the texts a model may write, for constrained decoding. |
| [gizai:h3](/docs/reference/gizai/h3) | giz's worker of MiniMax H3 (gizai:diffusion's D.H3): its video.generate jobs as WanGP makes them (sliding windows, a video to extend, control videos, Ref2VA references, a soundtrack) and the enhance-video and interpolate-video jobs (RTX Video and optical flow on the same GPU). |
| [gizai:higgs](/docs/reference/gizai/higgs) | Boson AI's Higgs audio tokenizer v2, the codec of OmniVoice: 24 kHz mono audio as 8 codes of 1024 values every 960 samples (25 a second), and back.  Encoding joins two views of the audio at 25 Hz: DAC's acoustic encoder of the 24 kHz samples (256 values a frame), and the mean of HuBERT base's 13 hidden states of the audio at 16 kHz, every second frame, through a semantic encoder (768 values). |
| [gizai:llm](/docs/reference/gizai/llm) | language models, fast on any hardware.  A model family is a recipe: its config.json read as an Arch, and the names of its tensors. |
| [gizai:ltx](/docs/reference/gizai/ltx) | Lightricks' LTX-2.5 (22B), the audio and video flow transformer of giz.ai's ltx2-5-distilled.  Two streams of rows run side by side through 48 blocks: video (4096 wide, 32 heads of 128; a row a latent pixel of 8 frames and 32 x 32 pixels) and audio (2048 wide, 32 heads of 64; a row a 40 ms latent). |
| [gizai:mcp](/docs/reference/gizai/mcp) | the Model Context Protocol (Streamable HTTP, revision 2025-06-18) for agents. |
| [gizai:ml](/docs/reference/gizai/ml) | one interface over the estimators of gizlang:{linalg,trees,svm,cluster}, with the algorithms and default hyperparameters of scikit-learn, xgboost, LightGBM and CatBoost: fit, predict, predict_proba, the metrics, and the model bytes. |
| [gizai:netops](/docs/reference/gizai/netops) | the operations of networks on gizai:tensor Tensors. |
| [gizai:nn](/docs/reference/gizai/nn) | what training a network needs on top of gizai:grad. |
| [gizai:npullm](/docs/reference/gizai/npullm) | Gemma 4 E2B's forward pass on an AMD NPU (XDNA 2), with the AIE kernels of FastFlowLM (the layer and lm_head xclbins, MIT runtime, binaries free to use unmodified). |
| [gizai:omnivoice](/docs/reference/gizai/omnivoice) | k2-fsa's OmniVoice, text to speech in more than 600 languages, as its reference package and giz's worker run it: a voice from a short recording and its text (a clone), or from a few attributes (a design).  The model is Qwen3 0.6B as a bidirectional transformer over [text tokens \| reference codes \| target codes] of the Higgs codec (gizai:higgs: 8 codes of 1025 values, 1024 the mask, 25 frames a second). |
| [gizai:onnx](/docs/reference/gizai/onnx) | ONNX model files (ModelProto, IR version 8) of gizai:ml models, and gizai:ml models of ONNX files. |
| [gizai:quant](/docs/reference/gizai/quant) | the formats of stored weights, apart from the model families and the kernels that run them. |
| [gizai:svm](/docs/reference/gizai/svm) | support vector machines. |
| [gizai:tensor](/docs/reference/gizai/tensor) | n-dimensional f32 arrays on the CPU's vector lanes and cores. |
| [gizai:tile](/docs/reference/gizai/tile) | GPU kernels as ordinary Gizlang over tiles, on every device.  A tile program says what one block of the output is; this module decides everything else: threads, which thread holds which element, workgroup memory, barriers, bounds masks, matrix-core fragments, layout conversions and the device. |
| [gizai:tokenizer](/docs/reference/gizai/tokenizer) | text into model inputs. |
| [gizai:trees](/docs/reference/gizai/trees) | decision trees and their ensembles: CART (scikit-learn's exact splitter and its random splitter for extra trees), forests, bagging, AdaBoost (SAMME and R2), gradient boosting, and one histogram GBDT with depth-wise, leaf-wise and oblivious growth that the xgboost, LightGBM, HistGradientBoosting and CatBoost defaults share. |
| [gizai:video](/docs/reference/gizai/video) | video on top of gizhw:codec and gizlang:mp4. |

gizai

Tensors on the CPU and GPU, gradients, GPU kernels, and language, image and video models. Import a module by name, for example import * as Store from "gizdb:store";. Each page lists the module's public declarations exactly as an installed toolchain provides them.

Module	Purpose
gizai:a2a	the Agent2Agent protocol (0.3, JSON-RPC over HTTP) for agents.
gizai:agent	an agent is a model, its instructions, typed tools and the conversation so far.
gizai:ai	AI models on this computer, by name.
gizai:cluster	clustering with a predict step: k-means (k-means++ and Lloyd), mini-batch k-means, BIRCH (CF subclusters, then Ward agglomeration by the nearest-neighbor chain), affinity propagation and mean shift.
gizai:diffusion	video and image generation models (diffusion transformers), fast on any hardware. A model family is a recipe: its config.json read as an Arch, and the names of its tensors.
gizai:grad	derivatives of tensor programs.
gizai:grammar	the texts a model may write, for constrained decoding.
gizai:h3	giz's worker of MiniMax H3 (gizai:diffusion's D.H3): its video.generate jobs as WanGP makes them (sliding windows, a video to extend, control videos, Ref2VA references, a soundtrack) and the enhance-video and interpolate-video jobs (RTX Video and optical flow on the same GPU).
gizai:higgs	Boson AI's Higgs audio tokenizer v2, the codec of OmniVoice: 24 kHz mono audio as 8 codes of 1024 values every 960 samples (25 a second), and back. Encoding joins two views of the audio at 25 Hz: DAC's acoustic encoder of the 24 kHz samples (256 values a frame), and the mean of HuBERT base's 13 hidden states of the audio at 16 kHz, every second frame, through a semantic encoder (768 values).
gizai:llm	language models, fast on any hardware. A model family is a recipe: its config.json read as an Arch, and the names of its tensors.
gizai:ltx	Lightricks' LTX-2.5 (22B), the audio and video flow transformer of giz.ai's ltx2-5-distilled. Two streams of rows run side by side through 48 blocks: video (4096 wide, 32 heads of 128; a row a latent pixel of 8 frames and 32 x 32 pixels) and audio (2048 wide, 32 heads of 64; a row a 40 ms latent).
gizai:mcp	the Model Context Protocol (Streamable HTTP, revision 2025-06-18) for agents.
gizai:ml	one interface over the estimators of gizlang:{linalg,trees,svm,cluster}, with the algorithms and default hyperparameters of scikit-learn, xgboost, LightGBM and CatBoost: fit, predict, predict_proba, the metrics, and the model bytes.
gizai:netops	the operations of networks on gizai:tensor Tensors.
gizai:nn	what training a network needs on top of gizai:grad.
gizai:npullm	Gemma 4 E2B's forward pass on an AMD NPU (XDNA 2), with the AIE kernels of FastFlowLM (the layer and lm_head xclbins, MIT runtime, binaries free to use unmodified).
gizai:omnivoice	k2-fsa's OmniVoice, text to speech in more than 600 languages, as its reference package and giz's worker run it: a voice from a short recording and its text (a clone), or from a few attributes (a design). The model is Qwen3 0.6B as a bidirectional transformer over [text tokens \| reference codes \| target codes] of the Higgs codec (gizai:higgs: 8 codes of 1025 values, 1024 the mask, 25 frames a second).
gizai:onnx	ONNX model files (ModelProto, IR version 8) of gizai:ml models, and gizai:ml models of ONNX files.
gizai:quant	the formats of stored weights, apart from the model families and the kernels that run them.
gizai:svm	support vector machines.
gizai:tensor	n-dimensional f32 arrays on the CPU's vector lanes and cores.
gizai:tile	GPU kernels as ordinary Gizlang over tiles, on every device. A tile program says what one block of the output is; this module decides everything else: threads, which thread holds which element, workgroup memory, barriers, bounds masks, matrix-core fragments, layout conversions and the device.
gizai:tokenizer	text into model inputs.
gizai:trees	decision trees and their ensembles: CART (scikit-learn's exact splitter and its random splitter for extra trees), forests, bagging, AdaBoost (SAMME and R2), gradient boosting, and one histogram GBDT with depth-wise, leaf-wise and oblivious growth that the xgboost, LightGBM, HistGradientBoosting and CatBoost defaults share.
gizai:video	video on top of gizhw:codec and gizlang:mp4.