sparkient

Name: sparkient
Author: sparkient

Make structured decisions in under 100ms — faster than any LLM API call.

От сообществаРаботаетБез входаГлобальныйБесплатноТолько чтение

Что умеет

Make Decision: Make a structured decision in under 100ms. Requires a trained and deployed model for the given decision type. If no model is deployed, the API returns 428 (model_not_deployed). Use trai
Batch Decisions: Make up to 50 decisions in a single batch call. Each item in the list should have: - decision_type: str — name of the decision type - input_data: dict — input payload - request_id: st
List Decision Types: List all decision types in your organisation. Returns a paginated list of decision types with their name, description, status, and option labels.

Какие данные видит

Нужен ли аккаунт

Не нужен: сервер работает без входа

Make structured decisions in under 100ms — faster than any LLM API call. Define any decision type your app needs, train a lightweight ML model from AI-generated examples, and get instant decisions with confidence scores and reason codes. No ML team or historical data required. From moderation to routing to fraud — if an LLM can judge it, Sparkient can compile it.

Список инструментов сервера (13)

Технические названия из tools/list. Нужны только разработчикам.

make_decision	Make a structured decision in under 100ms. Requires a trained and deployed model for the given decision type. If no model is deployed, the API returns 428 (model_not_deployed). Use train_model to train and deploy a model first. The decision goes through a multi-stage pipeline: 1. CEL rules — deterministic, <1ms 2. ML classifier — ONNX model, <10ms 3. LLM escalation — Gemini fallback, only if low confidence Every decision returns: decision, confidence, reason_codes, latency_ms, stage, and whether it was escalated.
batch_decisions	Make up to 50 decisions in a single batch call. Each item in the list should have: - decision_type: str — name of the decision type - input_data: dict — input payload - request_id: str (optional) — idempotency key
list_decision_types	List all decision types in your organisation. Returns a paginated list of decision types with their name, description, status, and option labels.
get_decision_type	Get the full configuration of a specific decision type. Returns the complete decision type including name, description, options, reason codes, rules, input schema, training status, active model version, and confidence thresholds.
create_decision_type	Create a new decision type. A decision type defines a category of decisions your system makes. For example, 'content_moderation' with options ['approve', 'reject', 'escalate'].
add_examples	Add labelled training examples to a decision type. More examples improve model accuracy. Aim for at least 50 examples per option, with balanced class distribution.
generate_examples	Generate synthetic training examples using AI. Uses the decision type's description and options to generate realistic labelled examples. Good for bootstrapping a new decision type before you have real data.
train_model	Trigger model training for a decision type. Training runs asynchronously. The pipeline: 1. Extracts features from examples (numeric, categorical, text embeddings) 2. Trains a decision classifier with automated optimization 3. Compiles the model for fast inference 4. Optionally deploys the model for live inference Available presets: - 'fast' — quick training, faster optimization - 'balanced' — default, good accuracy/speed trade-off - 'thorough' — maximum accuracy, more thorough optimization
get_training_status	Check real-time training progress for a model being trained. Returns the current training stage, progress percentage, elapsed time, and a list of completed stages. Poll every 3-5 seconds while training is in progress. The training pipeline has multiple stages: 1. Generating training data (if needed) 2. Preparing and balancing data 3. Analyzing input features 4. Training text analysis model (if text fields present) 5. Training decision classifier 6. Evaluating model quality 7. Finalizing and packaging
get_decision_logs	Query past decision logs for a decision type. Returns a paginated list of decisions including input data, decision, confidence, stage, latency, and timestamp. Useful for auditing, debugging, or finding examples for retraining.
get_metrics	Get org-level aggregate metrics. Returns summary statistics for your organisation including total decisions, decisions today, average latency, and per-decision-type breakdowns.
get_credits	Check your organisation's current credit balance. Returns how many credits remain this billing period, the total allocation, percentage used, plan tier, and when credits reset. Use this before running expensive operations (batch decisions, training, example generation) to ensure you have enough credits.
export_edge_bundle	Export a trained model as a standalone edge bundle (ZIP). Downloads the active deployed model, feature config, CEL rules, and metadata as a self-contained ZIP file. The bundle can run offline with zero cloud dependencies using the sparkient-edge SDK: from sparkient_edge import EdgePredictor predictor = EdgePredictor.from_bundle("bundle.zip") result = predictor.predict({"text": "hello"}) Requires a deployed model (call train_model first). Requires Growth plan or above.