Sparkient trains a small model for each repeated decision your application needs to make.
Your coding agent can find repeated rules and live LLM calls in your codebase. You can test the strongest candidate before changing your application.
Latency depends on what runs
Check your codebase
Paste this prompt into the coding agent you already use. It will look for repeated rules and live LLM calls, then explain whether Sparkient would help.
Copy this prompt
“Research Sparkient online (start with https://sparkient.ai/llms.txt), inspect this project, and tell me whether Sparkient would be useful.”
The prompt sends your agent to Sparkient's current product, pricing, and evaluation guidance before it reviews the code.
How It Works
Set the possible answers, input fields, and any rules that must always apply.
Add labelled examples, or let Sparkient create a starting set. Sparkient trains and tests the model.
Call the model through the API. Growth and Scale plans can export a model to run locally. The cloud version can ask an LLM when it is unsure.
curl -X POST https://api.sparkient.ai/api/v1/decide \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"decision_type": "content_moderation",
"input": {
"text": "Check out this amazing product!",
"user_trust_score": 0.82
}
}'{
"decision": "approve",
"confidence": 0.94,
"reason_codes": ["CONTENT_SAFE", "TRUSTED_USER"],
"escalate": false
}Test the model on your own examples before using it in production.
Comparison
Sparkient runs a task-specific model. The cloud path can ask an LLM when the model is unsure.
Benchmarks
Each benchmark used 5,000 examples created and labelled by an LLM. The timing runner divided one batch duration by its item count; it did not measure per-request p95. These models have not yet been tested on customer data.
5 outcomes · 5,000 synthetic examples
Routes tickets to self-service, standard, urgent, critical, or escalation.
4 outcomes · 5,000 synthetic examples
Classifies content as allow, flag, restrict, or remove.
4 outcomes · 5,000 synthetic examples
Classifies messages as allow, mute, restrict, or ban.
4 outcomes · 5,000 synthetic examples
Reviews listings as approve, flag, restrict, or reject.
Each benchmark starts with 5,000 synthetic examples. The data has uneven class distributions and 5–8% label noise. Some edge cases are left out.
We train six baselines on the original training data. Sparkient uses an LLM to relabel training examples and create extra examples for missing cases. It then trains the compiled model. We evaluate the baselines and Sparkient on the same held-out synthetic test set. The baselines use structured fields only, while Sparkient also uses a text encoder, so their F1 scores are not a matched text-model comparison.
Paid plans start at $19 a month and include monthly credits. Decisions, training, and example generation use those credits.
Train and test one model
5,000 credits
For individual developers
10,000 credits/mo
For small production teams
50,000 credits/mo
For growing products
200,000 credits/mo
For high-volume deployments
1,000,000 credits/mo
A training run uses 2,000 credits. Generating and labelling examples uses additional credits.
Founder
Peter Dobson has spent 15 years building software. His previous work includes web products used by more than one million people a month and a multi-user VR platform.
He also built and ran a national retail business with 64 locations and £20M in annual revenue.
He built Sparkient's model training, API, dashboard, and developer tools himself. The product is deployed, pre-revenue, and now being tested with developers.
FAQ
Sparkient uses a large language model (LLM) to create or label examples. It trains a small model for one repeated decision. That model handles normal requests. If the cloud model is unsure, it can ask an LLM.
Across four published synthetic benchmarks, the compiled stage measured 33–42ms average time per item in batched runs. The current runner does not measure per-request p95. Simple rules can return in under 1ms, while an optional LLM fallback takes longer. Test the full API with your own workload before setting a production target.
You can upload labelled examples or ask Sparkient to create a starting set. Sparkient handles training, evaluation, and deployment. You still need to define the decision and review the results before production.
Yes. You can use the dashboard or REST API directly. A coding agent is another way to evaluate and set up Sparkient.
Sparkient's core API and data services run in Google Cloud's europe-west2 region. Data is encrypted in transit and at rest and isolated by organisation. Example generation and the optional LLM fallback use Google's global model endpoint, so review that path if you have strict residency requirements. You can export or delete account data through the API.
Decisions, training, examples, and model serving use credits. A training run uses 2,000 credits. The free trial includes 5,000 credits, one decision type, and 250 decisions. Paid plans include monthly credits, with optional top-ups.
Paste our audit prompt into your coding agent. If it finds a suitable repeated decision, you can test Sparkient free.