Your question.
Your data.
Your model.

Build a specialized prediction model for your own work or the product you’re creating. Prepare the data, check the results, and use the version that fits your task.

Get early access

Describe what you want to learn.

Connect Attio, Pylon, or Fireflies and describe your outcome. Decisions Lab drafts a data preparation plan for review, helping turn existing records into examples for training.

Review the inputs and known answers, then keep separate examples for testing.

Explore integrations

Train. Test. Compare.

Train a specialized model and evaluate it on examples it hasn’t learned from. Inspect mistakes, available baseline comparisons, and results for each question. Each experiment keeps its own data and evaluation.

Compare the results before choosing a version. Predictions are estimates, and training does not guarantee better accuracy or future outcomes.

Use your model where you work.

Ask your deployed model a question with text context, or use supported questions over fresh connected data. Call it through the API to add predictions to your app, or process records in a saved workflow.

Your project can stay for your own use. Publishing and monetization are optional future capabilities.

Try a sample prediction

A growing world of decision models.

We’re starting with Clef-Flash and Kev, fine-tuned to your data with LoRA. Get to know the wider ecosystem, too.

Initial modelOpen source
Clef Cloudflare · Clef-Flash
Train with Decisions Lab
Initial modelOpen source
Kev Jared Palmer · 0.8B & 4B
Train with Decisions Lab
Explore the ecosystemClosed source
Jev TypeSafe AI
Read research
Explore the ecosystemClosed source
Decision-1 Microsoft · hosted API
Read research
Explore the ecosystemOpen source
Convai Innovations
Train with Decisions Lab
Explore the ecosystemOpen source
OpenJev AlexWortega
Train with Decisions Lab
Explore the ecosystemOpen source
SemIf TheoLeeCJ
Train with Decisions Lab
Explore the ecosystemOpen source
Nimble Bespoke Labs
Train with Decisions Lab
Fine-tuning methodOpen source
LoRA Low-rank adaptation
Train with Decisions Lab

Ready to train today: Clef-Flash, Kev 0.8B and Kev 4B. More open models are coming to Decisions Lab; closed models link to their research.

One model per question.

Example models for Attio, Pylon, and Fireflies. Each answers new records with how likely each answer is.

Deal close forecastTrained on Attio deals
Will this open deal be won in the next 30 days? Yes / no
1. Renewal, 40 seatsBudget approved. Contract in review.
  Yes78%
2. New customer, mid-marketNo reply since the demo two weeks ago.
  No81%
3. Upgrade to annual planPricing agreed. Waiting on signature.
  Yes86%
4. Expansion, EU teamChampion engaged. Security review pending.
  Yes62%
5. Pilot extension, retailPilot ends next week. No buyer on calls.
  No67%
6. Inbound, 12 seatsTrial active. Asked about SSO.
  Yes58%
Ticket SLA riskTrained on Pylon tickets
Will this ticket breach its SLA in 24 hours? Yes / no
1. Login errors in one workspaceHigh priority. Waiting on engineering.
  Yes69%
2. Invoice address changeRoutine request. Agent assigned.
  No88%
3. Data export stuckEscalated twice. No owner yet.
  Yes74%
4. API rate limit questionSecond follow-up from the customer.
  Yes57%
5. Password reset requestSelf-serve link sent.
  No92%
6. Webhook failures after updateSeveral customers affected. Fix in progress.
  Yes63%
Meeting follow-upTrained on Fireflies meetings
Will this meeting lead to a next step within a week? Yes / no
1. Pricing call with ops leadAsked for a proposal by Friday.
  Yes84%
2. Intro call, early stageNo timeline discussed.
  No73%
3. Quarterly review, key accountRenewal date confirmed on the call.
  Yes81%
4. Technical deep diveAgreed on a pilot scope.
  Yes77%
5. Check-in with past championContact has changed roles.
  No66%
6. Demo for the finance teamRequested a security questionnaire.
  Yes70%

Keep building on
what you learn.

Schedule refreshes and retraining using your saved setup. Each run creates a new version, with the data and results available to inspect.

Evaluate the candidate and choose when to deploy. Scheduled retraining does not automatically replace the model you’re using.

Initial training and deployment currently need setup support. The training workflow starts with a supported integration. Self-service local deployment and paid publishing are not available today.

Talk through your setup