Your email model
Trained on the openers, follow-ups, and cadences that booked meetings in your campaigns, in your brand voice.
Custom models
Your data, your own model
Fine-tune one specialized octa model per GTM task on your campaigns, replies, and outcomes, in early access.
tune on your data
Status
Early accessWant a model trained on your own data? Let us know. Custom models are in early access, not yet generally available. The capability below describes where they are headed: request early access to shape it with your campaigns.
01 / Your own octa series
A private model series specialized to your motion.
The shared Octa models learn from the whole corpus. A custom octa series learns from yours: your campaigns, your replies, your outcomes, your brand. One specialized model per GTM task, your email model, your campaign model, your qualification model, each tuned to what wins for you.
It compounds on your data
As your outcomes accumulate, your models sharpen on your data, not a shared average. Every booked meeting and every disqualified reply is a label that tunes the next pass toward your best results.
Your campaign model
Learns the multi-channel motions that worked for your offers and your segments, not a shared average.
Your qualification model
Scores inbound against your real ICP and the disqualification reasons your reps actually tagged.
02 / What it will do
A model that learns your motion, not the average one.
Continued pre-training plus long-horizon reinforcement learning on your own campaign data and outcomes, producing a model specialized to your org.
-
Train on your campaign telemetry
Continued pre-training on your own outreach, replies, and outcomes. The model learns the language and signals of your specific motion, not a generic average.
-
Optimize for your outcomes
Long-horizon reinforcement learning against your booked meetings and advanced deals. The objective is your GTM result, not token likelihood.
-
Specialize per org
An org-specific model that encodes your ICP, your proof, and your winning patterns directly in its weights, available to every Skill and agent you run.
03 / Skills and RAG vs custom
One composes a model with context. The other changes the weights.
Skills and RAG take a third-party model and supply your context at call time. A custom model trains on your data and changes the model itself. Use the first today; the second is where early access goes.
| Dimension | Skills + RAG Available now | Custom model Early access |
|---|---|---|
| Where the knowledge lives | Skills and RAG inject your context at call time, into the prompt. | Custom models move that knowledge into the weights, learned once. |
| What changes | Skills and RAG compose a third-party model with your data. | Custom models change the model itself on your data. |
| How it improves | Better prompts and better retrieval improve Skills and RAG. | More outcome data and more training improve a custom model. |
| When to reach for it | Reach for Skills and RAG first. They ship today and cover most needs. | Reach for a custom model when org-specific behavior is the differentiator. |
04 / Request access
Want a model trained on your motion? Let us know.
Custom models are an early-access program. We onboard orgs with enough campaign history to train against and a clear specialization to chase. Be honest with us about your data and we will be honest about fit.