Run the panel with
your models.
Add your own custom, fine-tuned, or self-hosted models to the deliberation — from any OpenAI-compatible endpoint, or running inside your own AWS or GCP account. Your credentials, your billing. We’re the layer that makes them all reason together, not another model to lock into.
Two ways to shape the panel
Pick your variants, or bring your own
Tune the four-model panel without leaving TruVerifAI — or add models we don't host at all.
Preselect from the panel
Choose specific variants for each of the four providers — a budget model for routine work, a premium one when the stakes are high. Your choice sets the credit multiplier, so you only pay for the horsepower you pick.
Bring your own model
Add a model we don't host — a fine-tuned checkpoint, a specialist open model, a self-hosted endpoint, or one running in your own cloud. It joins the deliberation alongside (or instead of) the defaults, on your credentials.
Connect almost anything
The providers you already use
Direct keys, any OpenAI-compatible host, or inference inside your own cloud account. Add a model, paste its credentials, give it a name, and test it before it goes live.
Direct provider keys
OpenAI, Anthropic, and Google AI Studio with your own API key and a model ID — reach the exact variant you want, billed to your account.
Any OpenAI-compatible host
Together, OpenRouter, Fireworks, Groq, a local Ollama server, or any self-hosted /v1/chat/completions endpoint — one base URL plus a key.
AWS Bedrock
Invoke Bedrock models — Nova, Llama, Mistral, Titan, Claude, custom imports — on your own IAM credentials, in your region. Inference never leaves your account.
GCP Vertex AI
Gemini, Claude on Vertex, and Model Garden models (Llama, DeepSeek, Mistral) on your service account and project — including fine-tuned endpoints.
Fine-tuned & self-hosted
Your own fine-tunes via Bedrock custom imports or Vertex Model Garden, and local models over an OpenAI-compatible server — all first-class panel members.
Declare its capabilities
Optionally tell us a model's context window, max output, and whether it handles images, PDFs, tools, or JSON mode — so the panel routes work to it correctly.
Your keys, your billing
Zero variable credits on your models
A BYOM isn't a markup. You pay your provider directly at their rates; TruVerifAI charges only the fixed orchestration fee for running the deliberation.
No variable cost on BYOMs
Your custom models contribute zero to the credit multiplier — you're billed only the fixed per-mode orchestration credits, never a token markup. The inference bill lands with your provider, at their price.
Encrypted, never shared
Keys, AWS credentials, and service-account JSON are encrypted at rest (Fernet) and scoped to your account. They're never returned in an API response and never visible to other users — only a fingerprint is shown back to you.
Tested before it's trusted
When you add a model we run one tiny round-trip with your credentials to confirm they work before saving — so a typo'd key fails at setup, not mid-deliberation.
Setup
Add a model in a minute
One form, one test call, and it's in the panel.
Pick the connection
Choose how to reach the model — a direct key, an OpenAI-compatible base URL, AWS Bedrock, or GCP Vertex AI.
Add credentials
Paste the API key (or AWS keys / Vertex service-account JSON + project & region), then the model's ID in the provider's format.
Name it & test
Give it a friendly label and hit Test & add. We run a minimal call to verify the credentials before saving.
Use it in the panel
Your model appears in the preselection picker — add it to any query, in the app or over the API, alongside or instead of the defaults.
Bring your models to the table
Add a fine-tuned, specialist, or self-hosted model and let it deliberate with the rest — on your credentials, at zero variable credits.