AgentNava model
Our own model,
on your own servers.
cerebro-llm-1 is our own 30-billion-parameter model. Only the on-prem version of AgentNava uses it. It runs on your GPUs, inside your network, with no internet connection, and your documents and conversations stay with you.
- Name
cerebro-llm-1- Size
- 30B parameters
- Runs on
- Your own GPUs
- Used by
- On-prem AgentNava only
01 / What it is
A model tuned for the work agents do.
We start from a strong open model, one whose weights anyone can download, and tune it at Cerebro Labs. Three skills matter most for agents, and we test each one on real work.
Tools
It uses the tools an agent gives it.
It calls functions with structured arguments, so an agent can look things up and take the next step.
Documents
It answers from your documents.
It reads the files you give an agent and answers from them. When the documents do not cover a question, it says so.
Reasoning
It thinks when a task needs it.
For harder questions it can reason step by step before it answers.
Only the on-prem version of AgentNava uses cerebro-llm-1.
On agentnava.com, agents run on leading hosted models. Install AgentNava on your own servers and the same agents run on cerebro-llm-1, on your hardware.
02 / How on-prem works
Everything runs inside your network.
An on-prem deployment puts three parts on your own machines: the AgentNava platform that runs your agents, cerebro-llm-1 on your GPUs, and storage for the documents your agents read. Your developers reach the agents through the AgentNava API, from your own apps.
Your network
Your apps
Call agents through the AgentNava API
AgentNava
Runs your agents and their conversations
cerebro-llm-1
On your GPUs
Your documents
On your own storage
Internet
Not needed while agents run
03 / What stays with you
Your data stays where it is.
Nothing in an on-prem deployment depends on a service outside your network.
- Your documents
- Stored on your own storage. Agents read them inside your network.
- Conversations
- Questions and answers are kept on your machines.
- The model
- cerebro-llm-1 runs on your GPUs. No request goes to an outside AI provider.
- Internet
- Not needed while agents run.
- Your developers
- Call agents through the AgentNava API, or call the model directly with any OpenAI-compatible client.
04 / Questions
What teams ask us first.
Do agents on agentnava.com use cerebro-llm-1?
No. Only the on-prem version of AgentNava uses cerebro-llm-1. Agents on agentnava.com run on hosted models from leading providers.
Who is the on-prem deployment for?
Teams whose data has to stay in their own building: lenders, insurers, healthcare and HR.
Does any data leave our network?
No. The agents, the model and your documents all run on your machines, and nothing needs an internet connection while agents run.
Can our developers call the model directly?
Yes. It speaks the standard OpenAI-style chat API, so any OpenAI-compatible client works.
On-prem
Run agents inside
your own walls.
Tell us about your data and your hardware. We will tell you plainly what a deployment needs.