OrbySoft

Product

Four parts. One channel.

This is not a white-label chatbot. It is the place a network trains, isolates, and serves AI.

Agents

An agent per partner, with its own tools and memory. A stable endpoint, a login, a cap. You decide what they can see. The runtime keeps that promise.

  • Gemini or Gemma through Vertex AI, or another model you name
  • No shared context between tenants
  • Hooks into the CRM or ERP the partner already has

Training

Fine-tune as a job, not a six-week project. Metrics go to BigQuery. You get a versioned endpoint.

  • SFT or LoRA on Gemma, or an adapter you bring
  • Datasets in Cloud Storage, with a life cycle
  • Promote a model to one partner without promoting it to all

Sandboxes

A short-lived box for code, a file, or a tool the agent should not run on a laptop. Snapshot, restore, a ceiling.

  • Our image or the channel’s
  • Tight network, secrets injected
  • It turns off when the job ends

Functions

A function becomes HTTP. It follows channel traffic and sits quiet after hours. You get a log and a line on the bill.

  • Python SDK to declare the agent
  • A gateway with a cap per partner
  • Queues for batch, webhooks for events

How a first install goes

01 · A call

We map the network, the data that cannot leak, and the first agent worth standing up.

02 · One environment

A Google Cloud project, São Paulo, one pilot partner, a spend cap.

03 · The rest of the channel

Onboard the others, keep your brand on the front, one invoice on the back.

Ask for access