An AI workspace is where you set the model, cost limits, knowledge and tools once. Hooshvares are built on that base, tested, and work under your control.
01 · WORKSPACE
Connect a provider, set a model per capability, and own the prices and cost caps. With a decision model, the conversation itself decides which model or tool handles each step.
OpenAI today; aggregators like OpenRouter, Iranian providers AvalAI and Liara, and local models through an OpenAI-compatible endpoint.
Set separate models for chat, vision, audio and embeddings, from one provider or several.
A decision model picks the right model or tool for each step and stops risky tool calls before they run.
Editable price per model, a cost cap per workspace, and usage reports.
02 · RELATIONSHIP
A hooshvare is Sufi's name for an AI companion. Each has a display name, source module, workspace and status, and takes its model, key, cost cap, knowledge and tools from its workspace.
03 · BUILD & MAINTAIN
A hooshvare isn't built once and forgotten. Its whole life cycle is in the console:
04 · THREE KINDS
Each product has its own hooshvares, covered on that product's page. Here are the kinds:
Reads and recommends
Changes things after you confirm
Carries multi-step work
05 · COMPARE
| Assistant | Copilot | Agent | |
|---|---|---|---|
| Read & suggest | ✓ | ✓ | ✓ |
| Change data | — | After confirm | After confirm |
| Multi-step work | — | — | Coming |
| Use allowed MCP tools | ✓ | ✓ | ✓ |
06 · CONTROL
Every page and action has its own SufiAI permission.
The hooshvare proposes, a person decides.
Per workspace.
So nothing breaks by accident.
By default, and always cites sources.
AI infrastructure is part of the business modules. The free licence covers up to 100 users and 100 GB of database; AI provider costs are yours.