Local-only
On desktop, Sanad Client can connect directly to Sanad Agent on the same computer. With a supported local model runtime, supported conversations and tools can run without an internet connection.
Sanad Agent · Open source · Local first · Cross platform
An AI agent that works where your work lives.
Sanad Agent combines a native agent that owns execution and local state with a Flutter client for this computer or paired remote devices. Choose local or hosted model providers, keep workspaces separate, and control the tools each workspace can use.
✓ MIT licensed✓ Local-first execution✓ Ollama, LM Studio, and llama.cpp✓ Verified release manifests
Download Sanad
Get the Sanad interface for your device.
Choose your platform below. Sanad recommends the right version for this device while keeping every option easy to find.
Downloads are verified against the stable release manifest. Unpublished apps stay visible and become available here when released.
Two modes, one codebase
Run local. Or connect your devices.
Start with one computer, then reach paired computers and headless servers from the same client when you choose.
Connected devices
Optional EastStar AI identity and relay services let you reach paired computers or headless servers from desktop, mobile, or web. Workspace files and execution remain on the agent device.
Built for controlled, durable work
What you get out of the box
Practical agent capabilities with visible context, explicit boundaries, and workspace-level control.
Durable, parallel conversations
Keep independent conversations running without mixing messages, providers, tools, or workspace context. Capacity depends on your computer and model providers.
Redirect work without throwing it away
Steer eligible active work at a safe boundary, queue separate requests, manage queued work, and recover input that did not run after Stop.
Providers on your terms
Configure hosted APIs, ChatGPT Subscription, supported local runtimes, and custom compatible endpoints with separate accounts and controlled failover.
Workspace-level control
Separate sessions, drafts, provider choice, context, permissions, and tool policy. Sensitive tools remain subject to approval and OS permissions.
Tools that grow with your workflow
Use built-in file, shell, web, memory, question, and scheduling tools. Add compatible MCP servers and skills.
Inspectable context
Project instructions, skills, memory, and identity files remain visible and editable instead of becoming hidden product state.
Experimental work stays clearly labeled
Realtime voice is hidden-by-default experimental work and is not part of the stable first-release promise.
Open source and hosted services
Sanad Agent and Sanad Client are available under the MIT License. Optional EastStar AI Portal and Gateway services provide identity, pairing, remote device inventory, and relay under the published Terms and Privacy Policy.
Start with one machine. Scale when you are ready.
Use local-only mode without an EastStar AI account, choose your model providers, and add connected devices only when your workflow needs them.