Local by default
Use the model server you already trust and keep your workspace close to the tools doing the work.
Why Truss
Coding agents should adapt to your workflow.
Truss is a local-first agent runtime that keeps models, tools, context, and permissions consistent across every way you work.
Why Truss exists
The interface should never become the platform.
Most coding agents are built around one editor, one provider, or one hosted service. That makes the first experience simple, but it also makes every future choice harder.
Truss starts with the runtime instead. The CLI, terminal UI, VS Code extension, desktop app, and future clients all use the same agent foundation. You can change how you work without rebuilding the agent around a new interface.
A different foundation
Control where it matters. Consistency everywhere else.
Truss keeps the important choices explicit without making you configure the same workflow twice.
Boundaries you can see
Choose what the agent can read, change, and run, then review sensitive actions before they happen.
One runtime, many ways to work
Carry the same models, tools, permissions, and workspace state between the shell, terminal, editor, and desktop.
Built to evolve
Change the parts without replacing the whole.
Truss is TypeScript-first, event-driven, and organized around focused interfaces. New providers, tools, memory systems, and clients can grow independently while the core agent loop stays understandable and testable.
- 01Model providers
- 02Tools and permissions
- 03Context and memory
- 04Plans and sessions
- 05Client interfaces
- 06Storage and execution
What that means for you
More control without more friction.
Start where you are
Connect Ollama, LM Studio, llama.cpp, or another compatible endpoint without adopting a hosted control plane.
Stay in the loop
See the context, tool activity, diffs, and approvals that shape the agent's work.
Keep moving
Resume conversations and plans with useful workspace state instead of rebuilding context every time.
Ready to try it?