Local model discovery
Find Ollama, LM Studio, llama.cpp, and other compatible local servers, choose a model, and keep reusable workspace-aware profiles.
Current capabilities
One local agent platform, ready in every workspace.
Truss keeps the runtime independent from its interfaces. Its CLI, terminal UI, VS Code extension, and desktop app share the same local-model discovery, tools, permissions, and durable workspace state.
Available today
Four clients. One runtime.
CLI for automation, TUI for terminal-native work, VS Code for editor workflows, and a desktop app for a dedicated local workspace.
CLI and service mode
Run agents and deterministic workspace commands from the shell, or host the newline-delimited JSON service used by editor clients.
Full-screen terminal UI
Work in a keyboard-first TUI with files, editor, Git diff preview, chat, shell output, model settings, and approvals.
VS Code extension
Use streaming chat, bounded file attachments, inline completions, tool approvals, agent modes, and commit-message generation in the editor.
Standalone desktop app
Open a three-pane Electron workspace with a file tree, editor and diff preview, terminal, persistent chat, safe IPC, and Git stage, commit, pull, and push actions.
Durable agent workflow
Save workspace memory, implementation plans, Git-aware status, task notes, and conversation history—then resume with useful context.
Controlled tool execution
Read, write, list, search, grep, and terminal tools run through typed registration, Chat/Plan/Edit modes, and a chosen approval policy.
Streaming and interruption
Every client renders the runtime's token, tool, approval, and lifecycle events, including cancellation and fresh-session recovery.
Replaceable foundations
Providers, tools, context sources, session storage, memory, plans, and clients are isolated behind focused TypeScript interfaces.
Implementation details