Agent sessions that persist
Keep thread history, native models, modes, commands, authentication state, and streaming output across browser sessions.
Local-first agent workspace
Trit
Run coding agents with durable context, real terminals, and permissions you can actually see.
curl -fsSL https://trit.wtf/install | bashACP nativeRegistry and custom agents
DurableSQLite threads and audit history
Host awarePTY, tmux, and project roots
Self-hostedYour machine, your credentials
Workspace
One place for the parts of agent work that should not be hidden.
Terminals with real process state
Work in project-scoped PTYs or persistent tmux sessions beside the agent conversation instead of behind an abstraction.
Permission decisions with context
Review filesystem and terminal operations, retain their audit trail, and scope repeat approvals to the thread that earned them.
Local data with an exit plan
Store state in SQLite, back it up before migrations, and keep releases, configuration, and data independently replaceable.
Security
Private by default, explicit when exposed.
Agents and shells run as your service user. Trit keeps the origin on loopback and makes the deployment boundary visible.
Loopback-only origin
Verified remote identity
Exact browser-origin enforcement
Project-root filesystem boundaries
Distribution
A release you can inspect, install, and roll back.
Checksum-verified archives with a bundled Node runtime.
Atomic version switching with health-checked rollback.
Database backup and numbered migrations before restart.
User-level systemd service with journal-backed logs.