Multi-agent coding
Several agents. Separate sessions. One place to supervise.
tlbx keeps terminal-native workers and structured ACP agents in durable, separate sessions with their own repository and browser context.
The working model
Each worker gets the session surface it actually supports.
Normal agent CLIs run unchanged in native PTYs. ACP-capable agents can instead use provider-backed Agent Controller sessions with structured turns, tools, diffs, questions and approvals. tlbx keeps both surfaces visible without reconstructing agent meaning from terminal text.
Separate the work
Give concurrent writers separate branches or checkouts. Keep read-only analysis together only when file ownership cannot collide.
Choose the surface
Use a Terminal Session for normal CLIs and TUIs, or Agent Controller for a compatible ACP runtime.
Supervise by state
Use session activity, attention, Git state, terminal output and browser evidence to see what is actually blocked or complete.
Reconcile deliberately
Review and integrate the useful work into the primary checkout, then retire temporary workers instead of leaving an invisible fleet behind.
What tlbx adds
Two execution surfaces. One supervision layer.
- Terminal
- Any CLI, shell or TUI runs unchanged in a real PTY with server-side scrollback.
- Agent Controller
- Compatible ACP agents expose canonical history, tool calls, diffs, questions and approvals as structured UI.
- Control
- CLI and API helpers can select sessions, send prompts, inspect output and drive scoped browser validation.
- Shared context
- Both session types keep repository state, files, browser tooling and navigation in the same self-hosted workspace.
Good fit
Parallel feature lanes, independent review and implementation, long builds, browser-backed validation and work that must survive laptop or browser changes.
Bad fit
Two agents editing the same files without ownership, phantom background workers with no terminal identity, or concurrency added only to make a small task look larger.
Human boundary
tlbx makes work observable and steerable. It does not remove the need to review destructive actions, reconcile branches or decide what should ship.
Common questions
Multi-agent coding FAQ
Does tlbx call the model provider?
Terminal Sessions run the provider’s normal local CLI. Agent Controller Sessions connect to the provider’s local ACP runtime. In both cases, authentication and model usage remain subject to that provider.
Can different agents run at the same time?
Yes. Each runs in an independent Terminal or Agent Controller session. Concurrent writers should use separate branches or checkouts to prevent file collisions.
Can an agent control the browser?
Yes, when explicitly authorized. tlbx exposes scoped browser status, navigation, query, click, text and diagnostic helpers.
Does tlbx replace Git worktrees?
No. tlbx supervises sessions and repositories. Git branches and worktrees remain the isolation mechanism when multiple workers write concurrently.