What Tenon is
Tenon is a native macOS terminal workspace for people running several CLI coding agents at the same time — Claude Code, Codex, or anything else that lives in a terminal.
It is a supervision layer, not an orchestrator. Agents keep running in their own harness, in real PTYs, with their own planning and their own tools. Tenon owns the part nobody else does: the operator's situation awareness, and the return path from a summary back to the thing that actually happened.
The problem it is shaped around
Starting five agents is easy. Following five agents is not.
The binding constraint on parallel agent work is not how many processes your machine will run. It is how many workstreams one person can hold, and — more expensively — how long it takes to re-enter one after looking away. Every context switch costs re-reading a transcript to find out what changed, whether the agent is stuck, whether its claim is true, and what it needs from you.
Tenon exists to make that cost smaller. It should let you answer five questions without reopening every transcript:
- What materially changed since I last looked?
- What requires my judgment now?
- What is the agent claiming, and what evidence supports it?
- Which work is blocked, drifting, stale, or in conflict?
- What can I safely ignore for now?
What it deliberately does not do
This list is as much the product as the feature list is.
| Tenon does not | Because |
|---|---|
| Plan work or decompose tasks | Your agent's harness already does, better, with its own context |
| Spawn, schedule, or supervise agents for you | That is orchestration; it belongs to the harness |
| Drive another agent's keyboard | No intent exists for it, on purpose |
| Re-implement the terminal | Panes are real libghostty surfaces with real PTYs |
| Replace the transcript with a summary | A summary that cannot be checked is a rumour |
| Run in a browser or over a network | It is a local, native macOS app |
Use your existing harness to start and coordinate agents; Tenon focuses on supervising their output.
What it is made of
A workspace is a directory, and it appears in the left sidebar. Each workspace owns tabs. A new tab opens as one terminal filling the whole canvas. That canvas is a 12 × 12 grid of panes you can split, drag, swap and resize; a pane can hold a terminal, a file, a diff, a web preview, or a view contributed by a plugin.
A pane keeps its identity — and its live process — through every move, resize, tab switch and workspace switch. Close the app and the layout comes back; the processes do not, because Tenon never pretends to resurrect a process it did not keep alive.
Around that sit four things worth knowing about early:
- Agent Lens reads a supported agent's own session record and renders one chronological timeline of what it did, with the raw evidence one click away. When it cannot bind a session with authority, it says so and hands you the terminal instead of guessing.
- The command palette (
⌘K) is where plugin-owned actions surface. It projects intent contracts; it is not a second, parallel command system. tenon-cliis in every terminal Tenon opens, so a shell — or an agent inside a pane — can inspect and drive the workspace it is sitting in.- Plugins declare a
manifest.jsonand share one public boundary. JavaScript plugins use amain.jsthat hot-reloads when you save; bundled implementations use the same boundary.
Who it is for
People who supervise agents rather than watch a single one. If you run one agent in one terminal and read every line it prints, a terminal you already like is a better tool than this. Tenon starts paying at three concurrent workstreams and keeps paying as that number grows.
Where to go next
- Install — download, verify, and run it.
- Your first workspace — a walkthrough from empty window to two agents working side by side.
- Concepts — why the architecture looks like this, if you want the reasoning before the buttons.