Installs locally. Approves on the web. Free.
The CLI runs on your machine, sees traffic from all your agents, and builds a single queue with pre-calculated risk. The same queue appears in the browser when you link the machine. Without changing Claude, Cursor, Codex, or OpenCode.
CLI for macOS, Linux, and Windows · Free on every plan · Desktop under construction
apply_coupon.rbinside the expected surface of AURA-118low riskrails db:migratecreates coupons table · reversiblemedium riskcheckout/read onlylow risklegacy/payments.rbtouches critical path · 312 lineshigh riskThree facts. Nothing more.
1 · It installs locally
The CLI watches your agents' traffic on your machine. macOS, Linux, and Windows. Automatic detection of known agents. You stay in the same terminal as always.
2 · You see and approve on the web
You link the machine with a pairing code. The queue appears in the browser. You approve or reject; the decision reaches the terminal. Without that link, nobody approves from outside.
3 · It is free
Runtime — CLI and web approval — is free on every plan. It does not call models to decide: risk comes from rules. Every person who uses it makes the platform better. Many projects at once? Orchestra organizes them by mission and by role. Included in Pro from 2027.
Better a short list than a long promise.
- —It does not write code. It is not an agent. It governs the ones you already use.
- —It does not orchestrate by mission or role. That is Orchestra, in Pro from 2027.
- —It does not test behavior. That is AURA QA, under construction, no date.
- —It is not a desktop app yet. Runtime Desktop is under construction.
- —It does not replace Review. The queue approves agent actions; the PR verdict lives on GitHub.
Runtime Desktop
A native window for the queue, without opening the browser. Same machine, same CLI, different shell. No date. The CLI and the web cover today's act: install, see, approve.
When Desktop exists, it will still be Runtime: free, local, without changing agents.
View Orchestra →No vendor is going to build this for you.
Anthropic, OpenAI, Google, and Cursor each build their own orchestrator — for their agent, with their subscription. Nobody will govern competitors' agents. Meanwhile, you jump between terminals.
Missed approvals
A hidden tab, a stopped agent, a lost afternoon. With ten agents, the problem multiplies by ten.
Blind decisions
Which model plans better? Which writes more reliable tests? Today: perceptions. Nobody has data.
Vendor cages
Every tool wants to be the only one. Switching agents when credits run out means switching the entire flow.
Install the CLI and keep working as usual.
Runtime watches your agents' traffic locally. Nothing changes in your flow, except that everything now goes through one place.
- macOS, Linux, and Windows
- Automatic detection of known agents
- Low / medium / high risk per action
# install (closed beta · the link arrives with your invite)$ aura runtime init✓ machine linked: cerebro (linux)✓ agents detected: claude-code, opencode, codex✓ approval queue active$ aura runtime watch[claude-code] write apply_coupon.rb low risk[codex] rm legacy/payments.rb high risk ⚠→ a / r / d (approve / reject / detail)
Free. On every plan. Forever.
Today it is in closed beta. Tell us how you work and we send access in order of arrival. Teams using three or more different agents get priority. Orchestra — flows, missions, roles — arrives with Pro from 2027.