
See who said yes to your team's AI agents
Pushary Teams makes AI coding agents like Claude Code, Codex and Cursor pause before risky steps and ask the right teammate on their Mac or phone, or a designated lead. Every question and answer is logged on a shared team page, so the whole team can see who approved what and when. Teams can also set routing rules, flag steps worth a second look, and export records for retros.
Agents like Claude Code, Codex and Cursor pause before a step you choose (e.g. a git push) and ask the responsible person, who can approve, deny, or answer from their Mac or phone.
One page shows every question and answer across tools: whose agent asked, who answered and when, including repository names, file paths and commands.
Write rules so certain steps route to a chosen lead — e.g. pushes to the infra repository ask the engineering lead first — while other questions go to whoever's agent it is.
Agents' questions appear in the Mac's notch on desktop and on the phone lock screen when away; both apps are included in every plan.
Mark rows as worth a look or follow-up, record the responsible person and next step, and export the record as CSV or JSON. No scores or ranking of people.
Works with Claude Code, Codex, Cursor, Antigravity CLI, VS Code, Hermes, OpenCode and more, with full MCP and API access (TypeScript, Python, REST and MCP on Partner plans).
Counts requests and answers for each person and flags steps worth a second look, such as a very fast yes on a risky step.
Companies that ship an AI agent can let their own users approve what it does from their phone, with usage limits by plan size.
You mark steps your agent should not take without a human — like pushing to main — and the agent stops and asks instead of running it.
On the Mac the question opens in the notch; away from your desk it appears on your phone's lock screen where you can approve, deny, or answer.
Each answer carries the name of who gave it, when, and what was asked, forming a shared record for the whole team across all connected agents.
Write a rule once — e.g. tool pattern Bash(git push:*) on the infra repository routes to the engineering lead — so the next incident ends with a better rule, not a guess.
Three self-serve plans: Agent at $9.99/month (5K notifications/month, 30-day history), Agent Pro at $19.99/month (unlimited notifications, budgets, Policy Autopilot suggestions, 90-day history), and Team at $59/month for 5 people (shared audit trail with export, team activity and summary email, 365-day history; extra seats $12/month each). Partner plans for product makers run $99 (Launch, up to 5K
$9.99/month
For one person, every agent. 5K notifications/month, every agent you run, approve/deny/answer from your phone, Mac app included, permission policies, full MCP and API access, 30-day history.
$19.99/month
For power users and small teams. Everything in Agent plus unlimited notifications, budgets, priority support, Policy Autopilot suggestions, 90-day history.
$59/month
For teams that need a shared record. Everything in Agent Pro, 5 people included (extra seats $12/month each), shared audit trail with export, team activity and a summary email, 365-day history, invite your security team.
$99/month
Approvals inside your product. Up to 5K of your users, 25K approval requests a month, 3 people included, 90 days of answer history, every answer sent to your server, TypeScript/Python/REST/MCP.
$299/month
Up to 25K of your users, 100K approval requests a month, 10 people included, 365 days of answer history, priority support.
$799/month
Up to 100K of your users, 500K approval requests a month, 25 people included, 730 days of answer history, priority support.
They send your own agent's questions to you, inside that one tool. Pushary Team gives everyone one record of who asked and who answered, across tools, without an Enterprise plan.
Everyone sees the same page, you included. It shows sessions, questions, answers, repository names, file paths and commands, and counts requests and answers per person. It flags steps worth a second look, such as a very fast yes on a risky step, but never scores or ranks anyone. You choose each person's access when you invite them.
It records what connected agents send to Pushary: requests, answers, and the shell commands and file edits they report. Answers in the phone app, Mac app and dashboard carry a name; answers from Slack, the web inbox or a web push link do not. A yes typed at the agent's own prompt is not recorded, and steps your rules approve on their own have no answer. An approval does not mean the work succeeded.
No. The workspace pays for its seats. Each teammate uses their own account and connects their own devices, and does not need a separate personal subscription.
Invite them from the Overview or Manage people. On the current Team plan, an invitation holds no seat; when someone joins and all seats are taken, a seat is added at $12/month, prorated on your next invoice. When someone leaves, the seat goes, but never below the 5 included.
A new workspace gets 3 days free, then pays $59/month for 5 people plus any added seats. Cancel before the trial ends and you pay nothing. Outside the US, checkout may charge the same numbers in € or £. Moving up from Agent or Agent Pro charges the difference that day unless you're still in your trial.
Contact Pushary for your security and data-processing review before purchase. SSO, SCIM and enterprise certifications are not included in this self-serve offer.
Remote Control is enough if Claude Code is the only agent you run and you sign in through claude.ai on a paid Claude plan. Add Pushary if you also run Codex, Cursor, Gemini CLI or OpenCode, or run Claude Code with an API key or on Bedrock.