Pushary

Pushary is built by RalphNex OÜ, an Estonian company founded in 2024, and created by founder Aadil Ghani to solve a specific pain point: AI coding agents that pause mid-task waiting for a human decision, sometimes for 30+ minutes. The tool routes those decision requests to your phone, Slack, or web dashboard so you can respond from anywhere and let the agent continue. It is available as a web app with companion iOS and Android apps.

What It Is

Pushary is a human-in-the-loop notification and approval layer for AI agents. When an agent like Claude Code, Codex, Cursor, Gemini CLI, Hermes, or Windsurf reaches a step it should not take autonomously — running a shell command, pushing to git, deploying to production — it pauses and sends the question to Pushary. Pushary then delivers that question as a push notification to your phone, a Slack message with Approve/Deny buttons, or a web dashboard card. Your answer flows back to the agent, which then continues. The tool also supports three question types: approve/deny, multiple choice, and free-text reply.

How the Workflow Works

The setup takes under two minutes via a single CLI command (npx @pushary/agent-hooks setup) that configures an MCP server, permission hooks, event tracking, and an API key. Once connected:

• The agent pauses at a risky or irreversible step and sends the question title, body, and tool name to Pushary
• Pushary delivers the question to your phone (iOS/Android), Slack, or web app
• You approve, deny, choose an option, or type a reply from your lock screen
• The answer flows back to the agent, which resumes
• When the task finishes, you get a completion ping

Source code, file contents, and secrets never leave your machine. Only the decision question itself travels through Pushary's servers.

Permission Policies and Fleet Board

Pushary lets you set per-tool permission rules once and apply them across every connected agent. Rules map a tool pattern to a timeout, a timeout action (auto-deny or hand back to terminal), and a delivery channel. Low-risk reversible steps can be auto-approved; destructive or irreversible actions always escalate. The Fleet Board shows all running agents in one view, which ones are waiting on you, and the full audit trail. Every question, tool involved, and decision is recorded with timestamp and actor for review and export.

Integrations and Platform Support

Pushary works with a broad set of AI coding agents and surfaces:

• Agents: Claude Code, Codex, Gemini CLI, Cursor, Hermes, Windsurf, Cline, GitHub Copilot, Warp, Kilo Code, Zed, Continue, Lovable, and any MCP-compatible agent
• Delivery channels: iOS app, Android app, Slack (with inline Approve/Deny buttons), web dashboard, and n8n (via a single workflow node)
• Developer access: TypeScript SDK, REST API, MCP server, and signed webhooks for teams embedding approvals in their own products

Architecture and Privacy Model

According to the About page, Pushary's backend uses Next.js with React Server Components, PostgreSQL on Neon with Row-Level Security, Apache Kafka for event streaming, and a dedicated worker service with exponential-backoff retry and dead-letter queuing. Infrastructure is hosted in Germany within the EU. The company is registered in Estonia (RalphNex OÜ, Registry Code: 16932562). An optional transcript sync feature is end-to-end encrypted on the user's device before transmission, so Pushary stores only ciphertext it cannot read.

Current Status

Pushary is actively available with iOS and Android apps published in their respective stores. The product was featured on Product Hunt (top post badge) and Hacker News. The homepage shows live agent integrations and a growing list of supported tools. The pricing page lists three developer tiers plus a Partner tier for teams embedding approvals in their own products, and a separate e-commerce push notification product on the same account.