Steer AI agents from the palm of your hand.
A universal, multi-provider approval gate for developers building with coding agents. Route decisions from any provider—including Claude, Codex, OpenCode, or custom workflows—to a single mobile inbox, keeping your agent stack flexible and open.
Execute database schema migration for billing tables
How it works
A short approval loop for autonomous agents.
Don't stall your workflows or leave terminals vulnerable. Route gated actions directly to push-ready cards.
Agent Reaches a Gate
An AI agent (running code locally or on a server) requests permission for a terminal command, file change, or API call.
Mobile Card Delivered
UpSteer translates the raw execution context into a beautifully structured mobile notification with the logs needed to make a choice.
Decide & Resume
Approve, reject, or request changes with a single tap. The agent resumes execution instantly and the audit logs are cataloged.
Terraform Apply?
Production AWS Infrastructure Change requested.
Features
Full operational awareness, in your pocket.
We designed UpSteer to bring high-fidelity control without desktop clutter.
Focused Mobile Cards
Rich metadata cards render exactly what the agent wants to do. Read diffs, command-line arguments, environment targets, and risk scores immediately on a layout optimized for quick decision-making.
Policy Engine
Define policies to auto-approve low-impact commands (like code formats or reading logs) while reserving mobile alerts for critical operations, database modifications, or external API relays.
Zero-Knowledge Context
Your raw code, tokens, and data values stay locally in your environment. The secure relay only receives structure and generalized metadata to generate notifications.
Always Audited
Every response, gate traversal, code suggestion, and request is signed and cataloged locally, giving you a comprehensive cryptographic audit trail.
Security & Privacy
Route decisions without exposing your source code.
UpSteer separates control signaling from data delivery. Your workspace content is never transferred to our servers, keeping operations secure by default.
End-to-End Signature Checks
Decisions are cryptographically signed using keys generated during agent onboarding. Relays verify authenticity before resuming runs.
Zero Storage Relay
Once a decision packet (approval, reject statement) is delivered back to your runner, context data is immediately wiped from the relay.
Local Agent Runner
Holds project code & sensitive env variables
UpSteer Secure Relay
Passes metadata with end-to-end encryption
Alex's iPhone
Approves commands with secure device keys
Pricing
Clear plans, zero surprise hosting costs.
Start free. Upgrade as your agent deployments grow and you require real-time status streams.
Developer
Ideal for solo testing and basic automation loops.
- 50 manual approval cards / month
- 1 active agent slot(Ideal for local coding loops like Claude/Codex, or any local agentic workflow)
- 7-day audit logs history
Solo Pro
Vibe Coder ChoiceNo-brainer safety for your background assistants.
- 500 manual approval cards / month
- 3 active agent slots(Run background dev loops like Claude/Codex, or custom agentic workflows concurrently)
- 2,000 live streaming minutes / month
- Auto-approve policy engine
- 7-day signed audit logs history
Power Solo
Power BuilderFor running a suite of background automations.
- 2,000 manual approval cards / month
- 10 active agent slots(Connect up to 10 independent coding, deployment, or custom agentic workflows concurrently)
- 5,000 live streaming minutes / month
- Auto-approve policy engine
- 14-day signed audit logs history
Team
CollaborativeFor startups and commercial builders.
- Unlimited manual approval cards / month
- Unlimited active agent slots(Run infinite multi-agent coding, CI/CD, or collaborative agentic workflows across your team)
- Unlimited live streaming minutes / month
- Collaborative policy rule engine
- 30-day signed audit logs history
FAQ
Frequently Asked Questions
Everything you need to know about setting up and using mobile-first agent approvals.
UpSteer provides a lightweight SDK and CLI bridge. You wrap your agent's terminal execution or tool invocation block (e.g., in Python, TypeScript, or Go). When the agent attempts a gated call, the SDK pauses execution and sends an encrypted request to the UpSteer relay, wait-looping until you tap a choice on your mobile device.
No. UpSteer is built with privacy at its core. Only the command query, description, and environment metadata (like risk parameters or labels) that you explicitly approve for transmission are sent to the relay. All actual execution, database handshakes, and codebase operations remain on your private machine or server.
If you're offline or do not respond within a configurable timeout window, the agent runner will trigger a fallback policy. Typically, it will gracefully pause or save its state to a resume point, allowing you to review and resume manually via the terminal once you are back online.
Yes. The UpSteer policy engine allows you to set up rules based on risk levels. For instance, commands that perform read-only actions (like git status, listing directory contents, or checking server health) can be configured to auto-approve, whereas writing actions (like schema changes or git push) will always require a mobile tap.