Watcher is now self-serve

Apollo Research · August 2026

Your coding agent (Claude Code, Codex, Copilot, etc.) is taking hundreds of actions a minute. It reads files, runs commands, installs packages, edits configs, and pushes code. It runs autonomously, with your permissions, faster than you can read the diff. Most of the time that's exactly what you want. 1% of the time it's how credentials get exposed, your environment gets corrupted, or a model nukes your project while you're getting coffee.

We think everyone running agents deserves frontier-grade security, so we're making Watcher free and self-serve. Sign up and start monitoring your agents in real time.

Coding agent security is non-optional in 2026

Scan a running list of real agent incidents like this one and a pattern jumps out: the vast majority of incidents come from coding agents.

Agents already have as much access as the engineers running them, and they run around the clock without getting tired. The incidents are showing up in the data:

And existing guardrails don't suffice and mostly get ignored: 80% of developers say they bypass their organization's AI code security policies (Snyk, 2025). Security that slows you down gets circumvented. This is why we built Watcher from the ground up with the philosophy of keeping developers highly productive while improving their security posture.

What Watcher gives you

Watcher runs alongside your coding agent and scores every action in real time, only interrupting when it matters:

• Dangerous actions are blocked before they run. Leaking a secret, tearing down infra, piping curl to a shell, pushing code to a public repo: all caught inline, with an explanation for you to review.
• You continue to move fast. Reading files, running benign tests, and standard edits are approved without requiring your attention.
• Grounded in Apollo's safety research. The default monitoring pipeline is tuned from our research into tens of thousands of real coding-agent trajectories and 40+ failure modes, and extensively backtested against our golden datasets. You can have confidence in your guardrails.
• Organizational visibility and audit history. See every action your agent has taken, across your org's coding agent sessions. Watcher Analyzer flags dangerous actions and provides a full audit trail for you to review.

Get started (for free!) in under a minute

Run the installer: it sets up Watcher locally, wires hooks into your coding agent (Claude Code and Codex today), and connects to Apollo's cloud for grading and session history. There's nothing to provision.

curl -fsSL https://github.com/ApolloResearch/watcher-bin/releases/latest/download/install.sh | bash -s -- --cloud
# Restart or open a new terminal
watcher

Start your agent as you normally would and open Watcher Live to see your sessions, approval decisions, and anything that needs your attention in one place.

Learn more to protect yourself from agent incidents

If you want to learn about the different components of Watcher (Watcher Live / Analyzer), understand how the grading pipeline works, how to define organization-wide rules in YAML, or how self-hosted and admin-managed deployments fit in, read Watcher: technical details for teams.