Stacktrace

Organizations are increasingly running autonomous coding agents (Claude Code, Devin, Codex) that take actions unsupervised, but teams lack a central way to see what these agents actually did, set limits on what they are allowed to do, and intervene when their behavior drifts from intent. Without this, risky actions can slip through, such as a credential-shaped value being sent to an external integration or a forced push to a main branch. Separately, companies whose products are used by customers' own personal AI agents have no way to verify those agents or hold back risky actions like refunds before they execute. Today these gaps appear to be handled ad hoc or not at all, since the page presents Stacktrace as the tool built to close them.

Stacktrace is a control plane for coding agents with two products: Fleet, which gives organizations a central place to see every agent session's findings, set policies ahead of time (for example blocking a specific MCP integration for every agent), and receive real-time alerts (including Slack notifications) when something like a blocked git push or a suspicious outbound request occurs; and Personal Agent Gateway (in preview), which verifies personal agents customers bring to a product and holds risky actions, such as a refund request, for customer confirmation before they execute. The product works with existing agents including Claude Code, Devin, and Codex, and also offers an in-browser sandbox so users can check an agent session directly.