Bedrock: The Human-in-the-Loop Control Room for Coding Agents

Description and use

Bedrock consolidates management of multiple coding agents into a single control room, providing persistent terminal sessions, AI‑based monitoring, and human approval gates to reduce developer supervision overhead.

Problem

Supervising multiple coding agents requires constant terminal switching, monitoring output, handling permission prompts, and reconstructing context, which consumes developer attention.

Outcome

Bedrock centralizes agent management, preserves agent lifecycles across UI restarts, and introduces human approval gates for risky actions, thereby streamlining supervision of multiple coding agents.

Workflow

• Launch Bedrock web server and daemon
• Create or select a project in the Bedrock UI
• Start one or more coding agent sessions via the unified console
• Monitor real-time terminal output and status in the shared interface
• AI Supervisor performs deterministic pattern checks and model-based classification
• When a risky action or plan is detected, the Supervisor presents a pending plan to the user
• Human reviews the proposal and approves or rejects it
• Approved plans are injected into the agent’s runtime
• Audit trail records the decision and action
• If the web server restarts, the daemon preserves agent lifecycles and buffers output for reconnection

Evidence

Bedrock is an open-source, local-first control plane for running, observing, coordinating, and supervising multiple coding agents through persistent terminal sessions, structured runtimes, AI-assisted monitoring, and explicit human approval gates.

Source

The fast path checks normalized terminal output against known patterns for potentially dangerous commands and interactive prompts.

Source

The Supervisor can create a pending plan. The application presents that proposal to the human, who can approve or reject it.

Source