Show HN: Pulse – Dashboard for Claude Code, approve tool calls from your phone
Pulse, developed by Nikita Doudikov and introduced on Hacker News, is a local, zero-dependency companion dashboard designed specifically for Claude Code, Anthropic’s terminal-based AI assistant. While terminal-based developer agents offer powerful local code manipulation, they inherently tie the developer to the command line to monitor active progress and manually approve tool executions. Pulse addresses this limitation by transforming local terminal logs into a real-time web dashboard and a remote authorization interface. This software is built for engineers and developers who run long-running agentic workflows and require a way to monitor token spend, recover interrupted sessions, and approve tool execution without being anchored to a single terminal window.
Technically, Pulse operates through three core mechanisms. First, it utilizes passive log parsing. Instead of wrapping or modifying the Claude Code binary, Pulse reads the agent's raw JSONL session files under ~/.claude/ in a read-only manner. It caches these files based on modification timestamps to minimize CPU overhead, aggregates token usage, and pushes live updates to a dashboard served on 127.0.0.1 via Server-Sent Events (SSE). Second, Pulse hooks directly into Claude Code's execution lifecycle using native JSON configuration parameters (Notification, Stop, and PreToolUse). When Claude Code encounters a tool call requiring user permission, the PreToolUse hook routes the request to Pulse, which can dispatch push notifications to a mobile device. This allows developers to safely authorize or deny actions from outside their local network without configuring complex port forwarding or VPNs. Third, the tool ensures session resilience by auto-snapshotting active JSONL logs into lightweight Markdown files, enabling developers to reconstruct transcripts and execute full-text search across all historical terminal sessions.
Going forward, Pulse establishes a compelling blueprint for non-custodial, local-first agent orchestration. It demonstrates that developers can achieve asynchronous, human-in-the-loop control over autonomous agents without relying on third-party SaaS wrappers, cloud telemetry, or external network dependencies. This paradigm could heavily influence the design of future developer tools, proving that agentic workflows can be made secure, observable, and highly mobile-friendly while maintaining complete local data sovereignty.
Please note that the source technical document analyzed for this article was truncated at the end of the text.