Datadog Lapdog gives real-time visibility into AI coding agent sessions
Datadog Lapdog is a free local development tool that lets developers monitor AI coding agents like Claude Code, Codex, and Pi in real time, surfacing prompts, tool calls, token costs, and model usage in a browser dashboard.
Score breakdown
Lapdog offers a single-command alternative to setting up a full OTEL/Prometheus observability stack, giving developers local, real-time visibility into agent prompts, tool calls, and token costs without requiring a paid Datadog account.
- 01Datadog released Lapdog in May 2026 as a free local development tool — no Datadog subscription required.
- 02Lapdog supports Claude Code, Codex, and Pi agents.
- 03It captures data via two methods: a proxy for model traffic (prompts, responses, token metrics) and non-blocking hooks (tool calls, permission events, lifecycle events).
Datadog released Lapdog in May 2026 as a free local development tool for monitoring AI coding agents. The post, written by Chris Ebert, walks through installation, pairing with Claude Code, and interpreting a real session in the dashboard. Lapdog installs on macOS, Linux, Windows, or via Docker; on macOS with Homebrew, a single command (`brew install datadog/lapdog/lapdog`) is all that's needed. After installation, running `lapdog claude` registers non-blocking hooks with Claude Code and launches the agent through a proxy, enabling full token usage capture. Sessions, traces, and spans are then viewable in a browser dashboard hosted at lapdog.datadoghq.com, which reads data directly from localhost rather than a remote server.
First, it acts as a proxy for model request traffic, collecting prompts, responses, and token metrics.
Lapdog captures agent activity through two complementary mechanisms. First, it acts as a proxy for model request traffic, collecting prompts, responses, and token metrics. Second, it subscribes to non-blocking hooks that fire on tool calls, permission events, and lifecycle events. Support levels vary by agent: both Claude Code and Codex are monitored via proxy and hooks, while Pi is currently monitored via hooks only, meaning Pi sessions do not capture token usage. Sessions launched without the proxy path (plain `claude` or `codex`) also omit token data. By default, all session data remains in the local agent's memory and is lost on restart; Datadog customers can optionally forward sessions to longer-term storage using the `--forward` flag with an API key. Sessions can be cleared at any time by running `lapdog stop && lapdog start`.
Key facts
- 01Datadog released Lapdog in May 2026 as a free local development tool — no Datadog subscription required.
- 02Lapdog supports Claude Code, Codex, and Pi agents.
- 03It captures data via two methods: a proxy for model traffic (prompts, responses, token metrics) and non-blocking hooks (tool calls, permission events, lifecycle events).
- 04Claude Code and Codex are monitored via both proxy and hooks; Pi is currently hooks-only.
- 05Token usage data is only captured when the agent is launched through the proxy path (e.g., `lapdog claude`).
- 06On macOS with Homebrew, Lapdog installs with a single command: `brew install datadog/lapdog/lapdog`.
- 07Session data stays on-device by default; Datadog customers can forward sessions to longer-term storage using the `--forward` flag.
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 11, 2026 · 08:34 UTC. How this works →