Preflight

Preflight Observability for AI Coding Assistants

Docs • What's New • Examples • Community • Contributing

Why Your AI Tool Needs Observability

Your AI coding assistant makes hundreds of decisions every session — what to read, what to edit, when to run commands. But you can't see any of it. You know it was fast, but was it _efficient_? You got a PR merged, but how much did it cost? You fixed a bug, but did it get stuck in a loop first?

Preflight is observability for agentic coding — the actions, cost, and efficiency of your AI coding assistant as it works. See exactly what's happening, how much it costs, and where your AI is wasting time.

Local-first by design. Preflight runs entirely on your machine and sends your data nowhere by default. A live dashboard at localhost:7777 shows your sessions in real time, fully offline. Connect a New Relic account only when you want more — team rollups, alerting, and cross-session history. You choose: local-only, New Relic, or both.

Demo

See cost breakdown, efficiency scoring, anti-patterns, and live session tracking in action.

What You Get

Visibility

• Every action captured — file reads, edits, commands, searches
• Live session dashboard — see what's happening right now
• Historical trends — analyze patterns over weeks and months

Cost Control

• USD spend tracking — per session, day, and week
• Per-model and cache breakdown — know which models cost most and how efficiently context is reused
• Budget alerts — get notified before you overspend
• Forecasting — project monthly burn rate

Efficiency Insights

• Efficiency score — 0–100 score per task, based on how directly the AI worked
• Anti-pattern detection — catches re-reads, blind edits, stuck loops
• Personalized recommendations — optimize your AI workflow
• Weekly coaching reports — narrative analysis vs. your historical baseline

Dashboards

• Local dashboard — live session view at localhost:7777, no account required
• 8 pre-built New Relic dashboards — deploy in seconds _(New Relic mode)_:
• Overview — session stats, cost summary, top tools
• Personal — 30-day self-reflection scoped to you
• Session Detail — deep-dive into a single session's tool calls
• Team View — aggregated cost and efficiency across developers
• Manager View — high-level team metrics, no tool-call content
• Adoption & Cost — adoption momentum, spend, skill and MCP usage, and per-developer outcomes for engineering managers
• Platform Comparison — Claude Code vs. Cursor vs. Windsurf, etc.
• Security Audit — audit trail of sensitive file access

Quick Start

1. Install

npm install -g @newrelic/preflight

Using [Smithery](https://smithery.ai)? Installing Preflight from the Smithery MCP registry wires up the MCP server for you, but Smithery has no mechanism to write Claude Code hooks. After install, either run preflight install in a terminal, or set "allowHeadlessInstall": true in your Preflight config (or NEW_RELIC_AI_ALLOW_HEADLESS_INSTALL=true) and ask Claude Code to call the nr_observe_install_hooks MCP tool. The tool refuses by default, because any connected MCP client can call it. With the flag on, it offers to do this automatically once nr_observe_health reports setup_required: true. Then restart Claude Code to activate monitoring.

2. Run setup

preflight setup

The wizard defaults to local mode — press Enter through the prompts and you're set. It wires Preflight into your AI tool (hooks + MCP server) and writes config to ~/.newrelic-preflight/. Takes under a minute, no account required.

Using GitHub Copilot? The wizard also asks to install Copilot hooks — saying yes configures both the Copilot CLI and VS Code Copilot Chat automatically (hooks, MCP registration, and the fix for VS Code's hook double-capture), so Copilot gets the same tool-call and cost metrics Claude Code does. Run it standalone anytime with preflight install --copilot. See docs/ADAPTERS.md for details.

When prompted, pick a mode:

3. Start coding

Restart your AI tool — hooks and the MCP server load at session start. Every tool call is captured automatically. Open http://localhost:7777 to watch your session live.

Using Kiro? Add Preflight as a Kiro Power for the

nr_observe_* query tools — see the doc for install steps and how to add

automatic tool-call capture on top.

Using Claude Code? You can skip the npm install above and add Preflight as a Claude Code plugin instead:

`

/plugin marketplace add newrelic-experimental/preflight

/plugin install newrelic-preflight@newrelic-preflight-marketplace

`

Other ways to install

Cursor — Add to Cursor installs the MCP server via a one-click deeplink; run preflight setup afterwards for hook capture.

Homebrew (macOS):

brew tap newrelic-experimental/preflight

brew trust newrelic-experimental/preflight brew install preflight

Run preflight setup afterwards for hook capture.

Works With

Claude Code • Amazon Kiro • Amazon Q Developer CLI • Factory Droid • OpenAI Codex • opencode • Kilo Code • Pi • GitHub Copilot • GitHub Copilot SDK • GitHub Copilot app • Google Gemini CLI • Cursor • Windsurf • Google Antigravity • Zed • Continue.dev • Cline

Coverage isn't uniform — some platforms capture every built-in tool call, others (Zed, Continue.dev, Cline) only see calls routed to Preflight's own MCP tools. See ADAPTERS.md for what each platform can and can't observe, and per-platform setup steps.

Connect New Relic (optional)

Local mode is fully featured on its own. Connect a New Relic account to unlock:

• Team & manager dashboards across multiple developers
• Alerting on cost spikes, low efficiency, and stuck loops
• Cross-session history, trends, and weekly coaching reports

Re-run preflight setup and choose cloud or both, or configure it non-interactively:

preflight install \

--mode cloud \ --license-key YOUR_LICENSE_KEY \ --account-id YOUR_ACCOUNT_ID

EU accounts add --eu. FedRAMP accounts add --fedramp. Japan accounts add --jp.

A key passed with --license-key is visible to every local user through ps. For scripts, read it from a file instead (--license-key-file ~/.nr-key, which warns if other users can read the file) or set NEW_RELIC_LICENSE_KEY and NEW_RELIC_ACCOUNT_ID in the environment. The flag wins over the file, and the file wins over the environment. The environment is used only when config.json has no stored licenseKey or licenseKeyCommand, so a shell that already exports these variables for an APM agent never overwrites your stored key. To replace a stored key, pass --license-key-file. With --copilot, the key is still handed to copilot mcp add as an argument, so it is briefly visible in ps for that one call.

Then deploy the prebuilt dashboards:

NEW_RELIC_API_KEY=NRAK-... NEW_RELIC_ACCOUNT_ID=12345 \

preflight deploy-dashboards --all

You'll need a license key (telemetry ingest) and your account ID, plus a user API key (NRAK-…) to deploy dashboards and alerts. See ADVANCED.md for alerts, OTLP export to other backends, and Terraform.

No dashboard until you run `deploy-dashboards`. Cloud mode ships telemetry to New Relic as soon as it's configured, but nothing creates a dashboard automatically — that's the separate step above, and it needs a different credential (a user API key, not your license key). Until you run it, there's no UI to look at. In the meantime, query the raw events directly in New Relic's Query Builder, e.g.:

`sql

SELECT * FROM AiToolCall SINCE 1 hour ago

`

Once deploy-dashboards succeeds, find the dashboards under your New Relic account's Dashboards section.

Data ingest note: Telemetry sent to New Relic counts against your account's data ingest. On paid plans, standard ingest rates apply. Monitor your usage under NR One → Data Management → Data Ingestion.

Requirements

Required

• Node.js v22 or higher (get it or use nvm)
• An AI coding tool (Claude Code recommended for deepest integration)

Optional

• New Relic account — only for cloud/both mode. Skip it to run local-only (the default).
• User API key (NRAK-…) — only needed to deploy dashboards and alerts

Other Commands

preflight doctor               # Run 10 diagnostic checks and print actionable fix commands

preflight validate # Check config for syntax errors and unknown keys preflight update # Pull latest version, rebuild, and offer to restart a running dashboard (source installs only — npm installs: npm install -g @newrelic/preflight@latest; Homebrew installs: brew upgrade preflight) preflight local # List running --local dashboard processes and live --stdio MCP processes preflight local --clean # Kill orphaned --local processes and --stdio processes with a missing binary (prompts for confirmation) preflight uninstall # Remove hooks and MCP config (prompts with a summary first) preflight uninstall --yes # Skip the confirmation prompt (for scripts and CI) preflight uninstall --daemon # Remove only the background dashboard daemon

Add --project to install/uninstall to scope changes to the current directory only.

WSL users: preflight setup will ask which Claude Code you're running. You can also set it explicitly:

• --windows-cc — Windows Claude Code (the desktop app); uses wsl.exe hooks and Windows paths
• --linux-cc — Linux Claude Code installed via npm inside WSL

Documentation

• ADVANCED.md — Configuration, dashboards, alerts, Terraform
• SCORECARDS.md — New Relic Scorecard rules for team attribution
• ARCHITECTURE.md — Data flow, component reference, and operating modes
• ADAPTERS.md — Per-platform integration mechanism, setup steps, and known gaps
• [**FLEET_