Quickai

quickai

**Profiler for Claude Code.** See exactly where your tokens, money and *time* go — per task, per subagent, per model — straight from the transcripts Claude Code already writes to disk. No instrumentation, no telemetry, nothing leaves your machine.


Most Claude Code users are on a **flat subscription**, so quickai leads with **tokens** (and time), not dollars. The `$` figures are an **API-equivalent** (what those tokens would cost pay-as-you-go) — a measure of *volume and ROI*, not a bill.

• Where the money/tokens went — top tasks, projects, models, agent types
• Where the time went — wall vs idle, per-task timeline, "why was the harness slow?"
• Behavior profiling — tool usage & error rates, cache-hit health, spinning tasks, waste
• Per-project reports and a self-contained HTML report you open in the browser
• MCP server — ask all of the above straight from a chat

---

1. Installation

Homebrew (macOS / Linux)

brew install AlexGladkov/tap/quickai

As a Claude Code plugin (adds the MCP server, ask from a chat)

/plugin marketplace add AlexGladkov/claude
/plugin install quickai@alexgladkov

The plugin registers the `quickai` MCP server, so you can just say *"profile the my-app project"* in a chat. It calls the `quickai` binary — install it first (Homebrew above, or `cargo install --git https://github.com/AlexGladkov/quickai`) so it's on your `PATH`.

Prebuilt for Apple Silicon (installs in ~1s); Intel-mac / Linux build from source (needs Rust).

From source

Requires **Rust** (1.75+). SQLite is bundled — no system deps.

git clone https://github.com/AlexGladkov/quickai.git
cd quickai
cargo build --release
# binary: ./target/release/quickai   (optionally: cp it into ~/.local/bin)

First run

Build the index from `~/.claude/projects/**/*.jsonl`:

quickai index

The index lives at `~/.claude/quickai.db` (pure derivative of the transcript