relai — see what your AI coding actually costs
One local dashboard for Claude Code, Codex, opencode, and Gemini.
The problem
Developers are using more than one AI coding tool — Claude Code for one task, Codex for another, opencode or Gemini for a third. Every tool keeps its own private record of what happened. When a team tries to answer basic questions — what did we build with AI this month? on which projects? was it worth it? — the data lives in four different places and nobody has a clean answer.
Telemetry reports end up blending real product work with weekend experiments. Signal gets lost inside the noise.
What relai does
relai reads the session files your AI tools already leave on disk — read-only, never writing back — and gives you one unified view. Every session, every project, every token, across Claude Code, Codex, opencode, and Gemini. Full-text search over every transcript, per-message token counts, custom titles instead of raw UUIDs, and an archive for stashing noisy sessions without touching the source files.
Know what it costs
A built-in pricing table covers the Claude 4.x and 3.x families, GPT-5 and GPT-4o variants, o1, and local models (Ollama, Gemma, Qwen, Llama at $0) — with cache reads and cache writes priced separately, the way providers actually bill. Stacked charts break tokens and estimated dollars down by day or hour, by source, category, or model.
Separate real work from experiments
One click — Auto-classify prototypes — and relai flags throwaway sessions automatically. Include or exclude sessions per row, then export telemetry for only the work that actually shipped: CSV or JSON, 19 columns including per-session estimated cost and tool breakdown. Finance sees accurate numbers, engineers keep their experiments private.
Move context between tools
Prototype in a cheaper local model, then hand the session to Claude with one click. relai summarizes the transcript map-reduce style, collects your project memory files, writes a handoff primer, and launches the next tool already knowing where you left off. Any source, any target — or resume the original session right in your terminal.
Session insights
relai can read a session back critically: automated checks flag security vulnerabilities, performance bottlenecks, and architectural oversights the session may have introduced, each with severity, explanation, and suggestions. A Turbo mode runs the same analysis through every AI tool you have installed and ensemble-votes the verdicts — one model's blind spot is another's catch.
Privacy by default
Everything runs on your laptop. relai opens your tools' data read-only and never writes back. Session data never leaves the machine unless you click Export — the only things stored in the cloud are your email and license status. That makes it usable in regulated environments where sending transcripts to a SaaS dashboard isn't an option.
Download
relai ships as a signed, notarized macOS app. No Python, no Terminal, no setup — drag it into Applications and launch. Sign in to start a 7-day free trial; after that it's $9/mo.
Feedback, ideas, and partnerships welcome —reach out.