# Introduction

Relayloop is the system of record for how your team works with AI agents — searchable history, token spend, and patterns that compound.

Rendered page: https://agentrelay.com/docs/loop/introduction
Markdown endpoint: https://agentrelay.com/docs/loop/markdown/introduction.md

---

Relayloop captures every conversation you and your team have with AI coding agents — Claude Code, Codex, Cursor, Agent Relay, and more — into one searchable record, and turns that record into leverage. It is local-first: a single Rust-backed CLI (`ai-hist`) syncs your agent history into a local SQLite database with full-text search, on your machine, with no account required. The hosted layer, **Relayloop Cloud**, is opt-in.

The local engine binary is `ai-hist`. Wherever you see that command in these docs, it is the real CLI you install and run.

## The problem

Three things go wrong once a team starts leaning on AI agents, and they get worse with scale.

- **AI leverage is bimodal.** Your strongest engineers pull 5–10× out of coding agents; most of the team gets 1×. The gap is prompting skill — tribal, undocumented, and gone the day someone leaves.
- **Spend is invisible.** Real money flows to tokens every month with almost no visibility into where it goes, which model burned it, or which loop ran away.
- **Nothing compounds.** Every session — the prompts that worked, the dead ends, the decisions — evaporates when the terminal closes. There is no shared playbook and no improvement loop.

## The three lenses

Relayloop looks at your agent history through three lenses.

- **Recall**: A searchable history of how your team prompts. Your best engineers' patterns become a living playbook everyone can draw from — find the prompt, the session, and the resume command in seconds.
  - **Burn**: Token spend by model, tool, project, and engineer. Find the waste, forecast the cost, and kill runaway loops before they bill you twice. Cost attribution is captured in [Relayloop Cloud](/docs/loop/cloud).
  - **Insights (coming soon)**: Patterns become proactive guidance and repo rules, so your baseline rises every week and your codebase gets steadily more agent-ready.

## Local-first, cloud opt-in

The CLI works completely offline. `ai-hist sync` reads your local agent files, dedupes, and indexes everything into a SQLite database at `~/.local/share/ai-hist/ai-history.db`. Nothing leaves your machine unless you choose to turn on cloud sync.

When you want a durable off-machine backup and team features, you opt into [Relayloop Cloud](/docs/loop/cloud) with `ai-hist login` and `ai-hist push`. The local CLI does not change and keeps working offline either way.

> Relayloop indexes Agent Relay history as one of its sources. If you coordinate agents with [Agent Relay](/docs/introduction), those conversations land in the same searchable corpus as your Claude Code and Codex sessions.

## Start here

- [Install](https://agentrelay.com/docs/loop/install): Install the `ai-hist` CLI in one command and set up your PATH.
  - [Search](https://agentrelay.com/docs/loop/search): Full-text search across every agent session you have ever run.
  - [Cloud](https://agentrelay.com/docs/loop/cloud): Push for a durable off-machine backup and team features like `pair check`.

[Go from zero to a searchable agent history in five minutes.](https://agentrelay.com/docs/loop/quickstart)