Product

Agent Configurator

Author your AI coding-agent configuration once, deploy it everywhere, and see exactly what your agent sessions did — and what they cost. All on your Mac, all in your own files.

Request beta access Download the Helper (1.0.3)

Agent Configurator is currently in beta testing through TestFlight. The Helper is an optional companion described below.

Author once, deploy everywhere

Keep your rules, skills, commands, context files and subagent definitions in one library instead of scattered across a dozen project folders. Write an item once, then deploy it wherever it is needed — into individual projects or into your global configuration.

Agent Configurator understands the layout each tool expects and writes each file to the right place, in the right format, for every runtime it supports: Claude Code, Codex CLI, Cursor, Gemini CLI and Grok CLI. Group related items into Profiles and deploy them together — a “Swift project” profile, a “web frontend” profile — so setting up a new repository takes one action instead of twenty.

Your library rules · skills · commands · context · agents · profiles

Claude Code
Codex CLI
Cursor
Gemini CLI
Grok CLI

One item, authored once, lands in the right place and the right format for every runtime you deploy it to — in a project, or in your global configuration.

The Global configuration screen listing the same two items — a commit rule and a no-commit-without-approval rule — filed separately under Cursor and under Claude, each with its own date. — The same rule, deployed into two different runtimes and filed where each one expects it.

Bring in what already exists

Import configuration you have already written straight from your projects and home directory into the library. Sync from remote sources — Git-backed blueprint repositories, skills.sh and Context7 — with clear indicators showing which items are new, which have updates available and which are already current. Import individually or in bulk.

The Agent Configurator window showing a remote source, skills.sh, with a list of available skill entries and their file counts. Each entry shows whether it is already imported or available to download. — Remote sources list what is available, what is new, and what you already have.

See what your agents are doing

Agent Configurator reads the session transcripts Claude Code and Codex CLI already write to your Mac and turns them into something you can actually read:

Sessions — every session with its project, message counts, and activity over time
Search — full-text search across your entire transcript history
Tools — which tools your agents reach for most, and how often
Costs — token usage and estimated spend, broken down by model, runtime and day

The Activity screen's Tools tab, showing total tool calls, files touched, average calls per session, a bar chart of usage over fourteen days, and a ranked distribution of which tools were used most. — Activity → Tools: which tools your agents reach for, and how usage trends over time.

The Costs tab showing conversation tokens and estimated spend for today, this week, this month and all time, a cache read and write summary, and a model mix bar chart broken down by model. — Activity → Costs: token usage and estimated spend, by day and by model, across every runtime.

Know your configuration

See the runtime configuration your agents are actually using — memory files, hooks, MCP servers and permission rules — in a readable view instead of hand-parsed JSON. Browse the Git repositories across your workspace with branch and status detail, and check workspace health across every registered project.

A project's effective configuration, showing 315 lines loaded at launch across 3 files, and a list of the instruction files that load every session with each one's line count and where it was inherited from. — The configuration your agent actually loads, assembled from every file that contributes to it — with the line count each one costs you at launch.

A project's Health tab reporting two issues — a stale feature branch 82 days old with a ready-to-copy git command to delete it, and one uncommitted file — alongside lint, worktree, environment and dependency checks. — Health checks per project, each with the command to fix it ready to copy.

Your files, your machine

Your library is a plain folder of Markdown files with YAML frontmatter. No proprietary database and no lock-in — readable, diffable and version-controllable with the tools you already use. Store it locally, in iCloud Drive, or in any folder you choose, including a Dropbox or network location.

Your library, your projects and your session transcripts are read locally and stay on your Mac. Nothing is collected and no analytics are sent to us. The features that do reach the network only run when you invoke them: syncing a remote source you have added, searching Context7, and the optional draft assistant, which sends your draft to the AI provider you choose using your own API key, stored in the system Keychain. See the privacy policy for the full account.

The optional Helper

Agent Configurator is sandboxed, which means some features are outside what the app itself is permitted to do: enumerating running processes with their memory use, ending a stuck one, listing the ports your own processes have open, running the Claude Code or Codex command-line tools you already have installed to generate text, and letting the AI agents you connect read and update your Library over MCP.

Agent Configurator Helper is a small, separately downloaded companion that does exactly those things, and only when asked. It is signed with our Developer ID and notarized by Apple, installs as a login item, and talks to the app over a private XPC connection that verifies both sides' code signatures. The Helper makes no network connections of its own; when it runs a command-line tool for you, that tool talks to its own service under your own sign-in. The app works fully without it; features that need it say so, and offer to install it.

After installing, allow it in System Settings → General → Login Items & Extensions. If Agent Configurator ever reports that the Helper isn't responding, use Settings → Text Gen → Repair Helper.

Download Agent Configurator Helper 1.0.3

Built for macOS

A genuinely native app: full light and dark appearance, complete keyboard and menu-bar coverage, VoiceOver support throughout, and the standard macOS controls you already know. Sandboxed, with access only to the folders you explicitly grant.

Requirements

System: macOS 26 or later
Hardware: Apple Silicon Mac
Works with: Claude Code, Codex CLI, Cursor, Gemini CLI, Grok CLI
Imports from: Git blueprint repositories, skills.sh, Context7
Availability: Beta, via TestFlight — ask us for an invitation
Support: Contact page