Why We Built Styrby

Styrby exists because we needed it and nobody was building it. The problem was specific: developers using multiple AI coding agents had no unified way to track costs, approve permissions, or review sessions across agents. This article explains the gap we found and the tool we built to fill it.

The Problem

By early 2026, most professional developers were using at least one AI coding agent. Many were using two or three. Claude Code for deep reasoning work. Codex for fast boilerplate. Gemini CLI for research. Aider for legacy codebases. Each agent runs in its own terminal with its own interface.

Three specific problems emerged from this multi-agent workflow.

No Unified Cost Visibility

Each agent bills through its own provider. To know your total AI spend, you check the Anthropic billing page, the OpenAI billing page, and the Google Cloud billing page. Then you add them up. If you want per-client cost attribution, you track sessions manually.

This is workable for a solo developer spending $50/month. It breaks down for teams spending $1,000/month across multiple agents and projects.

No Remote Permission Control

AI agents ask permission before running commands. If you step away from your terminal, the agent blocks on a pending approval. Walk to the kitchen, come back ten minutes later, and the agent has been idle the entire time.

Worse: if you run an agent overnight, there is no way to approve permissions without being at your terminal. The session either runs with all permissions auto-approved (risky) or blocks on the first permission request (wasteful).

No Session Management Across Agents

Each agent maintains its own session history. Reviewing what happened across a day of mixed agent usage means checking each agent separately. There is no unified timeline, no cross-agent search, and no way to compare costs between agents for similar tasks.

The Market Gap

We looked at existing solutions. The agent providers focus on their own agent experience. Anthropic built Claude Code Channels for remote Claude access. OpenAI has its own monitoring. Google has Cloud monitoring. None of them address the multi-agent case because none of them have incentive to. Each vendor wants you using their agent exclusively.

Developer tool companies were building AI features into existing products (VS Code extensions, IDE plugins), not standalone management tools for multi-agent workflows.

The gap was clear: a cross-agent management layer providing cost tracking, remote permissions, and session management regardless of which agents you use.

What We Built

Styrby has three components:

CLI. Runs alongside your AI agents on your workstation. Captures session data, intercepts permission requests, tracks token costs. Encrypts everything with TweetNaCl before sending to the server.
Mobile app. iOS app (built with Expo) for remote monitoring, permission approval, and session review. Push notifications for important events.
Web dashboard. Full session management, cost analytics, and budget configuration.

The architecture is zero-knowledge: the server stores encrypted session data and cost metadata. It never sees your source code or agent conversations.

What We Deliberately Did Not Build

Another AI agent. The world has enough AI coding agents. We build the management layer, not the agents themselves.
An IDE integration. IDE plugins compete with existing tools. Styrby operates at the terminal/system level, which is where agents run.
A replacement for provider dashboards. If you want detailed Anthropic usage analytics, use Anthropic's dashboard. Styrby provides cross-agent aggregation, not deep single-provider analytics.

Where We Are Now

Styrby supports five agents: Claude Code, Codex, Gemini CLI, OpenCode, and Aider. The CLI is in public beta. The iOS app is in development. The web dashboard is live. We are a small team, building carefully, and shipping features when they are ready.

If you use multiple AI coding agents and want better visibility into costs and permissions, Styrby is what we built for you.