AgentBar
ProductDocsBlog

Blog

Notes on building AgentBar

What we learned from watching agents watch themselves: measurement without spawning, trace reconstruction from logs, and the incident that started the project.

  • Aug 1, 2026 · 6 min read · By tsuvic

    Why AgentBar reads files instead of spawning your agents

    CodexBar froze a Mac every minute because it spawned Claude CLI to read usage. AgentBar reads the files agents already write, and tests prove it spawns nothing.

    AI coding agentsmacOS toolsProcess spawns

  • Aug 1, 2026 · 8 min read · By tsuvic

    How AgentBar reconstructs multi-agent traces from local logs

    Three coding agents, one request, one calling the next. AgentBar rebuilds the call graph and state from files the agents already write, and labels every edge it infers.

    AI coding agentsDistributed tracingmacOS tools

RSS: feed.xml

Get AgentBar

See how lightly it runs on your own Mac.

AgentBar runs entirely on files your agents already write to disk. It spawns no CLI, MCP server, browser, or agent to get usage or history. The Diag tab reports AgentBar's own CPU, memory, and spawned process count — always 0. macOS 14+ · Swift 6 / SwiftUI · source on GitHub.

GitHub repositoryInstall instructions
git clone https://github.com/tsuvic/agentbar