agent-console

Agent Console by LockedIn Labs is an open-source, local-first observability console for AI coding agents. It reads Claude Code and Codex transcripts already on your machine to show token usage, cache reads and writes, model usage, and list-price cost estimates. It runs locally with no account, telemetry, update checks, or crash reporting, and opens in your browser at http://127.0.0.1:6787. Key features include a multi-machine team view through a self-hosted hub, where each machine keeps its own data lanes and nothing is counted twice. A presenting mode replaces project, machine, and person names with stand-ins for safe screen sharing. Optional interoperability supports Claude Code OpenTelemetry and Kong or LiteLLM token metrics, plus a Prometheus /metrics endpoint and a Grafana dashboard. Live alerts flag repeated tool calls, spending spikes, and stalled sessions. The Projects view correlates token spend with local Git activity such as commits and integrations. The tool is privacy-focused: only model IDs, minute timestamps, token counts, and salted hashes leave a machine, never prompts, replies, file paths, or file contents. Releases include Apple-signed and notarized macOS builds, SHA256SUMS, signed build attestations, and a CycloneDX SBOM. It requires Node.js 22 or newer, or standalone executables are available for macOS, Linux, and Windows. The project is MIT licensed and includes a dependency-free analysis core exported for other Node.js consumers.