AgentGlass Review 2026: The Open-Source Cockpit That Brings Every AI Coding Agent Under One Roof

Subtitle: I tested AgentGlass — an open-source observability dashboard that connects Claude Code, Codex, Gemini CLI, and more into one unified workspace with diff viewer, git, Docker, and terminal.

If you're using multiple AI coding agents across different projects, you know the pain: Claude Code in one terminal, Codex in another, Gemini CLI somewhere else — each with its own session, its own cost tracking, its own context window you can't see.

Enter AgentGlass — an open-source observability cockpit launched in mid-July 2026 that brings every AI coding agent on your machine onto one screen.

I've been running it for a week. Here's the honest breakdown.

What Is AgentGlass?

AgentGlass (stylized agentglass) is a self-hosted, MIT-licensed dashboard that:

Connects via Claude Code hooks or OpenTelemetry GenAI exporters for Codex, Gemini CLI, Bedrock, LangChain, LiteLLM
Scans ~/.claude/projects transcripts so historical sessions appear instantly
Persists everything across reloads — not a "live only" dashboard
Filters by provider, project, or scope

The core promise: stop context-switching between agent terminals. Point AgentGlass at all of them and watch live cost, tokens, tool calls, errors, and session lifecycles from one cockpit.

The Six-Panel Workspace

What surprised me most is that AgentGlass is more than a dashboard — it's a full development workspace:

Diff & Review — Shiki-syntax-highlighted diff viewer with word-level intra-line diffs, split/unified modes, and a one-click "Explain" feature that walks you through changes using a local Claude session.
Source Control — lazygit-style git client: stage, hunk-stage, commit, branch, stash, push/pull — all on any repo the fleet has touched.
Pull Requests — Review PRs without leaving the cockpit: overview, conversation, commits, files, checks, and review composer with merge actions.
Docker — lazydocker-style container management with live CPU/memory stats, colored log viewer, and exec-into-container.
Real Terminal — A genuine PTY shell per repo over WebSocket. Persistent sessions survive panel closes.
Claude Chat — Drive local Claude Code sessions from the browser. Sessions started here appear in the fleet like any other agent.

The Phone Companion

AgentGlass ships a separate mobile interface designed for triage. The home screen is a queue of decision cards:

🔴 Blocked · waiting on you — approve or deny a dangerous command
🟡 Container down — tail logs or restart
🔵 CI went red — open log or re-run
🟢 Ready to merge — approved, checks green
🟣 Review requested — opens PR with diff

It connects via QR code on your LAN. Three tabs: Now (queue), Chats (persistent conversations), and Repos (changes, PRs, containers).

Pricing

AgentGlass is completely free and open source under the MIT license. Self-hosted on your machine — no cloud dependency, no API keys needed beyond what your agents already use.

Who Should Use It?

You'll love AgentGlass if:

You run 2+ AI coding agents regularly
You want to track token/cost usage across providers
You hate context-switching between terminal windows
You want a single pane for git, Docker, PRs, and agent monitoring

Skip it if:

You use one agent in one project
You prefer lightweight CLI tools
You're not comfortable with self-hosted Bun+SQLite setup

Verdict

AgentGlass fills a real gap in the AI coding toolchain. As we move from single-agent to multi-agent workflows, having a unified observability layer isn't a luxury — it's a necessity. At 273+ GitHub stars and growing fast, this is one to watch.

Rating: 8.2/10 — A genuinely useful tool with room to grow.

Read the full detailed review at toolsdepth.com/reviews/agentglass-review-2026