lot-map is a free Claude Code plugin that draws a live map of tool calls, subagents, hooks and context usage beside your chat. Here's how to install and use it.
Vamsi Tallapudi
Manager, Architect Technology at Cognizant
lot-map is a free, local-only Claude Code plugin that draws a live map next to your chat window, showing every tool call, model step, hook, skill and subagent as a pulse of light moving from a center hub out to one of ten "stations." Built by a developer who goes by alexeyw, it hit roughly a thousand GitHub stars within about a week of release, which tells you plenty of people are tired of scrolling through raw tool-call logs to figure out what their agent is actually doing.
I installed it on a session running three background subagents and watched the map fill in real time instead of tailing a terminal log. It's a genuinely useful debugging aid, not just a pretty animation. It's also, very clearly, just a pretty animation half the time. Both things are true.
The map has a center circle (labeled CLAUDE by default) and ten stations arranged around it: You, Model, Memory, Files, Shell, Network, Browser, Agents, Skills, and Hooks. Every time Claude Code does something (reads a file, runs a shell command, calls an MCP tool, fires a hook), a pulse travels from the center to the matching station.
A few details make this more than a toy:
The classification logic is more careful than I expected. cat, grep, mv, and sed -i route to Files because they only touch the filesystem, while curl, gh, ssh, git push, and npm install route to Network because they talk to something outside your machine. Docker commands, here-documents, and piped chains all get parsed too, as of the 0.8.0 and 0.10.0 updates.
claude plugin marketplace add alexeyw/lot-map
claude plugin install lot-map@lot-map
If that shorthand doesn't resolve, the two-step version above is the reliable fallback. In the desktop app, it's + → Plugins → Add marketplace, type alexeyw/lot-map, then install. Once it's in, the pane opens on its own at the start of a new session: no slash command needed to turn it on.
Two requirements actually matter here:
/lot-map manually to confirm it's at least tracking in the background.lot-map picks a mode based on where it's running, but you can force one through config:
| Mode | Where it works | What you get |
|---|---|---|
| Desktop map | VS Code extension, desktop app | Full interactive map, zoom controls |
| Terminal picture | kitty, Ghostty | Bitmap rendering, capped at 30 fps |
| Terminal braille | Any terminal with a braille-capable font, including Windows Terminal | Text-rendered map using braille characters |
| Terminal text | Everything else | Plain row list, no visual map at all |
If you're on Ghostty specifically, it's worth pairing this with what we covered in Ghostty 1.4's Claude Code terminal support. Picture mode is the whole point of running lot-map in a real terminal instead of falling back to text rows.
This is the feature that sold me on it. Each active subagent shows up as a small dot circling the Agents station, and that subagent's own tool calls visibly originate from its dot rather than getting lumped in with the main session. If a subagent fails, its dot turns red immediately, so you don't have to scroll back through a transcript to find out which of your five parallel agents choked.
Turn on the subagentNames option (added in 0.8.0) and each subagent's model step shows the first 40 characters of whatever prompt it was launched with, instead of just "subagent." That's a small thing, but when you're running several subagents off something like the Antigravity-style teamwork pattern, being able to tell them apart at a glance saves real time.
Background tasks get the same treatment: a status indicator shows running, completed, or failed, which matters more than it sounds like once you've lost track of a long npm run build running in a detached shell.
Everything's set through plugin config, either interactively or piped in as JSON:
echo '{"language":"es","contextCost":true,"theme":"dark"}' | \
claude plugin configure lot-map@lot-map --values-stdin
| Option | Default | Does |
|---|---|---|
language | en | UI language (7 supported: English, Russian, Spanish, French, German, Italian, Portuguese) |
coreName | CLAUDE | Center label, 10 characters max |
terminalView | text | Forces text / picture / braille mode |
theme | auto | dark / light / auto |
contextCost | false | Shows estimated token cost per call |
layout | you-top | Station arrangement |
motion | full | full / reduced animation |
hookReasons | false | Shows why a hook blocked something |
serverStations | — | Pins specific MCP server tools to a chosen station (new in 0.10.0) |
paneBackground | — | Transparent or custom pane background (new in 0.10.0) |
The contextCost option is the one I'd actually turn on by default. Watching token spend per call in real time is a more honest signal than checking usage after the fact. If you've read our piece on Claude Code usage limits and cost, this is basically a live version of that math running next to your session.
Yes, if you run subagents, background tasks, or MCP servers regularly and want to catch a stuck or looping call before it burns through your context window. The privacy story also checks out: the plugin makes no network requests, doesn't read files beyond tool-call metadata, and doesn't touch the model or the conversation itself, according to its own README and separate PRIVACY.md doc.
No, if you mostly run short, single-session chats where you'd notice a stuck tool call from the transcript anyway. The map adds real visual overhead for something that simple, and on a narrow terminal you're stuck with the text-row fallback, which gives you nothing lot-map's own log view didn't already have.
One thing to watch out for: lot-map's popularity has spawned a pile of near-identical forks on GitHub (markdotdigital, dgfisher2021, Wesley-Gomes, and others), some even ported to OpenCode Desktop. Most look like legitimate republishes or ports, but a plugin runs hooks inside your Claude Code session, so install from alexeyw/lot-map directly rather than a rebrand you found in a random "awesome mods" list. Same caution applies to any Claude Code plugin marketplace entry, honestly. It's the same reasoning behind treating codebase map tools like floorplan as read first, trust later.
A GitHub snapshot from November 14, 2026 put the repo at 736 stars and 85 forks, a few days after its initial push. By November 18 it had crossed roughly 1,000 stars, with version 0.10.0 shipping the same day, adding serverStations, paneBackground, and a guardedHooks option for Team and Enterprise logins. That's a fast climb for a plugin with no corporate backing, and it's now listed as a featured entry in the community-maintained awesome-claude-code-mods repo, which tracks thousands of these.
No. It's a third-party, community-built plugin by a developer named alexeyw, distributed through Claude Code's plugin marketplace system. Anthropic doesn't maintain or endorse it.
No. According to the project's README and PRIVACY.md, it makes no network requests, reads only tool-call metadata already visible inside your session, and keeps everything local to that session.
Claude Code 2.1.321 or later. Earlier builds don't expose the plugin hooks lot-map depends on to track tool calls and context usage.
Your terminal is probably narrower than the 144-column minimum, or it doesn't support picture/braille rendering. Set terminalView to braille for a text-based map that still works in narrow terminals, or just use /lot-map in text mode.
Yes. Each active subagent appears as its own dot circling the Agents station, and calls made by that subagent trace back to its dot. A failed subagent turns its dot red so you can spot it without reading the full transcript.
No, and this is a real limitation. The ring shows how full the context window is as a percentage, not a breakdown of what's consuming it. For that level of detail you still need Claude Code's own context inspection tools.
New AI tools, automation workflows, and course drops — straight to your inbox. Join 2,400+ builders.
Claude Code, Codex CLI, Pi, and OpenCode each win on different models. Runta's eval found harness choice swinging cost up to 4.4x on the same model.
Gemini 4 Neon launches at $3/$15 per million tokens with a 1M context window, slotting between Hydrogen and Carbon. Here's what the pricing and benchmarks actually mean.
MAI Code 2 claims 96% of Claude Opus 5.5's coding performance at 5% of the price in GitHub Copilot. We checked the benchmarks, the promo pricing, and real-world reports.
Comments