Breadcrumbs

See where your coding-agent tokens actually go. Breadcrumbs is a Visual Studio Code extension that turns the GitHub Copilot, OpenAI Codex, and Anthropic Claude Code logs already on your disk into usage dashboards and per-chat breakdowns -- entirely locally. Your prompts and code never leave the machine.

Why Breadcrumbs

Local only. It reads files already on disk and never uploads anything. The persisted index keeps allowlisted metadata only -- captured prompts and tool output stay out of the cache, and content is suppressed in Restricted Mode.
One place for three agents. Copilot, Codex, and Claude Code side by side, with provider-scoped totals (token semantics differ per provider, so Breadcrumbs never guesses a unified bill).
From the big picture down to one request. Provider overviews, a searchable chat inventory, and a per-chat action tree with per-request and cumulative token development.

Breadcrumbs all-providers overview with daily-usage, per-provider, and workspace charts

Quick Start

Download the latest .vsix from the Releases page.
In Visual Studio Code, run Extensions: Install from VSIX... and select the downloaded file.
Open Breadcrumbs from the Activity Bar.
Run Breadcrumbs: Refresh Usage Index.

Requires Visual Studio Code 1.120.0 or newer. No account, API key, or network access required.

What You Get

Provider overviews

Daily tokens and requests, reasoning- and cache-share trends over time, model and workspace breakdowns, token composition, and index health -- per provider or across all of them.

Claude Code provider overview with token composition and usage charts

Chat details

Open any chat for a token-composition breakdown, per-iteration signals (cache, output, context-window usage), and a collapsible action tree of user messages, model requests, reasoning, tool calls, results, and subagent runs.

Chat detail showing token composition and per-iteration signal charts

Chat detail timeline tree of requests, tool calls, results, and subagents

Feature summary

Provider readiness and source diagnostics, including a guided Copilot telemetry setup.
Provider-specific usage, model, cache, token, and quality overviews with charts.
Chronological chat inventory with provider-maintained or derived titles, plus a searchable Open Chat quick pick.
Per-chat action tree and per-request / cumulative token development.
Incremental local indexing with parser diagnostics and stale-result fallback, so refreshes after the first run are fast.
Structured diagnostic log ("Breadcrumbs Log" in the Output view); verbosity via Developer: Set Log Level....

Contributing

See CONTRIBUTING.md and LICENSE.