TokensBurned 0.6.10

TokensBurned 0.6.10 expands usage collection and refreshes the profile-card experience.

• Collect recorded Gemini CLI usage and retained history.
• Capture live per-call GitHub Copilot CLI usage with the optional extension; earlier calls cannot be recovered from ordinary history.
• Read supported OpenCode v1/v2 SQLite and legacy JSON histories, plus supported Cline SDK and IDE records.
• Deduplicate supported records when the CLI and plugins share the same version, connection, and BURN_HOME directory.
• Refresh installation guidance, privacy documentation, translations, and the README card animation.

Keep the CLI and plugins on the same version. After upgrading a background collector, run tokensburned run again to refresh its runtime. Consult the collection guide for supported formats and custom paths.

TokensBurned 0.6.9

Plugin updates now follow published stable releases instead of development code. Plugin catalogs pin both the release tag and commit; development versions skip remote update checks. The background collector and server-controlled upload scheduling from the latest main branch are included.

Existing website and SVG styling are unchanged in this release.

The attached package is available for manual CLI installation. npm registry publication is pending; the GitHub/plugin release is available here.

TokensBurned 0.6.8

Fixes

• A deferred yesterday no longer blocks today. The API closes a past day's write window until the next UTC day but accepts today's data every UTC hour. The client stored that deferral globally and held today's upload back with it; deferrals are now recorded per day.
• Uploads align to the server's UTC hour windows. After a successful upload the client waits for the next hour boundary rather than a fixed 60 minutes, so data reaches the server on average 30 minutes sooner. The waiting worker and tokensburned update use the same view, and update now lists deferred days with their retry time.

🤖 Generated with Claude Code

TokensBurned 0.6.7

Changes

• tokensburned update now catches up and uploads. After the release check it re-merges the last two days of every installed harness, uploads whatever is due (or leaves the single waiting worker behind), and prints the pending state.
• Update notices no longer vanish after the daily check. While the installed version is older than the published one, every SessionStart reminds the assistant to mention the update. Applying it still requires an explicit request.

Notes

• The API had been advertising 0.6.4 as the latest version, so no install ever saw a notice for 0.6.5 or 0.6.6. That is fixed server-side alongside this release.

🤖 Generated with Claude Code

TokensBurned 0.6.6

Fixes

• Nothing is left behind at the end of a session. Stop and SessionEnd now merge the transcript after every turn (the 20-minute throttle from 0.6.5 is gone; even a 15 MB transcript parses in ~55 ms in the detached child). SessionStart re-merges transcripts modified in the last two days, so a session that never ended cleanly is caught up on the next start.
• Uploads no longer wait for the next activity. When days are pending but the hourly upload window is closed, hooks leave behind a single waiting burn upload-worker per machine. It holds a pid lock in ~/.burn, sleeps until the window opens, uploads once, follows server deferrals for at most two hours, and exits. It is the only process TokensBurned leaves running, never more than one, never longer than two hours.
• tokensburned hooks install writes SessionStart, Stop and SessionEnd for CLI-only installs; Cline and backfill reuse the same worker.

Notes

• Update the plugin in Claude Code and Codex to pick up the new Stop hook.
• Server upload spacing is unchanged: at most once per hour, plus any deferral the server requests.

🤖 Generated with Claude Code

TokensBurned 0.6.5

Fixes

• OpenAI was reported as Unknown for the default Codex config (no base_url / model_provider). Codex now defaults to OpenAI, Claude Code to Anthropic, and CLAUDE_CODE_USE_BEDROCK/VERTEX/FOUNDRY map to their platforms. [model_providers.<id>] tables are resolved (base_url and name only). (#23)
• Unknown providers now show a name. Unrecognized gateways keep their endpoint hostname (hostname only, never path, port, or credentials) instead of collapsing to "Custom"; about 30 common gateways (OpenRouter, Azure, Bedrock, xAI, Mistral, Groq, Moonshot, Alibaba, Zhipu, MiniMax, Volcengine, SiliconFlow, ...) get curated labels. (#23)
• Uploads no longer depend on SessionEnd alone. Desktop and long-lived sessions rarely end, which left the outbox untouched for days. The Stop hook now merges the transcript at most every 20 minutes and SessionStart flushes pending days; server uploads stay throttled to once per hour. (#25)

Notes

• Buckets already stored as unknown are not rewritten; run tokensburned backfill to re-tag them.
• A fresh backfill after upgrading also pushes anything the old SessionEnd-only path left behind.

TokensBurned 0.6.0

Hourly local aggregation and one device/day upload row.

• Install the standalone CLI with npm install -g tokensburned.
• Flushes v2 daily envelopes at hourly and session-end boundaries.
• Preserves exact hourly, harness, provider, and model breakdowns.
• Uses lazy one-hour SVG refresh backed by R2.
• Adds one-row warm summaries, bounded historical reads, monthly rollups, and materialized rankings.
• Retires protocol v1 snapshot writes; old clients receive 426 Upgrade Required without a D1 write.

Server support is deployed at 100%. The minimum supported client version is 0.6.0.