CodeBurn (MIT-licensed) tracks AI coding spend across 31 tools by parsing local transcripts. Orq already stores per-message token usage for its own sessions. Marrying the two gives Orq fleet-wide cost analytics with CodeBurn’s visual language. Below: the proposed architecture and ten clickable previews of where it could live in the Orq UI. All numbers are mocked; every preview supports light and dark themes (◐).
Each server runs codeburn --json on a 5-minute cron (Orq job runner).
A small ingest endpoint merges the rollups into Orq’s existing token_usage
aggregation path, keyed by session/project/model/day. Orq renders everything with its own
components. CodeBurn stays an unmodified upstream dependency — updates are free.
Vendor CodeBurn’s parsers (the RFC’d @codeburn/core) into Orq’s backend and
read transcript files directly. Tighter coupling, no subprocess, but Orq inherits maintenance
of 31 parsers. Revisit if/when the core library ships as a package.
Run CodeBurn’s MCP server per host so agents can query their own spend mid-session (“how much has this task cost so far?”). Complements A — it serves agents, not dashboards.
Why A: Orq’s pricing engine already normalizes ~40 usage-field variants for its own sessions, but CodeBurn covers editors Orq can’t see (Cursor, Codex CLI, Gemini CLI, …) and ships battle-tested pricing tables. The collector gets that coverage with zero fork risk. License: MIT — attribution kept in THIRD_PARTY_NOTICES.
Mockups use Orq’s “Editorial Intelligence” tokens (warm paper / olive accent light · teal accent dark) with a CodeBurn-inspired chart palette validated for colorblind safety in both themes against Orq’s real card surfaces. Every chart has hover tooltips, legends, and a “view as table” twin. Data is fabricated for the last 30 days ending Aug 10, 2026.