TokenOps – AI Coding Observability
PythonFastAPISQLiteReactASTGemini APICLI
TokenOps is a local-first AI engineering observability platform. It ingests Claude Code, Codex, and OpenCode session logs, groups captured activity into inspectable work units, and relates token usage to code-graph deltas, validation evidence, commits, cost, and rework — without manual task tracking. Scores are evidence-weighted signals with confidence labels, never a universal productivity score.
Key Features
Multi-provider ingestion: Claude Code JSONL, Codex sessions, OpenCode SQLite
Python AST code-graph snapshots with baseline-aware work attribution
Evidence-weighted session scoring with semantic churn and confidence labels
Local FastAPI + React dashboard with model efficiency comparison
Privacy modes, retention purge, and redacted public profile export
Challenges
Attributing code changes to AI sessions without manual tracking
Designing scoring math that stays honest about uncertainty
Unifying three different agent log formats into one work-unit model
Key Learnings
AST parsing and code-graph diffing for change attribution
Evidence-weighted metrics design over vanity productivity scores
Local-first architecture with privacy-preserving exports
Duration
Ongoing
Role
Solo Developer
Status
in-progress