TraceLens

Offline Codex analytics with traceable usage, validation evidence, and project memory for the next session.

WHEN

2026

CONTRIBUTION

Design & development

TECHNOLOGIES

PythonSQLiteHTMLJavaScriptCodex

LOCAL ANALYTICS & MEMORY02

SELECTED CODEX SESSION LOGSEvidence into context

Python · local SQLite metadata

Session evidence → offline dashboard
Reviewed context → CHATGPT.md

UsageValidationReworkMemory

Conceptual local evidence workflow

THE PROJECT AT A GLANCE

Understand the work. Carry the context forward.

Explicitly selected Codex logs become local metadata, an interactive offline dashboard, and optional reviewed project memory.

Runtime dependencies: 0
Self-contained dashboard file: 1
Per-project memory: CHATGPT.md

View repository

Private repository · GitHub access is required to view the source.

01 / OBJECTIVE

What the project set out to do

Make Codex work easier to inspect across sessions: explore usage and repeated work, distinguish conversation completion from validation evidence, and retain useful project context locally.

02 / MY CONTRIBUTION

My part in the work

I developed TraceLens as a Python analytics engine with a self-contained HTML dashboard and portable Codex plugin. It normalizes selected session logs into SQLite metadata, generates evidence-backed reports, and optionally maintains per-project CHATGPT.md memory files.

The engineering focus is evidence and continuity: link findings to recorded events, preserve reviewed context through refreshes, and report the limits of observed results and estimated measurements.

03 / TECHNICAL APPROACH

How it came together

Normalize explicitly selected session logs

Read bounded Codex JSONL records, normalize supported events, deduplicate observations, and update existing session snapshots in a local SQLite store. Transcript commands are analyzed as data rather than executed.

Make usage and working patterns explorable

Embed data and assets in one offline HTML dashboard with project search, model and date filters, token charts, tables, heatmaps, and session timelines. Provide keyboard-focus and tap readouts alongside chart views.

Keep outcomes tied to captured evidence

Track chat and validation states separately, with supported tool results, check attempts, file references, commit evidence, and parsing warnings. Rework and loop findings link back to their supporting records.

Carry reviewed project context forward

Generate stable per-project memory files with recorded activity and optional reviewed goals, decisions, completed work, and next steps. Preserve personal notes outside managed blocks and retain reviewed context on activity-only refreshes.

04 / OUTCOMES

What the work produced

Built a local workflow from selected Codex transcripts to SQLite metadata and interactive reports.
Connected usage, validation, and repeated-work findings to supporting session evidence.
Added project browsing and reviewed memory that can be refreshed without replacing personal notes.

The reviewed README describes release candidate 0.2.0rc5. API-equivalent cost is a lower-bound estimate rather than a bill, active time is estimated, and rework indicators are heuristics. The runtime makes no network requests. Browser painting and installed-host plugin integration are listed as unverified in that release.

UP NEXT / PROJECT 03

Wakeline

View case study

Want to talk about this project?

Get in touch