Codex Usage Dashboard: What OpenTelemetry Adds

Which Codex usage dashboard should you use?

Free tracking does not mean free Codex usage. Self-hosted monitoring also needs storage, infrastructure, and maintenance. For hosted cost tracking, see Ganivra’s current plans.

How do I check my Codex usage and remaining limits?

Open the Codex usage dashboard. During an active Codex CLI session, run:

/status

OpenAI documents these as the places to check current and remaining limits. This is the quickest starting point when you want to know how much capacity you have left. Official Codex usage guidance.

Where does OpenAI’s Codex dashboard stop?

OpenAI provides more than a personal usage meter. Its organization analytics supports interactive Codex adoption and activity reporting. The Analytics API supports automated aggregate reports and joins with internal organizational data. For auditable records, OpenAI points to the Compliance API. OpenAI’s reporting surfaces.

The boundary is the reporting contract. An interactive dashboard is not a stable integration schema, and an aggregate report is not a request-by-request event stream. Use the Analytics API when its supported fields answer the question; check workspace eligibility and permissions before building around it.

For example, a weekly adoption report may answer how broadly Codex is used. Investigating which tool failures accompanied a particular session, or allocating usage to your client codes, needs additional event detail or business mappings. OpenTelemetry supplies supported events; your reporting pipeline supplies those mappings.

What does Codex OpenTelemetry add?

Codex supports opt-in OpenTelemetry export. Its documented events cover conversations, API requests, stream activity, prompt metadata, approval decisions, and tool results. Export is off by default. Codex monitoring and telemetry documentation.

• Request monitoring: inspect attempts, duration, status, and errors.
• Token reporting: read usage from completed-response events that carry token counts.
• Tool investigation: correlate tool duration, success, and approval decisions where the events provide that context.
• Custom attribution: join collected usage to your developer, team, project, and client mappings.
• Shared reporting: normalize supported Codex and Claude Code usage in the same reporting system.

Do not count every stream event as another model response. In the documented SSE events, token counts appear on response.completed. Validate event coverage with your installed client and deduplicate the records you ingest.

How do I enable Codex OpenTelemetry?

Merge an exporter configuration into your existing ~/.codex/config.toml. This example uses a placeholder collector endpoint:

[otel]
environment = "production"
log_user_prompt = false

[otel.exporter.otlp-http]
endpoint = "https://otel.example.com/v1/logs"
protocol = "binary"

Replace the endpoint with your collector’s URL and configure the authentication it requires. Preserve existing settings and avoid duplicate TOML tables. Start a new session, run a small task, and check that your collector receives the expected events. Codex batches telemetry and flushes on shutdown; collection is not guaranteed to be instantaneous.

Keep log_user_prompt = false to redact prompt text. Tool-result events can still contain output snippets, so inspect exported fields and apply collector redaction. That setting alone is not a guarantee that telemetry contains no source code or sensitive data. Exporter and privacy guidance.

How do I track Codex usage by developer and project?

Enabling an exporter gets events into a system. Reliable attribution takes three more steps:

1. Map identities. Associate available identity fields with a stable developer record. Keep missing identities visible as unassigned.
2. Define project ownership. Decide whether a project means a repository, client engagement, or cost center. Use consistent labels across machines.
3. Verify coverage. Compare a known session with the imported report, including model, token counts, developer, and project. Track devices or sessions that have not exported data.

Ganivra’s CLI setup uses a shell wrapper to apply project labels. For desktop sessions without project attribution, assign the project in the Sessions view. This is Ganivra’s integration behavior, not a promise that every Codex client automatically reports repository ownership. See the Codex usage and cost tracking setup.

Does a Codex cost dashboard show my actual bill?

Only if the report explicitly uses billing data. A token-based estimate answers what recorded usage is worth at the chosen rates. Subscription fees, credits, discounts, and billing arrangements can produce a different amount paid.

Keep three columns: recorded usage, estimated usage cost, and actual billed spend. Do not add estimated usage value to a subscription fee as if both were separate charges. If you allocate a shared bill to projects, document the allocation method and retain an unassigned category.

For example, a team could allocate a subscription bill using each project’s share of estimated usage value. That is an internal allocation policy, not a provider-issued project bill. Our AI coding chargeback guide explains how to reconcile allocations.

When should I add a shared Codex usage monitor?

Use native usage checks for remaining capacity and native analytics for supported adoption reports. Add centralized telemetry when you need operational investigation, custom project attribution, or a consistent view across coding tools.

AI coding cost tracking with Ganivra brings supported Codex and Claude Code usage into developer, team, project, and client views. Compare the Claude Code dashboard options if your team uses both, or model seat costs with the AI coding tools cost calculator.