lotel
look
Colourwayautodaynight

Still choosing a look for lotel. Your pick sticks on this browser.

Join the waitlist

The UI

lotel serves its UI at http://127.0.0.1:8080. Pages are rendered on the server. A small script adds live updates, charts and faster navigation. The sidebar groups them: the Dashboard, then Agents, then Telemetry.

Most pages share one filter bar. See Filters at the end of this page.

Dashboard

/ gives an overview of the instance:

The dashboard: row counts per signal and the live trace error rate per serviceThe dashboard: row counts per signal and the live trace error rate per service

Agents

Sessions

/sessions lists one row per coding agent session, across Claude Code, Codex and opencode: the agent, event and span counts, cost, tokens in and out, and when it was last seen. Each row links to its logs, and to its traces when it has spans.

Click a session to open its timeline at /sessions?session=<id>. It shows the session's spans as a waterfall and every event in order, with runs of the same event folded into one row. Click a span to see its details.

A session timeline: each model call and tool run as a bar, with its durationA session timeline: each model call and tool run as a bar, with its duration

Spend

/spend shows what your agents cost, by day, model and developer: cost in USD, tokens in, out and from cache, and active time when the agent reports it. Totals for the view sit above the table.

The numbers come from the agents' cost and token metrics. Points that are running totals (cumulative temporality) are not added up, so nothing is counted twice. On days without those metrics, lotel fills in from the agents' api_request events, and the page notes it.

Tools

/tools shows one row per tool your agents called: calls, failures, p50, p95 and max duration, and how many calls you rejected. Each count links to the spans and logs behind it.

Durations come from tool spans and measure the execution, not the time spent waiting for your approval. If the view has no tool spans, the page falls back to the tool-call counters and shows no durations.

The Tools page: calls, failures and latency per toolThe Tools page: calls, failures and latency per tool

Setup

/setup has copy-paste blocks that point an agent at this instance, with the endpoint and token filled in. It has three tabs: Claude Code, Codex as a config file, and Codex for one run. The endpoint uses the host name your browser used to reach the UI. See Quickstart.

Telemetry

Logs

/logs lists recent log records, newest first. Filter by service, severity, time range and attributes. The severity and service facets show counts and narrow the list when you click them. The Fields panel lists the attributes in the current view.

The Logs page: severity and service facets above the matching rowsThe Logs page: severity and service facets above the matching rows

Traces

/traces shows spans. Filter by service, span name, status, minimum duration in milliseconds, time range and attributes. The page shows the overall p95 over time, a span duration distribution, a per-service summary and the recent spans.

Open a trace at /traces?trace=<id> for its waterfall: spans in tree order, placed by start time and duration, colored by service, with errors outlined. Click a span for its attributes. When logs carry the trace or span id, a link shows how many and opens them on /logs.

A trace waterfall across three servicesA trace waterfall across three services

Metrics

/metrics lists every metric by shape, name, unit and service, with its point count, latest time and an example value. Click one to chart it: gauges and sums as time series, histograms as a distribution.

Explore

/query queries the store in two modes:

Results go through the same content masking as the other pages. See Privacy.

Explore: p95 span duration by service, as a chart and a table, with facetsExplore: p95 span duration by service, as a chart and a table, with facets

Services

/services is a service map built from parent and child spans. Each node is a service, and each edge is labeled with its call count. An edge is red when at least 2% of its calls are errors. Click a service to open its spans. The page also shows error-rate and p95 trends.

The service map: services, the calls between them, error rates and latencyThe service map: services, the calls between them, error rates and latency

Storage

/storage shows what is on disk and what forgets next: every table and segment with its rows, bytes, time range and caps. See Retention.

The Storage page: each table's Parquet segments on its retention windowThe Storage page: each table's Parquet segments on its retention window

Filters

Logs, Traces, Sessions, Spend and Tools share a filter box. Write one filter per line. Filters combine with AND.

WriteMatches
key=valueRows whose attribute key is exactly value
key!=valueRows that have key with a different value
key~textRows whose key contains text, ignoring case. % and _ act as wildcards.
resource.key=valueThe same, against the resource attributes
body.key=valueThe same, against the parsed JSON body (logs only)

!= and ~ skip rows that don't have the key, so a filter only narrows. A line that doesn't parse is flagged next to the box.

Filters also work in the URL as repeated f parameters, for example /logs?f=tool_name%3DBash&f=success%3Dfalse. Time ranges are 15m, 1h, 6h, 24h, 7d, 30d or all, set with range=.

Look and feel

The settings panel in the top bar switches the palette, scheme (day, night or auto), font and style. The choice is stored in a cookie. Each option is also a URL parameter, so any look is a link, for example /?pal=riso&scheme=night&font=onest&style=crisp. The fonts are embedded in the binary, so the UI makes no outside network requests.

Beside the settings panel, the "forgets after" stickers show the retention lotel was started with for logs, traces and metrics.

Edit this page on GitHub