Features

Run Monitoring & Insights

Runs list -- See all workflow runs with search, status filtering, and sortable columns. Active runs refresh automatically.

Run detail -- Central hub for a single run: workflow info (repo, revision, commit), runtime stats (status, duration, peak concurrency), and tabbed access to the process summary, task timeline, cost & performance, config & params, logs, reports, AI analysis, error & resume, and run history.

Process table -- Per-process task counts by status (succeeded, failed, cached, retried, etc.). Click any process to see resource analytics.

Task timeline -- Gantt-style visualization of every task across three phases: queue wait, execution, and staging overhead. Group by hostname, instance type, or capacity type to spot infrastructure patterns.

Process analytics -- Box plots for CPU, memory, I/O, and duration across all tasks in a process. Compare actual usage against allocated resources to identify right-sizing opportunities.

Task drill-down -- Full execution details for any task: resource metrics, timing breakdown, instance info with pricing, exit codes. Direct links to Datadog pod metrics.

Resume history -- Full lineage of resumed runs with per-cycle error details, failed tasks, and cached task counts.

Stale run detection -- Heartbeat-based detection of crashed runners. Automatically marks abandoned runs as failed.

Rich infrastructure failure attribution -- Tracker labels infra-level failures so you can tell them apart from workflow bugs: spot evictions, disk exhaustion, image-pull errors, OOM kills, and other Kubernetes-level events. Visible on the task detail page and queryable via MCP.

Config and params -- View the full resolved configuration and parameters for any run. Together with the git repo and commit, this is what you need to reproduce a run.

Logs and notes -- View workflow runner logs from S3 directly in the UI. Annotate any run with freeform notes visible to all users.

Cost Analysis

Per-process cost breakdown -- See which processes cost the most. Uses 1-minute CPU-proportional time slicing for accurate attribution across shared nodes.

Cross-run cost sharing -- When multiple runs share an EC2 host, costs are split proportionally by CPU allocation. No double-counting.

Spot vs on-demand comparison -- Actual costs alongside a "what-if all spot" estimate for evaluating savings potential.

Wasted cost -- Quantifies the cost of failed and aborted task attempts to help prioritize reliability improvements.

Output storage estimate -- Total size of published outputs plus an annual S3 storage cost range, accounting for Intelligent-Tiering. Shown on the run detail page after the run reaches a terminal status, and on the Cost & Performance tab when that report is available.

Process timeline -- Visual timeline of when each process was active, revealing parallelism and sequential bottlenecks.

AI-Powered Tools

Auto AI analysis -- Every completed run is analyzed by Claude Code on Kubernetes. Failed runs get root-cause analysis with evidence and recommendations. Succeeded runs with MultiQC output get QC summaries. Results appear in the AI Analysis tab.

MCP server -- Read-only database access for AI assistants via Model Context Protocol. Six tools: schema (column-level DB schema), query (SQL), cost_and_performance (structured cost report), qc_report_summary (MultiQC data), ai_analysis_result (auto-analysis markdown), and url (build deep links to runs/tasks/processes). Works with Claude Code, Claude Desktop, and any MCP client.

Reports & File Browsing

Configurable reports -- Workflows declare reports via tracker.yml (glob patterns + display names). Tracker renders them inline: HTML, PDF, CSV/TSV tables, images, JSON, text. Compatible with Seqera's tower.yml.

Output browsing -- Directory tree navigation of published outputs with search and clipboard copy of S3 paths.

Work directory browser -- Browse any task's S3 workdir to inspect .command.sh, .command.err, .command.out, and other execution artifacts inline.

File preview -- Inline rendering for HTML, PDF, CSV/TSV, JSON, images, and text. Up to 25 MB.