The portal
Bifrost’s control plane is driven from a review-first portal: a portfolio heatmap of migration risk, a deterministic risk breakdown per pipeline, a three-pane review where a human approves or edits the converted workflow, and a queue that tracks every pipeline through the lifecycle.
Portfolio heatmap
Point Bifrost at an Azure DevOps org and get a portfolio-scale heatmap of migration risk — grouped by project, coloured Green / Amber / Red, with the converted-ratio and risk score on every tile. Classic (designer) pipelines — the hard tail — are flagged distinctly. The header shows the pinned Importer version and whether air-gap mode is active.

Deterministic risk, explained
Click any pipeline to see why it scored the way it did. The score is computed from weighted, explainable factors — container jobs, variable groups, service connections, multi-stage gates — never from the LLM. The model explains and fills gaps; it does not score.

Table view
The same portfolio as a dense, sortable table — type, risk, converted-ratio, manual tasks, review status, and forecast runner-minutes per pipeline.

The three-pane review
The heart of the review-first workflow. Left: the original ADO pipeline. Middle: the converted GitHub Actions workflow, with the gaps the LLM filled highlighted. Right: the rationale, the deterministic risk flags, the verify-before-approving steps, the manual-task runbook, and the immutable audit trail. A reviewer approves, or edits the workflow inline, before anything is committed.

Edits are made in place and recorded — the audit trail captures who changed what, and why.

Review queue
Every pipeline, tracked through the proposal lifecycle —
draft → in_review → approved → committed → validated — with a migration-progress bar and the
last action (who, when) on each row.

Docs and help, in the portal
Operator documentation ships inside the portal itself — getting started, using the portal, and connecting to live data — so it travels with the tool, air-gapped or not.

Screenshots show Bifrost running against synthetic enterprise sample data
(contoso). No real pipeline definitions or secrets are shown.