Archive an OpenReview Conference

unicli openreview conference creates a resumable research archive for one OpenReview venue. It keeps each public submission together with its review thread, rebuttals, decisions, edit history, hosted files, and external artifact links.

Start with the venue group

Pass the canonical group id or its OpenReview group URL:

unicli --auth-retry openreview conference \
  "ICML.cc/2026/Conference" \
  --output ~/openreview-archives \
  --rpm 20 \
  -f json

unicli --auth-retry openreview conference \
  "https://openreview.net/group?id=ICLR.cc/2025/Conference" \
  --output ~/openreview-archives \
  -f json

The adapter reads the venue group's submission_id, prefers OpenReview API v2, and falls back to API v1 for older venues. Group configuration remains the source of truth when invitation names or decision labels differ across years.

Authentication

Use the signed-in browser session when OpenReview requires an access token or a browser challenge:

unicli browser profiles --json
unicli browser doctor --json
unicli --auth-retry openreview conference "ICLR.cc/2024/Conference"

The OpenReview client may refresh an expired access token once during the invocation. Access and clearance credentials are restricted to OpenReview hosts; the refresh credential is restricted to OpenReview's token endpoint. Only records and fields whose OpenReview readers include everyone enter the archive.

Archive layout

openreview-archives/
└── ICML.cc_2026_Conference/
    ├── manifest.json
    ├── group.json
    └── threads/
        └── <forum-id>/
            ├── record.json
            ├── edits/<edit-id>.json
            └── artifacts/<note-id>/<field>-<identity>.*

record.json retains the raw public submission and replies plus a normalized, chronological timeline. Timeline entries carry stable note and parent ids, invitation-derived lifecycle types, source URLs, all content sections, and a focused concerns view for weakness, concern, question, limitation, and issue fields.

OpenReview-hosted PDFs, supplementary files, appendices, source bundles, code packages, artifacts, and rebuttal attachments are downloaded serially. Each binary has a JSON sidecar with its source URL, size, media headers, and SHA-256. External repository, project, and dataset URLs remain structured links so their own host can be archived under its own authentication and rate-limit contract.

Request pacing and recovery

The default is 20 requests per minute with one active OpenReview request per process. The client observes Retry-After and rate-limit reset headers, and retries transient API failures with bounded exponential backoff and jitter.

Run conference jobs sequentially so their aggregate rate stays predictable:

for year in 2026 2025 2024; do
  unicli --auth-retry openreview conference \
    "ICLR.cc/${year}/Conference" \
    --output ~/openreview-archives \
    --rpm 20 \
    -f json
done

The manifest remains in_progress until submissions, revisions, and the final edit-head catch-up pass finish. Re-run the same command after interruption; the committed cursor and artifact hashes prevent completed work from being fetched again.

Use metadata-only mode to retain notes, revisions, and artifact references without downloading binaries:

unicli --auth-retry openreview conference \
  "NeurIPS.cc/2025/Conference" \
  --metadata-only \
  --output ~/openreview-archives \
  -f json

Inspect a tab before archiving

openreview venue accepts the URL shown in the browser, including a venue tab:

unicli openreview venue \
  "https://openreview.net/group?id=ICML.cc/2026/Conference#tab-accept-spotlight" \
  --limit 25 \
  -f json

The command resolves the tab through the venue's live decision-heading map, so display-label capitalization does not become an API query assumption.