Links that open the run
Share a URL and it opens the WebAssembly viewer at the exact frame, camera angle and panel layout you were looking at. No install on the other end.
Hub — now in open beta
Recordings your team
Recordings your team
can actually find
Dalaran Hub keeps every run in one place: uploaded from CI or a laptop, indexed by metadata, opened in the browser by anyone with the link. The local tools stay free — Hub is what happens when a recording needs to outlive the machine that made it.
| Recording | Branch | Duration | Size | Result | When |
|---|---|---|---|---|---|
| warehouse_run_042 | main | 24.6 s | 1.4 GB | pass | 3 min ago |
| grasp_eval_sweep_11 | feat/grip-v3 | 3 m 12 s | 8.9 GB | 2 failed | 28 min ago |
| nightly_regression | main | 41 m | 62 GB | pass | 6 h ago |
| sim2real_delta_07 | exp/domain-rand | 1 m 04 s | 420 MB | pass | yesterday |
| splat_convergence_3 | research | 12 m | 14 GB | — | 2 days ago |
What Hub adds
The parts that need a server
Search across runs
Query by tag, commit, branch, duration, entity path or custom metadata. Find every recording where the tracking error crossed a threshold.
Streaming playback
Recordings stream from object storage on demand. Open a 60 GB nightly run without downloading a byte more than the frames you scrub to.
Access control
Org, team and per-recording permissions. SSO via SAML or OIDC, audit logs on every view and download, expiring links for external reviewers.
CI upload
One step in your pipeline pushes artifacts with the commit SHA attached. Failed tests link straight to the recording that failed them.
Comments & markers
Drop an annotation on a timestamp and an entity. Threads live with the recording, so the discussion and the evidence never drift apart.
Getting data in
Two lines from local to shared
Point the SDK at Hub instead of a file and the recording lands in your workspace as it streams. Or upload afterwards from the CLI.
- Resumable uploads that survive a dropped connection
- Content-addressed chunks — repeated assets stored once
- Retention policies per project, from 7 days to forever
- Export back to
.dlrat any time; no lock-in
import dalaran as dl
dl.init("nightly-regression")
dl.connect_hub(
workspace="halcyon-robotics",
project="perception",
tags=["nightly", commit_sha],
)
run_evaluation() # logs as usual — it just goes to Hub too
- name: Upload recordings
run: |
dalaran hub login --token "$DALARAN_TOKEN"
dalaran hub push artifacts/*.dlr \
--project perception \
--tag "sha:$GITHUB_SHA" \
--tag "branch:$GITHUB_REF_NAME"
env:
DALARAN_TOKEN: ${{ secrets.DALARAN_TOKEN }}
Security
Built for data you can't leak
Encrypted
TLS 1.3 in transit, AES-256 at rest, per-workspace keys with optional customer-managed KMS.
Regional
Choose US, EU or AP storage per workspace. Data never leaves the region you pick.
Self-hosted
Enterprise plans can run Hub inside your own VPC against your own object store.
Auditable
Every view, share and export is logged and exportable to your SIEM.
All systems operational
API 99.98% · Viewer CDN 100% · Upload 99.95% — trailing 90 days
Status history
Start with the free workspace
10 GB, unlimited public recordings, three seats. No card.