~95% on SimpleQA (e.g. Qwen3.6-27B on a 3090). Supports all local and cloud LLMs (llama.cpp, Ollama, Google, ...). 10+ search engines - arXiv, PubMed, your private documents. Everything Local & Encrypted.
Workflow Status
Live status of every GitHub Actions workflow in this repo. Auto-generated by
scripts/generate_workflow_status.py. Do not edit between the generated markers — regenerate withpdm run python scripts/generate_workflow_status.py. Anything outside the markers is preserved on regeneration.
How to read this page
- Live badges (right column on active gates) re-render on every page
view and reflect the current head-of-default-branch status from
GitHub. Click one to land on that workflow's runs page. The badge is
the source of truth for current status — there is intentionally no
static status column, because that would flip every regeneration as
the most-recent run cycles through
success → skipped → in_progress. - Last activity uses coarse calendar buckets —
last 30 days,1-3 months ago,3-6 months ago,long ago,never. Exact dates would change every regeneration; buckets only change when a workflow drifts, which is the signal worth seeing in version-bump diffs. - Disabled = a caller has the
uses:line commented out, or the workflow is disabled in the GitHub UI. Stale = scheduled trigger but no successful run within 2× its cron cadence (and ≥60 days). The three top sections are the action items. - Reusable workflows (whose only non-gate triggers are
workflow_dispatchor pull-request events) show their gated run — the most recent run of their parent (release.yml, release-gate.yml, ci-gate.yml) that included them — not their own empty direct-run history.
78 workflows: 1 disabled · 2 stale · 1 manual-only · 74 active
⚠ Disabled workflows
| Workflow | Disabled where | Last direct run |
|---|---|---|
nuclei.yml |
release-gate.yml:219 (commented) |
never |
⚠ Stale (scheduled but no recent successful run)
| Workflow | Cron | Last successful run |
|---|---|---|
update-npm-dependencies.yml |
0 8 * * 4 |
never |
update-precommit-hooks.yml |
0 8 * * 5 |
never |
ℹ Manual-only by design
| Workflow | Last manual run | Trigger |
|---|---|---|
check-config-docs.yml |
long ago | manual |
Release-blocking gates — daily (release-gate cron 02:00 UTC)
Release gates — release-time only
Scheduled (own cron)
PR / push checks
Repository-dispatch publishers
| Workflow | Last activity | Trigger | Live badge |
|---|---|---|---|
publish.yml |
1-3 months ago | repo_dispatch |
Other
| Workflow | Last activity | Trigger | Live badge |
|---|---|---|---|
issue-research.yml |
last 30 days | issues | |
ldr-research-reusable.yml |
never | workflow_call |
Pending regeneration
Workflows added since the last dashboard regeneration. Regenerating
(pdm run python scripts/generate_workflow_status.py) needs an
authenticated gh, and it rewrites every "last activity" bucket on the
page — noise a single-workflow PR shouldn't carry. This is the
placeholder region that check-workflow-status.yml documents for that
case. It sits outside the generated markers, so a regeneration preserves
it verbatim: delete the row below by hand once the real one appears in
the table above.
milvus-tests.yml— never run yet; push:main, PR, manual. Path-filtered tosrc/local_deep_research/vector_stores/**,tests/vector_stores/**,pyproject.toml,pdm.lock,.github/workflows/milvus-tests.yml. Self-gates until the Milvus backend and themilvusoptional-dependency group land (#5331). Partial setups fail; only an entirely absent feature is skipped.security-mxss-browser-tests.yml— real-browser mXSS regression tests for the markdown rendering path (#6295). PR, manual, schedule (daily 02:30 UTC). Path-filtered to the spec, its fixture,playwright.mxss.config.js, the Playwrightpackage*.json, the sanitizer and service sources under test (web/static/js/security/**,web/static/js/services/ui.js,web/static/js/app.js), the rootpackage*.json, and the workflow itself. Needs no application server, no auth and no secrets: the spec loads the production sanitizer modules and its fixture page from disk viasetContent(), so it is safe to run on fork PRs.windows-database-tests.yml— native Windows SQLCipher and file-I/O tests on Python 3.12 and 3.14; PR to main, push:main, manual. Path-filtered to database/settings sources, config paths, locked dependencies, the native test modules, runner, and workflow.osv-scan-reusable.yml— sharedworkflow_callscanner used byosv-scanner.ymlandosv-scanner-scheduled.yml. Verifies the native scanner binary and complete scan results before artifact and SARIF uploads. Its status is reported in the calling workflows.rag-vector-scale.yml— manual large-collection latency benchmark for segmented FAISS (#7810). Flat and HNSW shards build a synthetic collection (1,000,000 vectors of 1,024 dimensions by default) without PDFs or an embedding model, and report per-document write time, search time and peak RSS. It is not called byrelease.ymland never runs on pull requests.rag-index-scale.yml— required synthetic-PDF scale gate called byrelease.yml, with manual dispatch available. Separate flat and HNSW shards check bounded index writes, memory, search, rollback, deletion, staged rebuilding and encrypted storage using deterministic embeddings. The large workloads never run on pull requests. They run only throughrelease.yml— on pushes tomainorv*.*.*tags and on manual release dispatches, gated onshould_release(amainpush whose version already has a release skips them) — or through this workflow's own manual dispatch.
Offline performance assertions
performance-assertions.yml runs the available offline benchmark suites in a
serial, bounded CI profile on relevant pull requests and main-branch changes.
It explicitly enables integration checks and rejects missing, failed or skipped
JUnit evidence and unverified measurement reports. When present, the original
synthetic-PDF and vector-scale smoke cases cover flat and HNSW; larger release
and manual workloads retain their existing profiles.
Each invocation stores results in a fresh run-* directory under its output
path and updates ci-summary.json with the current run directory. Reports must
match the requested area/profile and contain finite samples consistent with
their counts and summaries. Existing manual-shard contracts join the selection
when available. Native database, HTTP and RAG smoke measurements and retrieval
quality reports are copied out of pytest temporary storage into the uploaded
native/ results; database/RAG/HTTP use their fixed, asserted smoke settings,
recorded in each report. Use their standalone CLIs for larger local workloads.
Only numeric JSON and JUnit evidence is retained, rather than temporary databases.