v0.1.4

What's Changed

• fix vllm issues by @rohitgr7 in #169
• Read upload sizes from the file server, not the NFS attribute cache by @freyam in #170
• Give each MCP request its own sync thread and close its connections by @freyam in #171
• Worker lifecycle sweep: heartbeat, redelivery, connector lease, workshop logs, optimizer lock by @freyam in #172
• Release the DB connection before a dataset event stream starts by @freyam in #179
• Keep Data Workshop memory bounded at any dataset size by @freyam in #182
• Make the Console dark-only by @freyam in #183
• Build each release from the family's latest revision and retire the capacity scripts by @freyam in #173
• Data Workshop without an agent: MCP versions, sandboxed runs, documents by @freyam in #187
• Match Console journey options by exact name by @freyam in #188
• Train and serve decision models from workshop datasets by @freyam in #189
• Fix verified bugs on main and remove dead PII NER code by @freyam in #193
• Datasets from versions, script attachments, version analytics and cancel by @freyam in #194
• Align Data Workshop journeys and skills with the merged MCP contract by @freyam in #196
• Offer no create_dataset hint after a stopped file import by @freyam in #197
• [fixes] Bound scoring DB connections, keep spans in their project, keep scan fields by @freyam in #199
• Integrate native workflows and DecisionJob identity on main infrastructure by @Tyler-Edwards in #198
• Use skeletons for Console content loading by @Tyler-Edwards in #202
• Bump docling-slim from 2.131.0 to 2.132.0 by @dependabot[bot] in #200
• Fix onboarding runtime recovery and endpoint consistency by @Tyler-Edwards in #203
• Simplify data onboarding and hand off SFT preparation by @Tyler-Edwards in #204
• Require detailed capability scans and make SDK analytics optional by @Tyler-Edwards in #205
• Use trained model decisions in inference code examples by @Tyler-Edwards in #206
• Start data onboarding with exploration by @Tyler-Edwards in #207

Full Changelog: v0.1.3...v0.1.4