Releasesby Glyphtruss 0.18.6
- feat(loops/cli): json output for truss loops view Release 0.18.6
Also shipped Jun 5, 2026Baseten in June 2026 →
What's new in Sigma
AI Create AI columns (Beta) You can now create AI columns to enrich your data using natural language prompts. AI columns allow you to construct dynamic prompts that reference specific table columns, and can be helpful for tasks like enriching, summarizing, and classifying data. See Create AI columns (Beta) for more information.
Sandboxes, Infrastructure as Code
This week, Sandboxes are available in Priority Boarding from the dashboard and CLI, and Infrastructure as Code is available experimentally in CLI 5.2.0. Both features use the [Railway TypeScript SDK](https://github.com/railwayapp/railway-ts-sdk) (yes, we released a TypeScript SDK too).
Inbound SCIM Groups for Enterprise Connections is now Generally Available!
We’re pleased to announce that support for Groups within Auth0’s [Inbound SCIM](https://auth0.com/docs/authenticate/protocols/scim/configure-inbound-scim) for Enterprise Connections capability is now Generally Available (GA)!
Taxonomy MCP: Branch management and Create tools
Description: You can now build, iterate on, and ship tracking plan changes entirely through AI-assisted workflows. The Amplitude MCP server now includes create and branch management tools for taxonomy, bringing a workflow that was previously only available in the Amplitude UI to any AI assistant.
Web Experiment: Mutex groups, Holdout groups, and Experiment dependencies [Beta]
Description: As your experimentation program scales, controlling how experiments interact becomes critical. Web Experiment now supports mutual exclusion groups, holdout groups, and flag dependencies so you can run a more rigorous, isolated testing program. Mutual exclusion groups prevent users from seeing multiple experiments at once.
Per-Condition Retry Configuration
The new X-Fal-Retry-Config request header replaces all-or-nothing retry behavior with a separate retry budget per failure condition, allowing retry policy to be tuned independently for each failure mode.
More from trussEvery truss release →
truss 0.18.5
- cli: surface model_id, deployment_id, and endpoint after a deployment Fail fast when weights metadata resolves no files Bump truss-transfer to 0.0.43 and truss to 0.18.5rc0 fix: performance-client -> 529 non retriable feat(truss-chains): enable existing trusses to become chainlets Release 0.18.5
truss 0.18.4
- trt_llm/validation: fix AttributeError on class-level assignments and misleading class-not-found error refactor(loops): deactivate by deployment id feat(cli): truss loops logs --trainer-deployment-id / --deployment-id feat(loops/cli): show sampler status in truss loops view Release 0.18.4
truss 0.18.7
- feat: log bundled files at DEBUG during truss push feat(train): support B200 truss train workstations Release 0.18.7
truss 0.18.8
- Scope truss watch model resolution to a single model to avoid fetching all signatures Release 0.18.8
truss 0.18.9
- fix(loops): render deployments with sampler: null without crashing Fix chain watch detection for external package changes Release 0.18.9
truss 0.18.10
- feat(cli): Egress restrictions in truss runtime config Release 0.18.10
truss 0.18.12
- Revert "Add validation to prevent no_build with predict_endpoint " Release 0.18.12
truss 0.18.11
- Add time-range and filter flags to truss model-logs fix(truss-transfer): Disable flaky HF truss-transfer test Add validation to prevent no_build with predict_endpoint feat(chains): Add --no-sleep to chains watch fix(workstation): scope multinode SLURM rendezvous dir by job id Release 0.18.11