Tale v0.5.1

0.5.1 repairs the fresh-deploy path that 0.5.0 shipped broken. Start new installs from 0.5.1; if a 0.5.0 tale deploy failed on you, redeploy with the 0.5.1 CLI.

Highlights

This is a fix release on the 0.5 line. v0.5.0 could not complete a fresh tale deploy on a clean host — seven independent defects sat between tale init and a working stack, all fixed in #3108:

• The deploy pull list requested tale-backend-api / tale-backend-worker images the release pipeline never builds — the backend tier runs the platform image — so every fresh deploy died at "Failed to pull image". Service→image resolution now goes through one shared map used by both the compose creators and the pull list.
• The CLI-generated compose never set DATABASE_URL on the backend services, which crash-looped on their env schema.
• object-store was in the compose but in no deploy tier, so the blob store never started: uploads dead, the worker ENOTFOUND at boot. It now deploys as a stateful, stop-gated service the backend depends on.
• The deploy health wait only accepted healthy and burned its full timeout on backend-worker, which deliberately has no healthcheck; running with no configured healthcheck now counts as ready.
• The proxy kept the 0.5 lanes — SSE /events, blob routes, both machine doors, SSO/SCIM — behind the cutover's BACKEND_UPSTREAM switch, which tale deploy never sets. The lanes are now unconditional, with BACKEND_UPSTREAM remaining an override for split deployments.
• The platform image was missing the shipped config catalogs (/app/system, /app/builtin) — every provider read failed and org scaffolding had no seed catalog — and never owned /app/data, so the backend died with EACCES on the org-config volume. Both catalogs are baked into the image now, and the entrypoint re-asserts ownership on volumes a 0.5.0 image already booted against.

The full journey — tale init → tale deploy → onboarding → provider credential → chat streaming → document upload with RAG indexing — was verified end to end on a clean host, and two new guards (a compose↔release-matrix parity test and a container-image check in the release gate) hold the pipeline to it.

Migration notes

No database migrations between 0.5.0 and 0.5.1. On the 0.5 line, SQL migrations apply automatically at boot; this release adds none.

Upgrading

• New install — 0.5.1 supersedes 0.5.0's artifacts:

  curl -fsSL https://raw.githubusercontent.com/tale-project/tale/main/scripts/install-cli.sh | bash
  mkdir tale-05 && cd tale-05
  tale init
  tale deploy

• A host where a 0.5.0 deploy failed or half-started: update the CLI (tale update, or the install script above) and run tale deploy again in the same project directory — volumes a 0.5.0 image booted against are healed automatically.

• Coming from 0.4.x: 0.5 remains a breaking minor with no in-place upgrade in either direction — read the v0.5.0 release notes and Upgrades → 0.4 → 0.5: breaking cutover before touching anything.

What's Changed

• fix(platform): make the 0.5 fresh deploy reach a working stack by @larryro in #3108

Full Changelog: v0.5.0...v0.5.1