[release] v0.104.0#4717
Merged
Merged
Conversation
… parity Assessment for making OSS multi-organization with AGENTA_ACCESS_* enforcement, plus the two-step schema parity plan (drift fixes on kept tables, legacy table drop). Includes live schema dumps of both editions' core/tracing DBs, their diffs, the dump script, and the PR execution plan.
…ble chain ids Record the v0.103.5 propagation (b3c4d5e6f7a9 re-point, App-class collision in _admin_detach_user_references, db_manager_ee import/symbol churn) and the 12-char park/oss/ee revision-id rename. Update checkpoint ids and the chain layout diagram. Flag the committed schema dumps as stale (predate the workflow_revisions repair). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Same scheme as core (no t-prefix); tracing has its own database so the version tables (alembic_version_tracing_oss/_ee) and ids do not collide. tracing_ee is root+proof only (no EE-only tracing tables to adopt on a switch). Update the layout note and runner order. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…e core Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
dump_pg_schema.sh used ${vt^^} (bash 4+ uppercase expansion) which fails on
macOS system bash 3.2 with 'bad substitution'; replace with a portable tr.
Delete the committed schema dumps and diffs (oss/ee core+tracing, diff_*) —
they predate the v0.103.5 workflow_revisions repair and are regenerated from
replays, so the stale copies are misleading.
is_auth_info_blocked() early-returned False outside EE, so the four AGENTA_ACCESS_* env vars were parsed but never enforced in OSS. Deleting the gate activates the shared blocklist/allowlist logic in both editions.
is_auth_info_blocked no longer reads is_ee, so patching it raised AttributeError; the tests now exercise the path both editions share.
Moves OrganizationMemberDB/WorkspaceMemberDB/ProjectMemberDB from EE to the shared OSS models (EE re-exports), creates the three tables in the OSS chain with the exact EE shapes, backfills membership rows from owners and used invitations, and aligns api_keys/projects with the shared model in both chains. get_user_organizations now uses the membership join in both editions.
…atabases Fresh replays seed a default project (911e6034d05e) before any organization can exist, so the parity migrations' NULL-scope guard killed the chain on new deployments. Delete NULL-scope projects when the database has no users (the fresh-replay signature); real deployments keep failing loudly. The OSS org backfill now prefers the legacy singleton (slug oss-default, NULLS LAST so unslugged orgs sort after) before falling back to the oldest org, making adoption deterministic with 0, 1, or N orgs and NULL or set slugs.
Lifecycle actor columns carry no FKs by convention (settled in the phase-2 review). The parity direction reverses for this column: instead of EE gaining the OSS FK, OSS drops its legacy FK in the cleanup migration. Only the nullable-ification stays here. Editing the unshipped migration avoids shipping an FK that the next PR would immediately drop.
…the chain The v0.103.5 merge dropped b3c4d5e6f7a9 (repair_workflow_revision_versions), the head of the v0.103.5 core chain, and re-rooted the membership/parity migrations onto its parent a2b3c4d5e6f8. Any database that ran v0.103.5 is stamped at b3c4d5e6f7a9, so upgrading it into the stack failed with 'Can't locate revision b3c4d5e6f7a9'. Restore the migration (OSS + EE) and its shared data_migration verbatim, and re-root 0a1b2c3d4e5f / 2c3d4e5f6a7b onto b3c4d5e6f7a9 so the chain is a2b3c4d5e6f8 -> b3c4d5e6f7a9 -> membership.
The 911e6034d05e boot seed creates an empty default project before any org exists. On a populated DB the app already made a real default project per workspace at signup, so adopting the seed left two is_default projects in one workspace and the members view (which reads the default project) showed only the seed's lone member. Drop the unscoped empty seed when a scoped default already exists; the org keeps exactly one default.
Default workspace resolution becomes membership-based in both editions (get_default_workspace_id moves OSS-ward; the OSS oldest-workspace fallback and get_oss_organization are gone). Organization and workspace listings are scoped to the requesting user's memberships. The admin singleton org/workspace delete guards and the admin_create_organization ON CONFLICT branch are removed. Web: org switcher, New Organization, and owner submenu are enabled in OSS.
…d-trip The workspace-resolution unit tests move to the OSS suite (the function moved OSS-ward; patching db_manager_ee's engine no longer reaches it). The admin org create/delete round-trip returns to the OSS acceptance suite now that the singleton slug-collapse and delete guard are gone.
OSS no longer needs the owner-signup + invite + re-login dance: any user can sign up directly and gets their own organization, same as EE. global-setup drops the license fork and inviteOssUser; AGENTA_TEST_OSS_OWNER_EMAIL stays as an optional fixed-account login for persistent CI deployments.
Moves the org-creation core (org + owner membership + default workspace/ project + seeded defaults) into shared OSS code (oss commoners + db_manager); EE commoners slims to subscription/entitlement wrappers over it. OSS signup converges on the EE flow: personal org per allowed user, no first-user bootstrap, no invite-only gate. Mounts create/update/delete/transfer organization endpoints in OSS, lifts the admin-API multi-org block, and writes membership rows on invitation accept.
The api-key auth path sets request.state.user_id to the user's DB id (api_key.created_by_id), but get_user only matched UserDB.id when is_ee(). In OSS the new user-facing POST /organizations/ resolved the api-key user via get_user and 404'd. Apply the id fallback in both editions, guarded so a non-UUID SuperTokens uid still matches on uid only. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The data migration imported live ORM models, so a later model change could break replaying revision 1b2c3d4e5f6a on a fresh or delayed upgrade. Use local sa.table()/sa.column() definitions pinned to the schema at this revision instead. Backfill logic is unchanged.
Adds acceptance tests for the OSS create/rename/transfer/delete organization endpoints and unit tests for can_create_organization. Lifts the leftover is_ee gates on the simple membership-create endpoints — they wrap the graph path, which already creates memberships in both editions.
meters_type is an enum, not a table: call it five schema objects (one enum, four tables). Specify how create_organization derives a unique slug (slugify the name, numeric suffix on collision) instead of the vague 'generate unique slugs'.
Rewrites the restrict-organization-creation guide for both editions around the canonical AGENTA_ACCESS_* vars, adds a migration page for OSS operators (open-signup posture flip, membership backfill, legacy-table drop), updates quick-start/upgrading/configuration/dynamic-access-controls/opensource/ organizations pages, and documents the GitButler workflow in AGENTS.md.
Record what bit us during the v0.103.5 propagation: series need linear history (merge-connected stacks collapse on unapply/apply), 'but pull' rebases on main not the branch's own upstream, stage-then-commit --only to target a mid-stack branch, and 'but oplog restore' to recover a collapsed stack. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Covers the in-place edition switch for both Docker Compose and Kubernetes: stop the stack, point EE at the existing agenta_oss_* databases (via POSTGRES_DB_PREFIX or explicit POSTGRES_URI_*), reuse the same env file so the Compose project name and volume are kept, then start EE. The EE migration creates the EE-only tables and backfills on first run.
Drops the 20 dead pre-project tables (apps/variants/bases, deployments, docker_images, templates, old environments, old testsets, old auto/human evaluations, ids_mapping) in both chains, deletes the 15 dead model classes, the unimported models/db/models.py shim and models/converters.py, and the dead create_deployment in db_manager_ee. The drop is the schema-parity fix for the legacy organization_id/workspace_id columns EE carried on these tables. deprecated_models scaffolding stays (historical migrations import it).
Alembic loads every version script at startup, and several historical migrations still imported the model classes deleted with the legacy table drop (create_free_plans imported AppDB, the evaluators/environments/ applications_workflow/projects data migrations imported the others). This broke the migration runner on boot. Freeze the final pre-drop shapes of the nine queried tables in deprecated_models.py on a dedicated DroppedBase (a separate base so extend_existing does not merge columns into the older partial shapes) and point the migrations at them. Verified all 175 version scripts across both chains now import cleanly.
The frozen classes keep their ForeignKey declarations, but projects/users/ folders are not in DroppedBase's metadata, so the ORM could not resolve FK targets when a data migration flushes an insert (55bdd2e9a465 adds a testset row). Stub tables with just the id column satisfy resolution; selects never needed it, which is why the import sweep passed.
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
[fix] Clean up `workflow` `uri`<>`type` mapping
[chore] Add 'Rename' action to row menu in Prompts
…OSTGRES_PORT
POSTGRES_PORT only remaps the host-published port (compose
"${POSTGRES_PORT:-5432}:5432"); the bundled Postgres always listens on 5432
inside the Docker network. Deriving the in-network URIs from POSTGRES_PORT
made every service dial postgres:<host-port> and fail on a fresh start
whenever POSTGRES_PORT was customized. Use 5432 for the derived URIs and
drop the now-unused port attribute; external databases on non-default ports
still set explicit POSTGRES_URI_*.
fix(api): derive Postgres URIs against the fixed internal port, not POSTGRES_PORT
junaway
approved these changes
Jun 18, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
New version v0.104.0 in