refactor(core): add shared runtime orchestration#1002
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: bb65be817b
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
phernandez
changed the title
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: f338a6d382
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 703206a432
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
phernandez
force-pushed
the
codex/repository-explicit-sessions
branch
from
703206a to
07a8882
Compare
|
@codex review |
1 similar comment
|
@codex review |
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 58e93ca7c8
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
3 similar comments
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
|
@codex review |
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
phernandez
force-pushed
the
codex/repository-explicit-sessions
branch
from
c3c5a7b to
bd4b25c
Compare
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: bd4b25ca3f
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
Addressed the five latest Codex P2 findings in 07c25b2:
Regression coverage was added for each bug. Verification:
@codex review |
|
Codex Review: Didn't find any major issues. More of your lovely PRs please. Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
Signed-off-by: phernandez <paul@basicmachines.co>
Signed-off-by: phernandez <paul@basicmachines.co>
Signed-off-by: phernandez <paul@basicmachines.co>
Signed-off-by: phernandez <paul@basicmachines.co>
Signed-off-by: phernandez <paul@basicmachines.co>
Signed-off-by: phernandez <paul@basicmachines.co>
Signed-off-by: phernandez <paul@basicmachines.co>
Signed-off-by: phernandez <paul@basicmachines.co>
Signed-off-by: phernandez <paul@basicmachines.co>
Signed-off-by: phernandez <paul@basicmachines.co>
The change-routing loop matched project roots in arbitrary repository order and broke on the first containing root, so a parent listed before a nested child stole the child's file events and indexed them into the wrong project. Match deepest (longest root) first so the most specific project wins. Test covers a nested child with the parent listed first. Addresses Codex on #1002. Signed-off-by: phernandez <paul@basicmachines.co>
The directory-delete enqueuer ran the guarded file delete inline and discarded its RuntimeFileDeleteResult, so finish_directory_delete_acceptance reported every DB-deleted row as a successful file delete — even when the guard intentionally left a file on disk (checksum changed), which then reappears on the next index. Return the result up through enqueue_directory_file_delete_jobs, aggregate skipped files in the accepted result, and report them as failed_deletes with reasons. Cloud (queue-based) enqueuers return None and stay deferred. Addresses Codex on #1002. Signed-off-by: phernandez <paul@basicmachines.co>
The relation-resolution scheduler coalesced any write arriving while a pass was pending into that pass. A write that committed after the scan read the unresolved rows (but before the pending flag cleared) was dropped, leaving its forward references unresolved until an unrelated later trigger. Track a dirty bit, cleared once the scan starts, and re-arm exactly one follow-up pass when a write lands during the scan. Bursts still coalesce to one pass. Addresses Codex on #1002. Signed-off-by: phernandez <paul@basicmachines.co>
Local writes accept and materialize (write the markdown file + index) off the request path via the bounded worker pool. One-shot clients (bm tool write-note, importers) returned right after enqueuing, and run_with_cleanup then shut the DB and event loop down with no drain, so the source-of-truth file could be lost even though the API reported the note accepted. Add drain_pending_materializations() (awaits the pool) and call it in run_with_cleanup before db.shutdown_db. Long-lived servers keep the loop alive and are unaffected. Addresses Codex on #1002. Signed-off-by: phernandez <paul@basicmachines.co>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 1394bfce55
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
The router scheduled relation resolution right after the accept enqueued the deferred materialization. Under write load the 0.5s debounce could scan before the queued index_file inserted the new note's rows, so inbound forward references to it stayed unresolved until an unrelated later write. Inject the relation scheduler into the local materializer and schedule a pass after index_file completes; the scheduler (with the dirty-bit re-arm) coalesces it with the eager router pass. The router call is kept for cloud parity. Addresses Codex on #1002. Signed-off-by: phernandez <paul@basicmachines.co>
|
Worked through the remaining Codex threads — all twelve fixed, each with a regression test:
(Plus the four earlier threads resolved as already-addressed / not-applicable.) Resolving these threads now. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 39a3ac28b7
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
The os.walk-based scan returned an empty snapshot when the project root was missing/unmounted (os.walk swallows the error), so the project-index coordinator classified every indexed path as deleted and removed the project's entities. Add an onerror that re-raises for the root (deeper failures still prune that subtree), and re-raise rather than return an empty delete-everything snapshot when nothing was discovered. Addresses Codex on #1002. Signed-off-by: phernandez <paul@basicmachines.co>
|
Re: Resolve permalink-prefixed local watch events — not applicable on this branch. The local watcher path resolves events with |
DirectoryDeleteService opened its transaction with a raw session_maker() session, where SQLite leaves PRAGMA foreign_keys OFF. The bulk delete issues a Core DELETE on entity and relies on ON DELETE CASCADE for note_content/observations/relations, so those child rows were orphaned (and later reindexed). Route through db.scoped_session, which enables foreign_keys for SQLite (and manages commit/ rollback), matching every other FK-dependent delete. Addresses Codex on #1002. Signed-off-by: phernandez <paul@basicmachines.co>
When a background materialization wrote the markdown file but publish then found the note had moved (stale_file_path) or disappeared (missing) before publish, the just-written file was left on disk while the DB no longer owned that path, so the watcher/project index could re-import it as a duplicate. Flag those outcomes (written_file_orphaned) and enqueue a checksum-guarded cleanup of the written file; stale_db_version (same path, re-materialized by a newer pending version) is left alone. Addresses Codex on #1002. Signed-off-by: phernandez <paul@basicmachines.co>
The accepted-note move only checked for a DB conflict on the destination, so a move onto a file that exists on disk but is not indexed committed the entity/ note_content path to the destination while the file stayed unchanged, leaving DB/search pointing at it and the source to be reindexed as a duplicate. Add a storage-existence guard (verify_move_destination_absent) mirroring the create/PUT check, gated by verify_storage_absent_on_create (local True, cloud DB-first), and reject with 409. Addresses Codex on #1002. Signed-off-by: phernandez <paul@basicmachines.co>
|
Addressed the five follow-up threads from the latest Codex re-review:
All have regression tests. Resolving the four real ones; the watch-permalink one is resolved as not-applicable. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 7feb3cd0e0
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Threading the CLI embeddings flag into the search/FTS rebuild (the search-only fix) meant a full reindex embedded twice: the project-index phase ran inline embedding sync, then the explicit embeddings phase called reindex_vectors(force_full=True), discarding and rebuilding the vectors. The FTS rebuild must never embed — the explicit phase owns vectors — so pass embeddings=False there. Test asserts a full reindex calls the embedding provider once. Addresses Codex on #1002. Signed-off-by: phernandez <paul@basicmachines.co>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 94812485e5
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| # local writes accept and materialize off the request path, so a | ||
| # one-shot client would otherwise exit before the markdown file is | ||
| # written (the API already reported the note accepted). | ||
| await drain_pending_materializations() |
There was a problem hiding this comment.
When a local one-shot client such as bm tool write-note accepts a write, run_with_cleanup() now waits for the deferred materialization/indexing, but that materializer schedules relation resolution via an untracked debounced asyncio.create_task. After this line returns, asyncio.run() closes the loop and cancels the pending relation task, so inbound forward references to the newly written note remain unresolved until another write or full reindex runs. Add a drain/foreground path for relation-resolution follow-ups before shutting the DB/loop down.
Useful? React with 👍 / 👎.
…icit-sessions # Conflicts: # tests/services/test_initialization_cloud_mode_branches.py
Add the new verify_move_destination_absent method to the move-preparer test fakes (_CreatePreparer, _MovePreparer) so they still satisfy the AcceptedNoteMovePreparer protocol, and guard the new reindex/materialization tests against Optional await_args/materialization. Fixes the Static Checks (ty) failure on #1002. Signed-off-by: phernandez <paul@basicmachines.co>
- change planning: move destinations must be paths with no indexed row, restoring new-files-only semantics so an in-place edit matching a deleted file's checksum is never reclassified as a move (regression test added) - background schedulers and startup indexing hold strong task references so GC cannot cancel vector sync, project index, search reindex, or initial index tasks mid-flight - move_entity validates destinations through normalize_note_move_destination_path, matching move_note semantics - markdown checks consolidated onto runtime contracts helpers; link resolution now registers extensionless aliases for .markdown files too - normalize_note_move_destination_path returns a plain str; the single-field RuntimeNoteMoveDestination wrapper is gone - dead RuntimeStorageEventSkipReason.non_markdown removed (never emitted by the planner) - watch_coordinator deferred import documented Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
run_note_materialization enqueues old-path cleanup after the materialized write and its DB state are durable. A transient queue error there used to propagate and mark an already-successful job as failed (and drop the cleanup silently). The runner now catches the enqueue failure, logs it, and surfaces cleanup_enqueue_failed on the result so runtimes can alert on the orphaned old-path file instead of re-running finished work. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
NoteContentMutationService resolves the actor context for each mutation through an optional NoteContentMutationActorResolver before building the accepted-note actor. Routes keep passing their defaults; a runtime adapter can replace them with request-derived identity (user profile, source header, MCP actor headers) without subclassing the service. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…to focused modules Mechanical module split, no behavior change (Step 2 of the identity split plan): - runtime/contracts.py (2703 lines) is deleted and split into one-way layered modules: storage.py -> projects.py -> jobs.py -> workflows.py / note_content.py -> cleanup.py. runtime/__init__.py re-exports the same public names from the new modules (temporary barrel until Step 4). - indexing/project_index_workflow.py (1919 lines) is split into project_index_coordinator.py (fan-out + workflow request), project_index_maintenance.py (move/delete maintenance + SQL), and a shrunken project_index_workflow.py (workflow metadata planning + queued/logical-key helpers that move to cloud later). indexing/__init__.py re-exports the same names. - Cloud-only symbols (RuntimeWorkflowBroker/Transport, RuntimeQueuedWorkflowMetadata, RuntimeJobStatus, RuntimeJobReference, TenantId) are relocated, not removed; Step 3 deletes them. - All direct `runtime.contracts` / `project_index_workflow` imports across src and tests are repointed to the new concrete modules; maintenance/coordinator tests move to test_project_index_maintenance.py / test_project_index_coordinator.py. Placement notes vs the design doc, forced by the one-way layer order: - WorkflowId lives in runtime/jobs.py (job request shapes carry workflow ids; workflows.py imports it from jobs). - ProjectIndexMaintenanceRunner lives in project_index_maintenance.py because StoreProjectIndexMaintenanceRunner subclasses it and the coordinator imports maintenance, never the reverse. - RuntimePendingNoteFileDelete lives in note_content.py (its planners are note-content write planning); cleanup.py imports it. - Project* aliases live in storage.py since storage-level values reference them. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ith safe semantics The change-planning fix classifies a rename onto an existing indexed path as modify+delete (never a move), because the scan snapshot cannot distinguish it from an in-place edit matching a deleted note's checksum — and misclassifying that edit as a move drops the edit. The project-index test now locks in the safe semantics. The startup-index test fake gains add_done_callback for the strong-task-reference scheduler. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…d contracts Identity split (Step 3 of the runtime-contracts decomposition, #1002): basic-memory core is single-tenant and queue-agnostic; tenancy and PGQ transport identity belong to the cloud runtime that owns them. De-tenanted contracts (fields removed, not made optional): - runtime/jobs.py: RuntimeStorageFileIndexJobIdentity/Context/Observation, RuntimeProjectIndexJobRequest, RuntimeIndexFileBatchJobRequest, RuntimeProjectDeleteJobRequest. Dedupe keys are now project-scoped (index-file:{project_id}:..., index-file-batch:{project_id}:{batch_index}, index-project:{project_id}, delete-project:{project_id}) and routing headers carry project identity only. - runtime/note_content.py + runtime/cleanup.py: note materialization and note-file delete job requests and their plan_* builders. - runtime/job_payloads.py + indexing/job_payloads.py: worker payload models lose tenant_id/workflow_id; cloud subclasses re-add them on its wire. - indexing runners/contexts: IndexFileRuntimeRequest, live-update and relation-resolution contexts/plans, embedding index requests, directory and project delete acceptance, accepted-note enqueue orchestration. - ProjectIndexWorkflowRequest/Source renamed to ProjectIndexRequest(Source) and now carry only the project reference plus force_full/search/embeddings. ProjectIndexFanoutFailureRecorder no longer takes workflow_id; scope-owning adapters carry their own workflow identity. Opaque queue transport: - build_project_index_workflow_start / plan_project_index_workflow_start take transport_metadata + transport_event_data mappings instead of (transport_broker, transport_entrypoint, transport_job_id). - RuntimeWorkflowAttemptMetadata.attempt_started_event_data takes an opaque transport_event_data mapping instead of pgq_job_id. Event/metadata key ordering is preserved so persisted rows stay byte-identical when cloud passes its pgq fields. Deleted cloud-only modules/symbols (re-homed by cloud in Step 5): runtime/job_identity.py, runtime/job_status.py, runtime/workflow_metadata.py, RuntimeWorkflowBroker/Transport, RuntimeQueuedWorkflowMetadata, RuntimeJobStatus, RuntimeWorkflowQueueEnvelope, RuntimeJobReference (dead), TenantId alias, indexing/project_index_enqueue_runner.py, project_index_workflow_logical_key + build_project_index_workflow_queued + ProjectIndexWorkflowQueued, importers/job_payloads.py. Their unit tests moved out of the core tree alongside them. Local sentinel removal: LOCAL_EVENT_INDEX_TENANT_ID, LOCAL_NOTE_CONTENT_TENANT_ID, uuid4() workflow ids in local project index, get_runtime_tenant_id/RuntimeTenantIdDep, and the tenant param on the delete-directory route/service are gone; local code now builds plain project-scoped requests. Barrels (runtime/indexing/index __init__) stay importable with the deleted names removed; full barrel removal is a follow-up step. The workflow-gating behavior for per-file live updates and relation resolution moves to the runtime adapters that track workflow membership. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Reduce basic_memory.runtime, basic_memory.indexing, and basic_memory.index __init__.py modules to docstring-only packages with no re-exports or __all__. Every consumer now imports each name from its concrete owning module (e.g. basic_memory.runtime.storage, basic_memory.indexing.models, basic_memory.index.local_project), including the documented deferred imports in services/initialization.py, cli/commands/db.py, and index/watch_coordinator.py, and TYPE_CHECKING-only imports. The index package's cross-package re-exports (runtime storage-event contracts and the project_index shim over basic_memory.indexing) stop too: src/basic_memory/index/project_index.py was a pure re-export shim and is deleted; its names resolve to their basic_memory.indexing owners. The basic_memory.cloud package facade is unchanged - it re-exports its own submodules, which is a real package surface, not a cross-package barrel. Test changes: - delete tests/indexing/test_indexing_exports.py (barrel contract test) - drop the runtime __all__ assertion test and pure export-surface tests in tests/index/test_project_index_surface.py; keep the behavioral signature/type-hint assertions with concrete imports - repoint monkeypatch targets from barrel attributes to owning modules (basic_memory.index.watch_service.WatchService, basic_memory.index.local_project.*) - entrypoint-name assertions import the constants from runtime.job_payloads / indexing.job_payloads directly 50 src files and 75 test files rewritten. No import cycles surfaced; the latent index <-> services.initialization cycle stays handled by the existing deferred imports, now pointed at concrete submodules. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 7ab8139c92
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| # local writes accept and materialize off the request path, so a | ||
| # one-shot client would otherwise exit before the markdown file is | ||
| # written (the API already reported the note accepted). | ||
| await drain_pending_materializations() |
There was a problem hiding this comment.
When semantic_search_enabled is true and a one-shot local client such as bm tool write-note accepts a note, the router schedules LocalEntityVectorSyncScheduler.schedule_entity_vector_sync() via _background_tasks, but this cleanup path only waits for note materializations before shutting down the DB and closing the event loop. In that CLI context the vector-sync coroutine can be cancelled before it refreshes embeddings, so the markdown/FTS state is current while semantic search remains stale until a later reindex; drain the shared scheduler tasks or run the vector follow-up synchronously before shutdown.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in ecf8040 (commit 3d4d632): run_with_cleanup now drains the shared scheduler task set via a new drain_background_tasks() after pending materializations and before DB shutdown, so one-shot clients wait for vector sync instead of cancelling it at loop close. This also covers relation resolution, which had the same exposure. The drain filters on task state rather than set membership (completed tasks are pruned by a call_soon done-callback that may not have fired yet, and awaiting only done tasks never suspends — membership checks alone would busy-spin), and loops in waves so follow-up tasks scheduled by a draining task are covered too. Regression tests added in tests/cli/test_command_utils.py and tests/services/test_task_scheduler_semantic.py.
A one-shot client like 'bm tool write-note' closes the event loop right after the command returns, cancelling in-flight vector sync and relation resolution scheduled by the write path. The accept path had already reported success, so semantic search and inbound wikilinks stayed stale until an unrelated reindex. Drain the shared scheduler task set in run_with_cleanup after pending materializations and before DB shutdown. The drain filters on task state rather than set membership: completed tasks are pruned by a call_soon done-callback that may not have fired yet, and awaiting only done tasks never suspends, so membership checks alone would busy-spin. Addresses PR #1002 review (discussion r3511346127). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Cloud main writes source="web_v2" for note edits made in the web-v2 UI (basic-memory-cloud #1407), and the source vocabulary now lives here after the parity split. Distinguishing web_v2 from api lets clients tell a genuine web-UI edit apart from api/materialization round-trips. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: ecf8040111
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| request.file_path, | ||
| source=source, | ||
| ) | ||
| except FileOperationError: |
There was a problem hiding this comment.
When a local markdown file is removed after the metadata check but before file_indexer.index_file() reads it, the local markdown indexer unwraps that read failure and re-raises the underlying FileNotFoundError, so this except FileOperationError branch is skipped. That turns an expected watcher/project-index race into a failed index job instead of the intended missing result, leaving noisy failure status and delaying delete reconciliation until a later event/reindex; catch the file-not-found case here as well as FileOperationError.
Useful? React with 👍 / 👎.
1 participant
Why
This PR started as the explicit-session repository refactor and grew into the core half of the local/cloud runtime parity work. The goal is for
basic-memoryto own portable orchestration while cloud and local runtimes provide their own dependency adapters.Paired cloud PR: basicmachines-co/basic-memory-cloud#1345.
Cloud parity document: basic-memory-cloud/docs/local-cloud-runtime-parity.md.
Closes #750.
What Changed
basic_memory.runtimefor storage events, note-content payloads, job identity/status, workflow metadata, materialization, and project/file indexing values.basic_memory.cloudfacades for note-content reads/writes, directory deletes, project deletes, and materialization so cloud/local APIs can share route-facing orchestration.basic_memory.indexas the event-based indexing orchestration layer for storage events, local filesystem watching, project fanout, and inline local processing.basic_memory.indexingwith provider-neutral runners for accepted-note mutation/write flows, file indexing, project indexing, directory deletes, relation/vector follow-ups, and progress/result accounting.basic_memory.synccode and tests out of the active package/test tree so it is reference material only while the new index path becomes the clear file-indexing path.Public Surface
Verification
Not rerun during PR publication. The branch has been validated incrementally during the refactor with focused core tests,
just fast-check,just package-check, and paired cloud checks. Final broad core and cloud checks should run before merge.Review Notes
This PR should be reviewed with the paired cloud PR because the ownership split is intentional: core owns portable orchestration; cloud owns hosted adapters and tenant/runtime wiring.