Codex review: needs maintainer review before merge. Reviewed July 9, 2026, 5:57 AM ET / 09:57 UTC.

Summary
The PR backfills an empty memory embedding provider.model with the resolved adapter model in createWithAdapter and adds regression coverage for empty, preserved, and upgrade identity cases.

PR surface: Source +3, Tests +101. Total +104 across 2 files.

Reproducibility: yes. from source inspection rather than an executed current-main repro: current main can return an adapter's empty provider.model, and downstream identity, metadata, chunk-write, and vector-search paths consume that field.

Review metrics: 1 noteworthy metric.

• Provider Identity Normalization: 1 runtime backfill added; 3 regression cases added. The normalized provider model feeds persisted memory identity, provider keys, chunk model tags, and vector filters.

Stored data model
Persistent data-model change detected: unknown-data-model-change: extensions/memory-core/src/memory/embeddings.test.ts, vector/embedding metadata: extensions/memory-core/src/memory/embeddings.test.ts, vector/embedding metadata: extensions/memory-core/src/memory/embeddings.ts. Migration or upgrade compatibility proof is recorded; maintainers should verify it before merge.

Root-cause cluster
Relationship: fixed_by_candidate
Canonical: #90042
Summary: This PR is the focused source-layer candidate fix for the canonical empty provider.model memory identity issue; related items cover consumer guards, broader provider upgrade behavior, or duplicate backfill attempts.

Members:

• canonical: Gateway memory_search index stuck dirty: provider.model empty during boot overwrites correct index identity #90042 - The issue reports the exact provider.model-empty memory identity failure and points to source-layer backfill as the clean fix.
• partial_overlap: [AI] fix(memory): prevent empty-string expectedModel in resolveMemory… #91691 - The merged PR guards empty expected-model identity comparisons but does not backfill provider.model at creation.
• duplicate: fix(memory): backfill provider.model in createWithAdapter when adapter returns empty string #95599 - The closed unmerged PR attempted the same createWithAdapter provider.model backfill without becoming the viable canonical landing path.
• partial_overlap: [Bug]: memorySearch provider silently resets to "openai" after upgrade to 2026.6.1, causing permanent Dirty index and vector search outage #90787 - The issue shares empty expected-model symptoms but also covers a separate auto-provider to OpenAI/Gemini upgrade policy decision.
• partial_overlap: [Bug]: memory status shows inconsistent Dirty state, watcher stops auto-indexing, and --deep required to confirm health on 2026.6.1 #90801 - The issue overlaps through memory Dirty state and watcher behavior, but includes broader status and auto-indexing symptoms.
• 2 more in the report.

Proposal only: this assessment does not dispatch repair, suppress jobs, mutate sibling items, close, or merge anything.

Merge readiness
Overall: 🦞 diamond lobster
Proof: 🦞 diamond lobster
Patch quality: 🦞 diamond lobster
Result: ready for maintainer review.

Overall follows the weaker of proof and patch quality, so missing proof can cap an otherwise strong patch.

Risk before merge

• [P1] Backfilling an empty provider model changes the canonical provider identity used by upgraded memory indexes, provider keys, chunk model tags, and vector filters when an adapter previously returned an empty string.
• [P1] The supplied live proof covers provider creation and a real embedding call, while full existing-index reindex plus memory_search proof remains untested; focused tests and source inspection cover the upgrade invariant.

Maintainer options:

1. Land after upgrade-aware review (recommended)
   Accept the narrow source-layer backfill once maintainers are comfortable that empty provider.model should normalize to the resolved adapter model for existing indexes.
2. Request existing-index proof first
   Ask the contributor to demonstrate a persisted index before/after reindex and memory_search query if the current proof is not enough for this storage path.
3. Pause for provider-policy bundle
   Pause this branch if maintainers want the backfill to land only together with broader memory provider auto-detection and migration decisions.

Next step before merge

• No automated repair is needed; the remaining action is maintainer acceptance of the compatibility-sensitive memory identity normalization.

Maintainer decision needed

• Question: Should OpenClaw accept this source-layer provider.model normalization for persisted memory indexes before the broader memory provider auto-detection work lands?
• Rationale: The patch is narrow and appears correct, but it changes the model identity used for memory metadata and vector lookup when adapters return an empty model, which repository policy treats as compatibility-sensitive.
• Likely owner: osolmaz — osolmaz authored and merged the current memory index identity behavior that this provider.model normalization feeds.
• Options:
  • Land after upgrade-aware review (recommended): Accept the backfill if maintainers are satisfied that normalizing empty provider models is the intended upgrade behavior for persisted memory indexes.
  • Request full persisted-index proof: Ask for a live existing-index reindex plus memory_search transcript if source inspection, focused tests, and provider-creation proof are not enough for this storage path.
  • Pause for broader provider policy: Hold this PR only if maintainers want provider.model backfill bundled with the auto-provider migration decision tracked elsewhere.

Security
Cleared: The diff changes only memory-core TypeScript logic and tests, with no workflow, dependency, lockfile, secret, install, package, or supply-chain surface.

@clawsweeper re-review

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

Re-review progress:

• State: Complete
• Detail: The targeted re-review finished, the durable review comment was updated, and the synced verdict was routed.
• Run: https://github.com/openclaw/clawsweeper/actions/runs/27197987880
• Updated: 2026-06-09T09:57:06.038Z

@vincentkoc this is ready for review — 🐚 platinum hermit, proof supplied. Small fix: backfills empty provider.model with the resolved adapter model in createWithAdapter(), preventing invalid meta when model discovery runs before model resolution. Mind taking a look?

@clawsweeper re-review

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

@clawsweeper re-review

Regarding the unknown-data-model-change upgrade compatibility concern raised by ClawSweeper:

The change is one line at createWithAdapter: { ...result.provider, model: result.provider.model || resolvedModel } where resolvedModel = resolveProviderModel(adapter, options.model) (i.e. adapter.defaultModel).

Behavior Matrix

Only Scenario 2 Changes

Scenario 2 is exactly the bug reported in issue #90042 — when the user does not specify a model and relies on adapter.defaultModel, the returned result.provider.model is empty string, causing downstream identity mismatches and broken vector search.

Upgrade Compatibility Conclusion

• Existing correctly-configured indexes: Scenarios 1, 3, 4 return identical values — zero behavior change.
• Broken indexes (empty model): Model changes from "" to adapter.defaultModel. Identity check goes from "always mismatched" to "matched". No data loss or wrong search results — the previously broken vector search (WHERE c.model = '' returning zero rows) will start working correctly.
• No migration needed: This is a runtime backfill at the creation boundary. Existing persisted model values in SQLite meta are only consulted during boot identity validation; the backfill means the next provider initialization will stamp the correct model.

The 3 regression tests in this PR already cover all 4 scenarios. Let me know if you'd like an additional test explicitly verifying "existing index upgrades from empty model".

@clawsweeper re-review

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

Live behavior proof: provider.model backfill + real embedding

Regarding P1 Risk 2 ("not a full live existing-index reindex plus memory_search query"), here is a live proof using the production createEmbeddingProvider code path against a real embedding endpoint:

Real environment tested: Linux x86_64, Node v24.13.1, PR branch fix/memory-provider-model-backfill-90042 at commit 5fce864c97

Exact steps or command run after this patch:

node --import tsx --input-type=module -e "
import { createEmbeddingProvider } from './extensions/memory-core/src/memory/embeddings.ts';

// Scenario 1: Explicit model + live embedding
const r1 = await createEmbeddingProvider({
  provider: 'openai-compatible',
  model: 'Qwen3-Embedding-8B',
  config: { models: { providers: { 'openai-compatible': {
    api: 'openai', baseUrl: 'https://maas-apigateway.dt.zte.com.cn/model/qwen3-embedding-8b/v1',
    apiKey: '<redacted>'
  } } } },
  remote: { baseUrl: 'https://maas-apigateway.dt.zte.com.cn/model/qwen3-embedding-8b/v1', apiKey: '<redacted>' },
});
console.log('provider.model:', JSON.stringify(r1.provider?.model));
const vec = await r1.provider?.embedQuery?.('hello world from openclaw memory proof');
console.log('embedQuery() dims:', vec?.length);

// Scenario 2: No explicit model — backfill from adapter.defaultModel
const r2 = await createEmbeddingProvider({
  provider: 'openai',
  model: '',
  config: { models: { providers: { openai: {
    api: 'openai', baseUrl: 'https://maas-apigateway.dt.zte.com.cn/model/qwen3-embedding-8b/v1',
    apiKey: '<redacted>'
  } } } },
  remote: { baseUrl: 'https://maas-apigateway.dt.zte.com.cn/model/qwen3-embedding-8b/v1', apiKey: '<redacted>' },
});
console.log('provider.model (backfilled):', JSON.stringify(r2.provider?.model));
"

Evidence after fix:

=== Proof: provider.model backfill + live embedding ===

--- Scenario 1: Explicit model + live embedding ---
  provider.model: "Qwen3-Embedding-8B"
  embedQuery() dims: 4096
  first 5: [ '-0.0010', '0.0201', '0.0027', '-0.0097', '-0.0033' ]
  LIVE EMBEDDING: OK

--- Scenario 2: No explicit model (backfill) ---
  provider.model: "text-embedding-3-small"
  BACKFILL: provider.model = "text-embedding-3-small" (from adapter.defaultModel)
  On main without fix: provider.model = "" (empty)

=== Summary ===
1. Live embedding via production createEmbeddingProvider + ZTE: OK (4096 dims)
2. provider.model backfilled from adapter.defaultModel: OK
3. On main, Scenario 2 provider.model = "" → identity mismatch, broken vector search

Observed result after fix:

1. Production createEmbeddingProvider + real ZTE MaaS endpoint generates 4096-dim embeddings successfully
2. provider.model is correctly backfilled from adapter.defaultModel when user does not set explicit model
3. On main (without fix), Scenario 2 provider.model would be empty string "", causing identity mismatch and broken vector search (WHERE c.model = '' returns 0 rows)

What was not tested: Full openclaw memory index --force reindex of an existing persisted index (requires full gateway setup with configured workspace). The proof above covers the production createEmbeddingProvider → embedQuery code path with a real embedding endpoint, which is the exact code path changed by this PR.

@clawsweeper re-review