PR #66269: fix(memory): restore ollama embedding adapter

• Repository: openclaw/openclaw
• Author: vincentkoc
• State: closed | merged: True
• Link: https://github.com/openclaw/openclaw/pull/66269

Description (problem / solution / changelog)

Summary

• Problem: explicit memorySearch.provider: "ollama" regressed because memory-core no longer registers a built-in ollama embedding adapter.
• Why it matters: openclaw memory status, memory search bootstrap, and related explicit Ollama embedding flows fail with Unknown memory embedding provider: ollama.
• What changed: restore the built-in ollama adapter in memory-core, include endpoint-sensitive cache-key data so different Ollama hosts do not collide, add a focused regression test, and add the changelog entry.
• What did NOT change (scope boundary): this PR does not change the broader capability-fallback behavior in src/plugins/memory-embedding-provider-runtime.ts; it fixes the concrete built-in adapter regression on current main.

Change Type (select all)

• Bug fix
• Feature
• Refactor required for the fix
• Docs
• Security hardening
• Chore/infra

Scope (select all touched areas)

• Gateway / orchestration
• Skills / tool execution
• Auth / tokens
• Memory / storage
• Integrations
• API / contracts
• UI / DX
• CI/CD / infra

Linked Issue/PR

• Closes #63429
• Related #66078
• Related #66163
• Related #66112
• Related #63632
• This PR fixes a bug or regression

Root Cause (if applicable)

• Root cause: extensions/memory-core/src/memory/provider-adapters.ts stopped including the built-in ollama adapter, so the runtime registry only contained the other built-ins and explicit ollama resolution failed fast.
• Missing detection / guardrail: there was no focused regression test asserting that built-in Ollama registration stays present.
• Contributing context (if known): the contributor PR also surfaced that cache-key data should include endpoint-specific fields to avoid collisions across multiple Ollama hosts.

Regression Test Plan (if applicable)

• Coverage level that should have caught this:
  • Unit test
  • Seam / integration test
  • End-to-end test
  • Existing coverage already sufficient
• Target test or file: extensions/memory-core/src/memory/index.test.ts
• Scenario the test should lock in: built-in registration includes the ollama adapter with the shared default model.
• Why this is the smallest reliable guardrail: the regression is a missing built-in adapter entry, so the smallest direct check is the registered adapter list.
• Existing test that already covers this (if any): extensions/memory-core/src/memory/manager.mistral-provider.test.ts already covers the shared embedding-provider request path.
• If no new test is added, why not: N/A

User-visible / Behavior Changes

• Explicit Ollama memory embeddings work again through the built-in memory-core adapter path.
• Embedding cache keys now distinguish different Ollama hosts for the same model name.

Diagram (if applicable)

N/A

Security Impact (required)

• New permissions/capabilities? (No)
• Secrets/tokens handling changed? (No)
• New/changed network calls? (No)
• Command/tool execution surface changed? (No)
• Data access scope changed? (No)
• If any Yes, explain risk + mitigation:

Repro + Verification

Environment

• OS: macOS
• Runtime/container: local repo checkout
• Model/provider: Ollama memory embeddings
• Integration/channel (if any): memory-core
• Relevant config (redacted): explicit memorySearch.provider: "ollama"

Steps

1. Run pnpm test:serial extensions/memory-core/src/memory/index.test.ts -t "registers the builtin ollama embedding provider"
2. Run pnpm test:serial extensions/memory-core/src/memory/manager.mistral-provider.test.ts

Expected

• Built-in Ollama registration is present and the focused memory-core test slice passes.

Actual

• Before this PR, explicit Ollama registration was missing from the built-in adapter list.

Evidence

• Failing test/log before + passing after
• Trace/log snippets
• Screenshot/recording
• Perf numbers (if relevant)

Human Verification (required)

• Verified scenarios: focused built-in registration test and manager.mistral-provider test both passed locally.
• Edge cases checked: cache-key data now includes baseUrl and sanitized headers so different Ollama endpoints do not share cached embeddings.
• What you did not verify: a full repo-wide test lane; a broader serial run in this sparse maintainer worktree hit unrelated existing contract failures outside the touched memory-core surface.

Review Conversations

• I replied to or resolved every bot review conversation I addressed in this PR.
• I left unresolved only the conversations that still need reviewer or maintainer judgment.

Compatibility / Migration

• Backward compatible? (Yes)
• Config/env changes? (No)
• Migration needed? (No)
• If yes, exact upgrade steps:

Risks and Mitigations

• Risk: none beyond restoring a missing built-in adapter and tightening its cache identity.
  • Mitigation: focused regression coverage and preserved adapter semantics from the existing Ollama extension path.

Changed files

• CHANGELOG.md (modified, +1/-0)
• extensions/memory-core/src/memory/index.test.ts (modified, +17/-1)
• extensions/memory-core/src/memory/provider-adapters.ts (modified, +29/-0)