feat: search quality improvements + Groq rate limiter#68
Merged
Conversation
Search quality (tasks A2, A4, A5, A6): - FTS5 expansion: index summary, tags, resolved_query alongside content for better keyword matches on enrichment metadata - Entity-aware routing: auto-detect KG entity names in queries and route to kg_hybrid_search for combined chunk + fact retrieval - Post-RRF reranking: boost results by importance (up to 1.5x) and recency (30-day half-life exponential decay) - Fix format=format bug: default search path was passing Python builtin format() instead of detail parameter Enrichment fix (task A1): - Add 2s rate limit between Groq API calls to avoid 429 spam on free tier - Prioritize Claude Code chunks over YouTube in enrichment queue - Source priority ordering in get_unenriched_chunks 20 new tests, 695 total passing, 0 regressions. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
There was a problem hiding this comment.
Your free trial has ended. If you'd like to continue receiving code reviews, you can add a payment method here.
|
Warning Rate limit exceeded
⌛ How to resolve this issue?After the wait time has elapsed, a review can be triggered using the We recommend that you space out your commits to avoid hitting the rate limit. 🚦 How do rate limits work?CodeRabbit enforces hourly rate limits for each developer per organization. Our paid plans have higher rate limits than the trial, open-source and free plans. In all cases, we re-allow further reviews after a brief timeout. Please see our FAQ for further information. 📒 Files selected for processing (4)
📝 WalkthroughWalkthroughAdds entity-aware routing to search (detects KG entities and routes to kg_hybrid_search), expands FTS5 to index enriched fields, applies post-RRF boosting by importance and recency, prioritizes chunk sources, and enforces Groq API rate limiting; accompanying tests validate these behaviors. Changes
Sequence Diagram(s)sequenceDiagram
actor Client
participant SearchHandler as search_handler\n(_brain_search/_search)
participant Detector as _detect_entities
participant KGDB as KG Tables\n(kg_entities, kg_entities_fts)
participant Vector as vector_store\n(kg_hybrid_search)
Client->>SearchHandler: submit query
SearchHandler->>Detector: extract candidate entities (bigrams, caps)
Detector->>KGDB: lookup exact / FTS matches
alt entity found
KGDB-->>Detector: entity record(s)
Detector-->>SearchHandler: matched entity list
SearchHandler->>Vector: kg_hybrid_search(query, entity_name, routing params)
Vector-->>SearchHandler: chunks + KG facts (formatted)
SearchHandler-->>Client: entity-aware results
else no entity
KGDB-->>Detector: no match
Detector-->>SearchHandler: []
SearchHandler->>SearchHandler: fallback to standard hybrid_search/_search path
SearchHandler-->>Client: standard search results
end
Estimated Code Review Effort🎯 4 (Complex) | ⏱️ ~45 minutes Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 3✅ Passed checks (3 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 4
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@src/brainlayer/mcp/search_handler.py`:
- Around line 225-231: _brain_search currently skips _search and calls
kg_hybrid_search without forwarding active filters, causing different semantics;
modify the kg_hybrid_search call in _brain_search to pass through the same
filter parameters that _search uses (e.g., content_type, source, tag, intent,
importance_min, date_from, date_to, sentiment and any other active filter keys)
by extracting them from the incoming filters/context and supplying them as named
arguments to kg_hybrid_search (keep existing args like query_embedding,
query_text, n_results, entity_name, project_filter) so entity-aware routes honor
the same filters as non-entity searches.
In `@src/brainlayer/pipeline/enrichment.py`:
- Around line 95-97: The shared _groq_last_call and GROQ_RATE_LIMIT_DELAY
rate-limiting logic is not thread-safe; protect reads/updates to _groq_last_call
with a lock and perform the wait inside the critical section to enforce spacing
across threads. Add a module-level threading.Lock (e.g., _groq_lock) and wrap
the check/update/sleep sequence that uses _groq_last_call and
GROQ_RATE_LIMIT_DELAY (the same logic around the GROQ call present near the top
with _groq_last_call and the duplicate block around lines ~535-547) so only one
thread can compute elapsed, sleep the required remaining time, update
_groq_last_call using time.monotonic(), then release the lock before making the
API request.
In `@tests/test_groq_rate_limit.py`:
- Around line 56-69: The test test_rate_limit_delay_env_override currently only
checks os.environ and never validates the module-level constant; update it to
reload the enrichment module and assert the module constant matches the env
value: after setting os.environ["BRAINLAYER_GROQ_RATE_DELAY"] = "5.0", call
importlib.reload(enrichment) and assert enrichment.GROQ_RATE_LIMIT_DELAY (or the
actual module-level constant name in brainlayer.pipeline.enrichment) equals 5.0,
keeping the existing try/finally that restores the original env.
In `@tests/test_search_quality.py`:
- Around line 329-360: The test test_entity_routing_calls_kg_hybrid_search is
calling store.kg_hybrid_search directly so it never exercises the routing logic
in _brain_search; change the test to call the public brain search entrypoint
(e.g., store._brain_search or the exposed brain_search wrapper) with the same
setup and mock the kg_hybrid_search method (or spy on store.kg_hybrid_search) to
assert it was invoked when the query contains the entity; keep the existing DB
fixtures (kg_entities, kg_entity_chunks, chunk) and ensure the test passes
query_text="Michal Cohen meetings" and query_embedding=mock_embed(...) into the
brain search call, then assert that store.kg_hybrid_search was called (and/or
that the brain search returns the entity-linked chunk) instead of calling
kg_hybrid_search directly.
ℹ️ Review info
Configuration used: Organization UI
Review profile: ASSERTIVE
Plan: Pro
📒 Files selected for processing (7)
src/brainlayer/mcp/search_handler.pysrc/brainlayer/pipeline/enrichment.pysrc/brainlayer/search_repo.pysrc/brainlayer/session_repo.pysrc/brainlayer/vector_store.pytests/test_groq_rate_limit.pytests/test_search_quality.py
📜 Review details
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (3)
- GitHub Check: test (3.11)
- GitHub Check: test (3.12)
- GitHub Check: test (3.13)
🧰 Additional context used
📓 Path-based instructions (6)
**/*test*.py
📄 CodeRabbit inference engine (CLAUDE.md)
Run tests with
pytest
Files:
tests/test_groq_rate_limit.pytests/test_search_quality.py
src/brainlayer/**/*.py
📄 CodeRabbit inference engine (CLAUDE.md)
Use Python package structure with Typer CLI located in
src/brainlayer/
Files:
src/brainlayer/pipeline/enrichment.pysrc/brainlayer/vector_store.pysrc/brainlayer/search_repo.pysrc/brainlayer/session_repo.pysrc/brainlayer/mcp/search_handler.py
**/enrichment*.py
📄 CodeRabbit inference engine (CLAUDE.md)
**/enrichment*.py: Use MLX (Qwen2.5-Coder-14B-Instruct-4bit) as primary enrichment backend on Apple Silicon (port 8080) with Ollama (glm-4.7-flash) fallback on port 11434
Auto-switch enrichment backend from MLX to Ollama after 3 consecutive MLX failures; allow override withBRAINLAYER_ENRICH_BACKENDenvironment variable
Files:
src/brainlayer/pipeline/enrichment.py
src/**/*.py
📄 CodeRabbit inference engine (CLAUDE.md)
Lint and format Python code using
ruff check src/andruff format src/
Files:
src/brainlayer/pipeline/enrichment.pysrc/brainlayer/vector_store.pysrc/brainlayer/search_repo.pysrc/brainlayer/session_repo.pysrc/brainlayer/mcp/search_handler.py
**/vector_store.py
📄 CodeRabbit inference engine (CLAUDE.md)
Use APSW with sqlite-vec for vector storage (in
vector_store.py)
Files:
src/brainlayer/vector_store.py
**/mcp/**/*.py
📄 CodeRabbit inference engine (CLAUDE.md)
Implement MCP tools:
brain_search,brain_store,brain_recall,brain_entity,brain_expand,brain_update,brain_digest,brain_get_personwith legacybrainlayer_*aliases
Files:
src/brainlayer/mcp/search_handler.py
🧠 Learnings (4)
📚 Learning: 2026-03-02T21:14:04.731Z
Learnt from: CR
Repo: EtanHey/brainlayer PR: 0
File: CLAUDE.md:0-0
Timestamp: 2026-03-02T21:14:04.731Z
Learning: Applies to **/enrichment*.py : Auto-switch enrichment backend from MLX to Ollama after 3 consecutive MLX failures; allow override with `BRAINLAYER_ENRICH_BACKEND` environment variable
Applied to files:
src/brainlayer/pipeline/enrichment.py
📚 Learning: 2026-03-02T21:14:04.731Z
Learnt from: CR
Repo: EtanHey/brainlayer PR: 0
File: CLAUDE.md:0-0
Timestamp: 2026-03-02T21:14:04.731Z
Learning: Applies to scripts/**/*.py : Drop FTS triggers (especially `chunks_fts_delete`) before bulk deletes and recreate after — FTS triggers cause massive performance degradation on large datasets
Applied to files:
src/brainlayer/vector_store.py
📚 Learning: 2026-03-02T21:14:04.731Z
Learnt from: CR
Repo: EtanHey/brainlayer PR: 0
File: CLAUDE.md:0-0
Timestamp: 2026-03-02T21:14:04.731Z
Learning: Applies to scripts/**/*.py : Never delete from `chunks` table while FTS trigger is active on large datasets
Applied to files:
src/brainlayer/vector_store.py
📚 Learning: 2026-03-02T21:14:04.731Z
Learnt from: CR
Repo: EtanHey/brainlayer PR: 0
File: CLAUDE.md:0-0
Timestamp: 2026-03-02T21:14:04.731Z
Learning: Applies to **/mcp/**/*.py : Implement MCP tools: `brain_search`, `brain_store`, `brain_recall`, `brain_entity`, `brain_expand`, `brain_update`, `brain_digest`, `brain_get_person` with legacy `brainlayer_*` aliases
Applied to files:
src/brainlayer/mcp/search_handler.py
🧬 Code graph analysis (4)
tests/test_groq_rate_limit.py (2)
src/brainlayer/pipeline/enrichment.py (1)
call_groq(525-581)src/brainlayer/session_repo.py (1)
get_unenriched_chunks(21-86)
tests/test_search_quality.py (4)
src/brainlayer/vector_store.py (1)
close(689-698)src/brainlayer/mcp/search_handler.py (3)
_embed(717-718)_embed(752-753)_detect_entities(29-97)src/brainlayer/search_repo.py (1)
hybrid_search(335-580)src/brainlayer/kg_repo.py (1)
kg_hybrid_search(769-818)
src/brainlayer/search_repo.py (1)
src/brainlayer/dashboard/search.py (1)
score(53-78)
src/brainlayer/mcp/search_handler.py (5)
src/brainlayer/vector_store.py (1)
_read_cursor(546-548)src/brainlayer/_helpers.py (1)
_escape_fts5_query(35-54)src/brainlayer/mcp/_shared.py (4)
_get_vector_store(16-24)_normalize_project_name(92-142)_get_embedding_model(27-34)_build_compact_result(239-250)src/brainlayer/cli/__init__.py (1)
_normalize_project_name(140-162)src/brainlayer/kg_repo.py (1)
kg_hybrid_search(769-818)
🔇 Additional comments (4)
src/brainlayer/session_repo.py (1)
54-67: Source-priority ordering and source propagation look correct.The prioritization logic and returned
sourcefield are consistent with the intended enrichment queue behavior.Also applies to: 83-83
src/brainlayer/search_repo.py (1)
542-568: Post-RRF reranking block is robust and well-guarded.The boost application is bounded and gracefully skips malformed temporal metadata without breaking ranking.
src/brainlayer/vector_store.py (1)
186-234: FTS5 schema expansion and migration/backfill path are consistent.The rebuild detection, trigger synchronization, and full-column backfill align with the expanded FTS search surface.
Also applies to: 519-526
src/brainlayer/mcp/search_handler.py (1)
301-301: Good fix on the default fallback argument forwarding.Switching to
detail=detailcloses theformat=formatbug in the default path.
Comment on lines
+225
to
+231
| kg_results = store.kg_hybrid_search( | ||
| query_embedding=query_embedding, | ||
| query_text=query, | ||
| n_results=num_results, | ||
| entity_name=entity_name, | ||
| project_filter=normalized_project, | ||
| ) |
There was a problem hiding this comment.
Entity-aware route drops active search filters.
When entity detection triggers, _brain_search bypasses _search and calls kg_hybrid_search without forwarding filters (content_type, source, tag, intent, importance_min, date_from/date_to, sentiment). This changes query semantics unexpectedly.
🔧 Suggested fix
- kg_results = store.kg_hybrid_search(
+ if source == "all":
+ source_filter = None
+ elif source:
+ source_filter = source
+ else:
+ source_filter = None
+
+ kg_results = store.kg_hybrid_search(
query_embedding=query_embedding,
query_text=query,
n_results=num_results,
entity_name=entity_name,
project_filter=normalized_project,
+ content_type_filter=content_type,
+ source_filter=source_filter,
+ tag_filter=tag,
+ intent_filter=intent,
+ importance_min=importance_min,
+ date_from=date_from,
+ date_to=date_to,
+ sentiment_filter=sentiment,
)🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@src/brainlayer/mcp/search_handler.py` around lines 225 - 231, _brain_search
currently skips _search and calls kg_hybrid_search without forwarding active
filters, causing different semantics; modify the kg_hybrid_search call in
_brain_search to pass through the same filter parameters that _search uses
(e.g., content_type, source, tag, intent, importance_min, date_from, date_to,
sentiment and any other active filter keys) by extracting them from the incoming
filters/context and supplying them as named arguments to kg_hybrid_search (keep
existing args like query_embedding, query_text, n_results, entity_name,
project_filter) so entity-aware routes honor the same filters as non-entity
searches.
Comment on lines
+329
to
+360
| def test_entity_routing_calls_kg_hybrid_search(self, store, mock_embed): | ||
| """When entities detected, brain_search should use kg_hybrid_search.""" | ||
| # Insert entity | ||
| cursor = store.conn.cursor() | ||
| cursor.execute( | ||
| """INSERT INTO kg_entities (id, entity_type, name, metadata, created_at) | ||
| VALUES (?, ?, ?, '{}', datetime('now'))""", | ||
| ("ent-route-1", "person", "Michal Cohen"), | ||
| ) | ||
| cursor.execute( | ||
| "INSERT INTO kg_entities_fts (name, metadata, entity_id) VALUES (?, '{}', ?)", | ||
| ("Michal Cohen", "ent-route-1"), | ||
| ) | ||
| # Insert a chunk linked to this entity | ||
| _insert_chunk( | ||
| store, "ent-chunk-1", | ||
| content="Michal Cohen prefers morning meetings before 10am", | ||
| importance=7.0, | ||
| created_at="2026-03-01T00:00:00", | ||
| ) | ||
| cursor.execute( | ||
| "INSERT INTO kg_entity_chunks (entity_id, chunk_id, relevance) VALUES (?, ?, ?)", | ||
| ("ent-route-1", "ent-chunk-1", 0.95), | ||
| ) | ||
|
|
||
| # The search should find the entity-linked chunk | ||
| results = store.kg_hybrid_search( | ||
| query_embedding=mock_embed("Michal Cohen meetings"), | ||
| query_text="Michal Cohen meetings", | ||
| n_results=5, | ||
| entity_name="Michal Cohen", | ||
| ) |
There was a problem hiding this comment.
test_entity_routing_calls_kg_hybrid_search does not actually test routing.
It calls store.kg_hybrid_search(...) directly, so _brain_search entity-detection/routing logic is never exercised. The test name and coverage intent are currently misleading.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@tests/test_search_quality.py` around lines 329 - 360, The test
test_entity_routing_calls_kg_hybrid_search is calling store.kg_hybrid_search
directly so it never exercises the routing logic in _brain_search; change the
test to call the public brain search entrypoint (e.g., store._brain_search or
the exposed brain_search wrapper) with the same setup and mock the
kg_hybrid_search method (or spy on store.kg_hybrid_search) to assert it was
invoked when the query contains the entity; keep the existing DB fixtures
(kg_entities, kg_entity_chunks, chunk) and ensure the test passes
query_text="Michal Cohen meetings" and query_embedding=mock_embed(...) into the
brain search call, then assert that store.kg_hybrid_search was called (and/or
that the brain search returns the entity-linked chunk) instead of calling
kg_hybrid_search directly.
There was a problem hiding this comment.
Actionable comments posted: 2
♻️ Duplicate comments (3)
tests/test_groq_rate_limit.py (1)
53-69:⚠️ Potential issue | 🟠 MajorValidate the loaded delay constant, not just the env var string.
Line 63 only checks
os.environ, so this test still passes ifbrainlayer.pipeline.enrichment.GROQ_RATE_LIMIT_DELAYwas already loaded with a stale value. Reload the module and assert the constant directly.🔧 Suggested fix
def test_rate_limit_delay_env_override(self): """BRAINLAYER_GROQ_RATE_DELAY env var should override the default delay.""" import os + from importlib import reload + from brainlayer.pipeline import enrichment original = os.environ.get("BRAINLAYER_GROQ_RATE_DELAY") try: os.environ["BRAINLAYER_GROQ_RATE_DELAY"] = "5.0" - # Need to reload to pick up env var - # Check the module-level constant respects the env var - # (This tests the env var mechanism, not the reloaded value) - assert float(os.environ["BRAINLAYER_GROQ_RATE_DELAY"]) == 5.0 + enrichment = reload(enrichment) + assert enrichment.GROQ_RATE_LIMIT_DELAY == 5.0 finally: if original is None: os.environ.pop("BRAINLAYER_GROQ_RATE_DELAY", None) else: os.environ["BRAINLAYER_GROQ_RATE_DELAY"] = original + reload(enrichment)🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@tests/test_groq_rate_limit.py` around lines 53 - 69, The test currently only asserts the environment variable string instead of the module constant; modify test_rate_limit_delay_env_override to set os.environ["BRAINLAYER_GROQ_RATE_DELAY"]="5.0", reload the enrichment module (e.g. import importlib; importlib.reload(brainlayer.pipeline.enrichment) or import the module after setting the env), then assert that brainlayer.pipeline.enrichment.GROQ_RATE_LIMIT_DELAY == 5.0 (or float(...) equals 5.0); ensure you restore the original env var in the finally block as before.src/brainlayer/mcp/search_handler.py (1)
230-236:⚠️ Potential issue | 🟠 MajorEntity-aware branch drops active filters before
kg_hybrid_search.Line 230 routes through
kg_hybrid_search, but it does not forward current filters (content_type,source,tag,intent,importance_min,date_from,date_to,sentiment). This changes query semantics versus the default_searchpath.🔧 Suggested fix
- kg_results = store.kg_hybrid_search( + if source == "all": + source_filter = None + elif source: + source_filter = source + else: + source_filter = None + + kg_results = store.kg_hybrid_search( query_embedding=query_embedding, query_text=query, n_results=num_results, entity_name=entity_name, project_filter=normalized_project, + content_type_filter=content_type, + source_filter=source_filter, + tag_filter=tag, + intent_filter=intent, + importance_min=importance_min, + date_from=date_from, + date_to=date_to, + sentiment_filter=sentiment, )🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/brainlayer/mcp/search_handler.py` around lines 230 - 236, The entity-aware branch calls store.kg_hybrid_search without forwarding the active filter parameters, which changes semantics compared to the default _search path; update the store.kg_hybrid_search call in search_handler.py (the block invoking store.kg_hybrid_search with query_embedding, query_text/query, n_results/num_results, entity_name, project_filter/normalized_project) to pass the current filter arguments (content_type, source, tag, intent, importance_min, date_from, date_to, sentiment) so that kg_hybrid_search receives the same filters as _search and preserves query semantics.tests/test_search_quality.py (1)
360-394:⚠️ Potential issue | 🟠 MajorThis test does not verify routing through
_brain_search.Line 387 calls
store.kg_hybrid_search(...)directly, so entity-detection/routing logic inbrainlayer.mcp.search_handler._brain_searchis never exercised. Rename or, preferably, route through_brain_searchand assertkg_hybrid_searchwas invoked.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@tests/test_search_quality.py` around lines 360 - 394, The test currently calls kg_hybrid_search directly so routing logic in _brain_search is not exercised; modify the test to call brainlayer.mcp.search_handler._brain_search (or the public entry that triggers it) with the same inputs (query_text and query_embedding) and use a spy/patch on the Store.kg_hybrid_search method (reference symbol: kg_hybrid_search) to assert it was invoked and returned the expected results; ensure the existing DB setup (kg_entities, kg_entities_fts, kg_entity_chunks, and _insert_chunk) remains the same and replace the direct call to store.kg_hybrid_search with a call to _brain_search, then assert the spy was called and that the returned payload includes "chunks" and "facts".
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@tests/test_groq_rate_limit.py`:
- Around line 20-45: The test relies on global mutable state
enrichment._groq_last_call so reset or patch that state at the start of
test_call_groq_sleeps_between_calls to isolate timing (e.g., set
enrichment._groq_last_call = 0 or patch it via monkeypatch) before invoking
enrichment.call_groq three times; update the test to assign or patch
enrichment._groq_last_call inside the with-block (or at test start) so previous
tests can't affect the timing assertions.
- Around line 102-108: The test currently only checks ordering when both
"claude_code" and "youtube" appear, allowing a vacuous pass if either is
missing; update the test in tests/test_groq_rate_limit.py around the use of
store.get_unenriched_chunks and the sources list to first assert both sources
are present (e.g., assert "claude_code" in sources and assert "youtube" in
sources) and only then compute cc_idx = sources.index("claude_code") and yt_idx
= sources.index("youtube") and assert cc_idx < yt_idx to enforce the priority
ordering.
---
Duplicate comments:
In `@src/brainlayer/mcp/search_handler.py`:
- Around line 230-236: The entity-aware branch calls store.kg_hybrid_search
without forwarding the active filter parameters, which changes semantics
compared to the default _search path; update the store.kg_hybrid_search call in
search_handler.py (the block invoking store.kg_hybrid_search with
query_embedding, query_text/query, n_results/num_results, entity_name,
project_filter/normalized_project) to pass the current filter arguments
(content_type, source, tag, intent, importance_min, date_from, date_to,
sentiment) so that kg_hybrid_search receives the same filters as _search and
preserves query semantics.
In `@tests/test_groq_rate_limit.py`:
- Around line 53-69: The test currently only asserts the environment variable
string instead of the module constant; modify test_rate_limit_delay_env_override
to set os.environ["BRAINLAYER_GROQ_RATE_DELAY"]="5.0", reload the enrichment
module (e.g. import importlib; importlib.reload(brainlayer.pipeline.enrichment)
or import the module after setting the env), then assert that
brainlayer.pipeline.enrichment.GROQ_RATE_LIMIT_DELAY == 5.0 (or float(...)
equals 5.0); ensure you restore the original env var in the finally block as
before.
In `@tests/test_search_quality.py`:
- Around line 360-394: The test currently calls kg_hybrid_search directly so
routing logic in _brain_search is not exercised; modify the test to call
brainlayer.mcp.search_handler._brain_search (or the public entry that triggers
it) with the same inputs (query_text and query_embedding) and use a spy/patch on
the Store.kg_hybrid_search method (reference symbol: kg_hybrid_search) to assert
it was invoked and returned the expected results; ensure the existing DB setup
(kg_entities, kg_entities_fts, kg_entity_chunks, and _insert_chunk) remains the
same and replace the direct call to store.kg_hybrid_search with a call to
_brain_search, then assert the spy was called and that the returned payload
includes "chunks" and "facts".
ℹ️ Review info
Configuration used: Organization UI
Review profile: ASSERTIVE
Plan: Pro
📒 Files selected for processing (3)
src/brainlayer/mcp/search_handler.pytests/test_groq_rate_limit.pytests/test_search_quality.py
📜 Review details
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (2)
- GitHub Check: test (3.12)
- GitHub Check: test (3.11)
🧰 Additional context used
📓 Path-based instructions (4)
**/*test*.py
📄 CodeRabbit inference engine (CLAUDE.md)
Run tests with
pytest
Files:
tests/test_groq_rate_limit.pytests/test_search_quality.py
src/brainlayer/**/*.py
📄 CodeRabbit inference engine (CLAUDE.md)
Use Python package structure with Typer CLI located in
src/brainlayer/
Files:
src/brainlayer/mcp/search_handler.py
**/mcp/**/*.py
📄 CodeRabbit inference engine (CLAUDE.md)
Implement MCP tools:
brain_search,brain_store,brain_recall,brain_entity,brain_expand,brain_update,brain_digest,brain_get_personwith legacybrainlayer_*aliases
Files:
src/brainlayer/mcp/search_handler.py
src/**/*.py
📄 CodeRabbit inference engine (CLAUDE.md)
Lint and format Python code using
ruff check src/andruff format src/
Files:
src/brainlayer/mcp/search_handler.py
🧠 Learnings (2)
📚 Learning: 2026-03-02T21:14:04.731Z
Learnt from: CR
Repo: EtanHey/brainlayer PR: 0
File: CLAUDE.md:0-0
Timestamp: 2026-03-02T21:14:04.731Z
Learning: Applies to **/enrichment*.py : Auto-switch enrichment backend from MLX to Ollama after 3 consecutive MLX failures; allow override with `BRAINLAYER_ENRICH_BACKEND` environment variable
Applied to files:
tests/test_groq_rate_limit.py
📚 Learning: 2026-03-02T21:14:04.731Z
Learnt from: CR
Repo: EtanHey/brainlayer PR: 0
File: CLAUDE.md:0-0
Timestamp: 2026-03-02T21:14:04.731Z
Learning: Applies to **/mcp/**/*.py : Implement MCP tools: `brain_search`, `brain_store`, `brain_recall`, `brain_entity`, `brain_expand`, `brain_update`, `brain_digest`, `brain_get_person` with legacy `brainlayer_*` aliases
Applied to files:
src/brainlayer/mcp/search_handler.py
🧬 Code graph analysis (2)
tests/test_groq_rate_limit.py (2)
src/brainlayer/pipeline/enrichment.py (1)
call_groq(525-581)src/brainlayer/session_repo.py (1)
get_unenriched_chunks(21-86)
tests/test_search_quality.py (4)
src/brainlayer/vector_store.py (1)
close(689-698)src/brainlayer/mcp/search_handler.py (3)
_embed(726-727)_embed(761-762)_detect_entities(29-102)src/brainlayer/search_repo.py (1)
hybrid_search(335-580)src/brainlayer/kg_repo.py (1)
kg_hybrid_search(769-818)
…rd, tests - Rate limiter: wrap _groq_last_call check/update in threading.Lock - Entity routing: skip kg_hybrid_search when active filters present (kg_hybrid_search doesn't support content_type/source/tag/etc.) - Test: env-override now reloads module and asserts constant value - Test: renamed misleading test, added filter-guard coverage Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This was referenced Mar 8, 2026
This was referenced Mar 31, 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
1 participant
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.
Summary
chunks_ftsnow indexessummary,tags,resolved_queryalongsidecontent— keyword searches match enrichment metadata (biggest relevance win, coachClaude 0.4666→0.7+)kg_hybrid_searchfor combined chunk + fact retrieval (was dead code, now connected)format()instead ofdetailparamTest plan
🤖 Generated with Claude Code
Summary by CodeRabbit
New Features
Improvements
Bug Fixes
Tests