Skip to content

BrainAPI is a knowledge graph–powered AI memory layer that transforms unstructured data into structured knowledge, enabling intelligent search, recommendations, and contextual memory for AI agents and applications.

Fixed coderabbit suggestions

Christianaa8c358eb5ba
diff --git a/src/adapters/graph.py b/src/adapters/graph.py
index 24f40d1..7e9c1ac 100644
--- a/src/adapters/graph.py
+++ b/src/adapters/graph.py
@@ -10,6 +10,7 @@ Modified By: Christian Nonis <[email protected]>
from typing import Dict, List, Literal, Optional, Tuple
from src.adapters.interfaces.graph import GraphClient
+from src.constants.embeddings import Vector
from src.constants.kg import (
IdentificationParams,
Node,
@@ -19,6 +20,7 @@ from src.constants.kg import (
)
from src.adapters.interfaces.embeddings import VectorStoreClient
from src.utils.normalization.list_reduction import reduce_list
+from src.utils.similarity.vectors import cosine_similarity
class GraphAdapter:
@@ -392,8 +394,6 @@ class GraphAdapter:
List[Tuple[Node, List[Tuple[Predicate, Node, List[Tuple[Predicate, Node]]]]]]: The list of 2nd degree hops.
"""
- # TODO: add a way to light weight the output
-
def flatten_node(n):
return (
{"uuid": n.uuid, "labels": n.labels, "name": n.name} if flattened else n
@@ -409,9 +409,18 @@ class GraphAdapter:
nodes = self.get_by_uuids(from_uuids, brain_id)
nodes_by_uuid = {n.uuid: n for n in nodes}
- vs = vector_store_adapter.get_by_ids(
- [n.properties["v_id"] for n in nodes], brain_id=brain_id, store="nodes"
- )
+ v_ids = [
+ n.properties["v_id"] for n in nodes if getattr(n.properties, "v_id", None)
+ ]
+ if len(v_ids) == 0:
+ print("[ ! ] No v_ids found for nodes:", from_uuids)
+ return []
+
+ vs = vector_store_adapter.get_by_ids(v_ids, brain_id=brain_id, store="nodes")
+ if len(vs) == 0:
+ print("[ ! ] No vectors found for nodes:", from_uuids)
+ return []
+
averaged_vector = [
sum(v.embeddings[i] for v in vs) / len(vs)
for i in range(len(vs[0].embeddings))
@@ -419,7 +428,7 @@ class GraphAdapter:
all_fd_nodes = self.get_neighbors(list(nodes_by_uuid.keys()), brain_id=brain_id)
- all_fd_v_ids = [
+ all_fd_v_ids: List[str] = [
fd[1].properties["v_id"]
for fds in all_fd_nodes.values()
for fd in fds
@@ -428,10 +437,10 @@ class GraphAdapter:
all_fd_vs = vector_store_adapter.get_by_ids(
all_fd_v_ids, brain_id=brain_id, store="nodes"
)
- fd_vs_by_uuid = {v.metadata["uuid"]: v for v in all_fd_vs}
+ fd_vs_by_uuid: Dict[str, Vector] = {v.metadata["uuid"]: v for v in all_fd_vs}
all_filtered_fd_uuids = []
- filtered_fd_by_origin = {}
+ filtered_fd_by_origin: Dict[str, List[Tuple[Predicate, Node]]] = {}
for node_uuid, fd_list in all_fd_nodes.items():
fd_nodes_by_uuid = {fd[1].uuid: fd[1] for fd in fd_list}
@@ -461,9 +470,22 @@ class GraphAdapter:
"with_": from_node.description,
},
)
- filtered_uuids = {v["metadata"]["uuid"] for v in filtered}
+ filtered_uuids = {
+ v["metadata"]["uuid"]
+ for v in filtered
+ if cosine_similarity(averaged_vector, v["embeddings"])
+ >= similarity_threshold
+ }
else:
- filtered_uuids = {fd[1].uuid for fd in fd_list}
+ filtered_uuids = {
+ fd[1].uuid
+ for fd in fd_list
+ if fd[1].uuid in fd_vs_by_uuid
+ and cosine_similarity(
+ averaged_vector, fd_vs_by_uuid[fd[1].uuid].embeddings
+ )
+ >= similarity_threshold
+ }
filtered_fd_by_origin[node_uuid] = [
fd for fd in fd_list if fd[1].uuid in filtered_uuids
]
@@ -471,7 +493,7 @@ class GraphAdapter:
all_sd_nodes = self.get_neighbors(all_filtered_fd_uuids, brain_id=brain_id)
- all_sd_v_ids = [
+ all_sd_v_ids: List[str] = [
getattr(sd[1], "properties", {}).get("v_id")
for sds in all_sd_nodes.values()
for sd in sds
@@ -523,9 +545,19 @@ class GraphAdapter:
"with_": from_node.description,
},
)
- reduced_uuids = {v["metadata"]["uuid"] for v in reduced}
+ reduced_uuids = {
+ v["metadata"]["uuid"]
+ for v in reduced
+ if cosine_similarity(averaged_vector, v["embeddings"])
+ >= similarity_threshold
+ }
else:
- reduced_uuids = {sd["metadata"]["uuid"] for sd in sd_vs_with_desc}
+ reduced_uuids = {
+ sd["metadata"]["uuid"]
+ for sd in sd_vs_with_desc
+ if cosine_similarity(averaged_vector, sd["embeddings"])
+ >= similarity_threshold
+ }
second_degree = [
(flatten_pred(sd[0]), flatten_node(sd[1]))
@@ -534,9 +566,6 @@ class GraphAdapter:
and sd[1].uuid not in exclude_set
and sd[1].uuid != from_uuid
]
-
- if similarity_threshold > 0.0:
- if fd_pred["metadata"]["v_id"]
node_hops.append(
(flatten_pred(fd_pred), flatten_node(fd_node), second_degree)
diff --git a/src/core/agents/architect_agent.py b/src/core/agents/architect_agent.py
index 4bbd366..b2fb9ff 100644
--- a/src/core/agents/architect_agent.py
+++ b/src/core/agents/architect_agent.py
@@ -61,6 +61,7 @@ from src.utils.tokens import token_detail_from_token_counts
HISTORY_MAX_MESSAGES = 25
HISTORY_MAX_MESSAGES_DELETE = 8
+MAX_RECURSION_LIMIT = 100
class ArchitectAgent:
@@ -313,7 +314,13 @@ class ArchitectAgent:
response = future.result(timeout=timeout)
for m in response.get("messages", []):
if hasattr(m, "usage_metadata"):
- self.get_token_summary(m.usage_metadata)
+ self._update_token_counts(m.usage_metadata)
+ self.token_detail = token_detail_from_token_counts(
+ self.input_tokens,
+ self.output_tokens,
+ self.cached_tokens,
+ self.reasoning_tokens,
+ )
return response
except FutureTimeoutError:
raise TimeoutError(
@@ -439,7 +446,10 @@ class ArchitectAgent:
)
messages_list.append(user_message)
- return self.agent.invoke({"messages": messages_list})
+ return self.agent.invoke(
+ {"messages": messages_list},
+ config={"recursion_limit": MAX_RECURSION_LIMIT},
+ )
@retry(
stop=stop_after_attempt(max_retries),
diff --git a/src/core/agents/janitor_agent.py b/src/core/agents/janitor_agent.py
index ddd61f7..dc862ed 100644
--- a/src/core/agents/janitor_agent.py
+++ b/src/core/agents/janitor_agent.py
@@ -197,6 +197,7 @@ class JanitorAgent:
flattened=True,
vector_store_adapter=self.vector_store,
brain_id=brain_id,
+ similarity_threshold=0.35,
)
self._get_agent(
diff --git a/src/core/agents/kg_agent.py b/src/core/agents/kg_agent.py
index daf20b1..62e54f4 100644
--- a/src/core/agents/kg_agent.py
+++ b/src/core/agents/kg_agent.py
@@ -168,7 +168,12 @@ class KGAgent:
Update the knowledge graph with new information.
"""
- self._get_agent(identification_params, metadata, brain_id=brain_id)
+ self._get_agent(
+ type_="normal",
+ identification_params=identification_params,
+ metadata=metadata,
+ brain_id=brain_id,
+ )
preferred_entities_prompt = f"""
You must prioritize the extraction of the following entities: {preferred_entities},
@@ -207,7 +212,12 @@ class KGAgent:
Update the knowledge graph with new structured information.
"""
- self._get_agent(identification_params, metadata={}, brain_id=brain_id)
+ self._get_agent(
+ type_="normal",
+ identification_params=identification_params,
+ metadata={},
+ brain_id=brain_id,
+ )
response = self.agent.invoke(
{
@@ -227,7 +237,11 @@ class KGAgent:
return response
def retrieve_neighbors(
- self, node: Node, looking_for: Optional[Union[str, Iterable[str]]], limit: int
+ self,
+ node: Node,
+ looking_for: Optional[Union[str, Iterable[str]]],
+ limit: int,
+ brain_id: str = "default",
) -> RetrieveNeighborsOutputSchema:
"""
Retrieve the neighbors of a node.
@@ -247,6 +261,7 @@ class KGAgent:
"""
self._get_agent(
+ type_="normal",
identification_params={},
metadata={},
tools=[
@@ -254,10 +269,12 @@ class KGAgent:
self,
self.kg,
self.database_desc,
+ brain_id=brain_id,
)
],
output_schema=RetrieveNeighborsOutputSchema,
extra_system_prompt=extra_system_prompt_str,
+ brain_id=brain_id,
)
if isinstance(looking_for, str):
diff --git a/src/core/agents/tools/janitor_agent/JanitorAgentSearchEntitiesTool.py b/src/core/agents/tools/janitor_agent/JanitorAgentSearchEntitiesTool.py
index 097544a..d940d4f 100644
--- a/src/core/agents/tools/janitor_agent/JanitorAgentSearchEntitiesTool.py
+++ b/src/core/agents/tools/janitor_agent/JanitorAgentSearchEntitiesTool.py
@@ -69,13 +69,11 @@ class JanitorAgentSearchEntitiesTool(BaseTool):
)
def _run(self, *args, **kwargs) -> str:
- _query = ""
+ _queries = []
if len(args) > 0:
- _query = args[0]
+ _queries = args[0] if isinstance(args[0], list) else [args[0]]
else:
_queries = kwargs.get("queries", [])
-
- if len(_queries) == 0:
return "No queries provided in the arguments or kwargs"
found_entities: Dict[str, Union[str, List[Node]]] = {}
@@ -116,7 +114,7 @@ class JanitorAgentSearchEntitiesTool(BaseTool):
except Exception as e:
print(f"Vector search failed, using KG results only: {e}")
- if len(found_entities) == 0:
+ if len(founds) == 0:
found_entities[_clean_query(_query)] = (
"Knowledge graph does not contain any entities that match the query"
)
diff --git a/src/core/search/entity_info.py b/src/core/search/entity_info.py
index c7ae5dc..6bfc95d 100644
--- a/src/core/search/entity_info.py
+++ b/src/core/search/entity_info.py
@@ -200,8 +200,13 @@ class EventSynergyRetriever:
days_ago = 0
if node.happened_at:
- days_ago = (datetime.now() - node.happened_at).days
-
+ try:
+ happened_at = node.happened_at
+ if isinstance(happened_at, str):
+ happened_at = datetime.fromisoformat(happened_at)
+ days_ago = max(0, (datetime.now() - happened_at).days)
+ except Exception:
+ days_ago = 0
recency = 1 / (1 + np.log1p(days_ago)) if days_ago > 0 else 1.0
base_score = similarity * 0.6 + recency * 0.2
diff --git a/src/core/search/entity_sibilings.py b/src/core/search/entity_sibilings.py
index ad5c3aa..9592220 100644
--- a/src/core/search/entity_sibilings.py
+++ b/src/core/search/entity_sibilings.py
@@ -33,10 +33,10 @@ class EntitySinergyRetriever:
target_node_vs = vector_store_adapter.search_vectors(
target_embedding.embeddings, store="nodes", brain_id=self.brain_id
)
-
+ if not target_node_vs:
+ return None, []
target_node_id = target_node_vs[0].metadata.get("uuid")
target_node = graph_adapter.get_by_uuid(target_node_id, brain_id=self.brain_id)
-
if not target_node:
return None, []
diff --git a/src/lib/llm/client_small.py b/src/lib/llm/client_small.py
index 1a0e01d..5ca18aa 100644
--- a/src/lib/llm/client_small.py
+++ b/src/lib/llm/client_small.py
@@ -96,7 +96,7 @@ class LLMClientSmall(LLM):
return self._langchain_model
def generate_text(
- self, prompt: str, max_new_tokens: int = 100000, timeout: int = None
+ self, prompt: str, max_new_tokens: int = 65536, timeout: int = None
) -> str:
if not prompt or len(prompt.strip()) == 0:
return "Input prompt is empty"
@@ -151,7 +151,7 @@ class LLMClientSmall(LLM):
def generate_json(
self,
prompt: str,
- max_new_tokens: int = 100000,
+ max_new_tokens: int = 65536,
max_retries: int = 3,
timeout: int = None,
) -> dict:
diff --git a/src/lib/neo4j/client.py b/src/lib/neo4j/client.py
index eb2b6b7..7ad9e61 100644
--- a/src/lib/neo4j/client.py
+++ b/src/lib/neo4j/client.py
@@ -199,9 +199,10 @@ class Neo4jClient(GraphClient):
for node in nodes:
identification_dict = {"name": node.name}
+ merged_metadata = {**(metadata or {}), **(node.metadata or {})}
all_properties = {
**(node.properties or {}),
- "metadata": metadata,
+ "metadata": merged_metadata or None,
}
if identification_params:
@@ -228,7 +229,9 @@ class Neo4jClient(GraphClient):
]
for attr in attributes:
- if getattr(node, attr, None):
+ if attr == "metadata":
+ continue
+ if getattr(node, attr, None) is not None:
all_properties[attr] = getattr(node, attr)
property_assignments = []
@@ -463,7 +466,10 @@ class Neo4jClient(GraphClient):
"""
cypher_query = """
MATCH (n) WHERE n.uuid = $uuid
- RETURN n.uuid as uuid, n.name as name, labels(n) as labels, n.description as description, properties(n) as properties
+ RETURN n.uuid as uuid, n.name as name, labels(n) as labels, n.description as description,
+ properties(n) as properties,
+ n.polarity as polarity, n.happened_at as happened_at, n.last_updated as last_updated,
+ n.observations as observations, n.metadata as metadata
"""
self.ensure_database(brain_id)
result = self.driver.execute_query(
@@ -485,7 +491,10 @@ class Neo4jClient(GraphClient):
"""
cypher_query = f"""
MATCH (n) WHERE n.uuid IN ["{'","'.join(uuids)}"]
- RETURN n.uuid as uuid, n.name as name, labels(n) as labels, n.description as description, properties(n) as properties
+ RETURN n.uuid as uuid, n.name as name, labels(n) as labels, n.description as description,
+ properties(n) as properties,
+ n.polarity as polarity, n.happened_at as happened_at, n.last_updated as last_updated,
+ n.observations as observations, n.metadata as metadata
"""
self.ensure_database(brain_id)
result = self.driver.execute_query(cypher_query, database_=brain_id)
@@ -497,10 +506,31 @@ class Neo4jClient(GraphClient):
description=record.get("description", ""),
properties=record.get("properties", {}),
polarity=record.get("polarity", "neutral"),
- happened_at=record.get("happened_at", None),
- last_updated=record.get("last_updated", datetime.now(timezone.utc)),
- observations=record.get("observations", []),
- metadata=record.get("metadata", {}),
+ **(
+ {"happened_at": record.get("happened_at", None)}
+ if record.get("happened_at", None) is not None
+ else {}
+ ),
+ **(
+ {
+ "last_updated": record.get(
+ "last_updated", datetime.now(timezone.utc)
+ )
+ }
+ if record.get("last_updated", datetime.now(timezone.utc))
+ is not None
+ else {}
+ ),
+ **(
+ {"observations": record.get("observations", [])}
+ if record.get("observations", []) is not None
+ else {}
+ ),
+ **(
+ {"metadata": record.get("metadata", {})}
+ if record.get("metadata", {}) is not None
+ else {}
+ ),
)
for record in result.records
]
@@ -542,7 +572,10 @@ class Neo4jClient(GraphClient):
cypher_query = f"""
MATCH (n{labels_str}) {where_str}
- RETURN n.uuid as uuid, n.name as name, labels(n) as labels, n.description as description, properties(n) as properties
+ RETURN n.uuid as uuid, n.name as name, labels(n) as labels, n.description as description,
+ properties(n) as properties,
+ n.polarity as polarity, n.happened_at as happened_at, n.last_updated as last_updated,
+ n.observations as observations, n.metadata as metadata
"""
self.ensure_database(brain_id)
@@ -601,10 +634,13 @@ class Neo4jClient(GraphClient):
)
cypher_query += """
- RETURN n.uuid AS uuid, n.name AS name, labels(n) AS labels, n.description AS description, properties(n) AS properties, r AS rel,
- CASE WHEN startNode(r) = n THEN 'out' ELSE 'in' END AS direction,
- type(r) AS rel_type, r.description AS rel_description, properties(r) AS rel_properties, r.flow_key as rel_flowkey, r.uuid as rel_uuid,
- c.uuid AS c_uuid, c.name AS c_name, labels(c) AS c_labels, c.description AS c_description, properties(c) AS c_properties
+ RETURN n.uuid as uuid, n.name as name, labels(n) as labels, n.description as description,
+ properties(n) as properties,
+ n.polarity as polarity, n.happened_at as happened_at, n.last_updated as last_updated,
+ n.observations as observations, n.metadata as metadata, r AS rel,
+ CASE WHEN startNode(r) = n THEN 'out' ELSE 'in' END AS direction,
+ type(r) AS rel_type, r.description AS rel_description, properties(r) AS rel_properties, r.flow_key as rel_flowkey, r.uuid as rel_uuid,
+ c.uuid AS c_uuid, c.name AS c_name, labels(c) AS c_labels, c.description AS c_description, properties(c) AS c_properties
"""
if limit:
cypher_query += f" LIMIT {limit}"
@@ -973,7 +1009,10 @@ class Neo4jClient(GraphClient):
cypher_query = f"""
MATCH (n)
{"WHERE " + " AND ".join(filters) if filters else ""}
- RETURN n.uuid as uuid, n.name as name, labels(n) as labels, n.description as description, properties(n) as properties
+ RETURN n.uuid as uuid, n.name as name, labels(n) as labels, n.description as description,
+ properties(n) as properties,
+ n.polarity as polarity, n.happened_at as happened_at, n.last_updated as last_updated,
+ n.observations as observations, n.metadata as metadata
SKIP {skip}
LIMIT {limit}
"""
diff --git a/src/services/api/controllers/kg.py b/src/services/api/controllers/kg.py
index 0dbe1fa..2dfda6f 100644
--- a/src/services/api/controllers/kg.py
+++ b/src/services/api/controllers/kg.py
@@ -4,7 +4,7 @@ Created Date: Wednesday December 31st 2025
Author: Christian Nonis <[email protected]>
-----
Last Modified: Wednesday December 31st 2025 7:45:55 pm
-Modified By: the developer formerly known as Christian Nonis at <[email protected]>
+Modified By: Christian Nonis <[email protected]>
-----
"""
@@ -21,6 +21,8 @@ async def get_hops(
):
"""
Get the hops in the graph for a given query.
+
+ Currently only supports 2nd degree hops.
"""
text_embeddings = embeddings_adapter.embed_text(query)
data_vectors = vector_store_adapter.search_vectors(
@@ -37,5 +39,6 @@ async def get_hops(
flattened=flattened,
vector_store_adapter=vector_store_adapter,
brain_id=brain_id,
+ similarity_threshold=0.0,
)
return hops
diff --git a/src/services/api/controllers/retrieve.py b/src/services/api/controllers/retrieve.py
index 5d4faa5..e48d896 100644
--- a/src/services/api/controllers/retrieve.py
+++ b/src/services/api/controllers/retrieve.py
@@ -147,7 +147,6 @@ async def retrieve_neighbors(
and not v.id.replace(
"-", ""
).isalpha() # likely not a UUID if all numeric (may have hyphens for uuid standard)
- and v.id.isdigit()
)
}
fd_v_neighbors_ids = list(fd_v_neighbors_embeddings_map.keys())
@@ -244,7 +243,7 @@ async def retrieve_neighbors_ai_mode(
if not node:
raise HTTPException(status_code=404, detail="Entity not found")
- result = kg_agent.retrieve_neighbors(node, looking_for, limit)
+ result = kg_agent.retrieve_neighbors(node, looking_for, limit, brain_id)
ids = [neighbor.uuid for neighbor in result.neighbors]
descriptions = [neighbor.description for neighbor in result.neighbors]
diff --git a/src/services/api/routes/system.py b/src/services/api/routes/system.py
index 00b073f..71b9d3a 100644
--- a/src/services/api/routes/system.py
+++ b/src/services/api/routes/system.py
@@ -4,7 +4,7 @@ Created Date: Monday December 1st 2025
Author: Christian Nonis <[email protected]>
-----
Last Modified: Monday December 1st 2025 10:12:48 pm
-Modified By: the developer formerly known as Christian Nonis at <[email protected]>
+Modified By: Christian Nonis <[email protected]>
-----
"""
@@ -32,3 +32,27 @@ async def create_brain(request: CreateBrainRequest):
Create a new brain
"""
return await create_new_brain_controller(request)
+
+
+@system_router.get(path="/brains/{brain_id}/reset") # TODO
+async def reset(brain_id: str):
+ """
+ Resets the brain
+ """
+ raise NotImplementedError("Not implemented")
+
+
+@system_router.get(path="/brains/{brain_id}/delete") # TODO
+async def delete(brain_id: str):
+ """
+ Deletes the brain and all its data
+ """
+ raise NotImplementedError("Not implemented")
+
+
+@system_router.post(path="/brains/{brain_id}/create-backup") # TODO
+async def create_backup(brain_id: str):
+ """
+ Creates a backup of the brain and returns a task ID
+ """
+ raise NotImplementedError("Not implemented")
diff --git a/src/workers/app.py b/src/workers/app.py
index d05f270..f7abc2a 100644
--- a/src/workers/app.py
+++ b/src/workers/app.py
@@ -48,7 +48,7 @@ ingestion_app.conf.update(
task_time_limit=86400,
task_soft_time_limit=86400,
broker_transport_options={
- "visibility_timeout": 7200,
+ "visibility_timeout": 86400,
"fanout_prefix": True,
"fanout_patterns": True,
},