Knowledge Bank Documentation

Build intelligent applications with the GuruCloud Knowledge Bank. Semantic search, multi-dimensional schemas, batch ingestion, and MCP server integration — all through a clean REST API and Python SDK.

Base URL: https://www.gurucloudai.com/api/v1/kb • All responses use a {"data": ...} envelope.

Authentication

All API requests require a KB API key passed as a Bearer token. API keys start with kb_ and can be created from the KB Dashboard.

HTTP Header Authorization: Bearer kb_your_api_key_here

Scopes

API keys can have one or more scopes that control access:

Scope	Access
read	Search, list entries, get schema, get stats, MCP definition
write	Add/update/delete entries, batch ingest
admin	Modify schema, delete KB, manage dimensions, generate PATs

MCP Authentication

MCP endpoints accept both KB API keys (kb_...) and Personal Access Tokens (PATs) as Bearer tokens. You can use your existing KB API key directly — no need to generate a separate token. If you prefer a dedicated MCP-only token, use the generate-pat endpoint.

Installation

Shell pip install gurucloud-kb

Requires Python 3.10+. The only dependency is httpx.

Quick Start

Python

from gurucloud_kb import GuruCloudClient

client = GuruCloudClient(api_key="kb_your_api_key")

# List your Knowledge Banks
kbs = client.list_kbs()

# Work with a specific KB
kb = client.get_kb("your-kb-uuid")

# Search
results = kb.search("how does authentication work?")

# Add an entry
kb.add_entry({
    "dimensions": {
        "content": "Auth uses JWT tokens with RS256 signing.",
        "useful_for": "Understanding auth architecture",
    }
})

# Batch ingest
kb.ingest([
    {"dimensions": {"content": "Entry 1"}},
    {"dimensions": {"content": "Entry 2"}},
])

# Get MCP server definition for agent injection
mcp_def = kb.get_mcp_server_definition()

Python (async)

from gurucloud_kb import AsyncGuruCloudClient

async with AsyncGuruCloudClient(api_key="kb_your_api_key") as client:
    # All methods are awaitable
    kb = await client.get_kb("your-kb-uuid")

    results = await kb.search("how does auth work?")

    await kb.ingest([
        {"dimensions": {"content": "Entry 1"}},
        {"dimensions": {"content": "Entry 2"}},
    ])

    mcp_def = await kb.get_mcp_server_definition()

Sync Client

GuruCloudClient

The main entry point. Supports context manager for automatic cleanup.

Method	Returns	Description
list_kbs()	list[KBInfo]	List all KBs owned by the user
get_kb(kb_id)	KnowledgeBank	Get a KB object with pre-bound methods
create_kb(name, ...)	KnowledgeBank	Create a new KB
update_kb(kb_id, ...)	KBInfo	Update name/description
delete_kb(kb_id)	dict	Delete a KB (admin scope)
get_mcp_server_definition(kb_id)	MCPServerDefinition	Get MCP server URL and tools
create_api_key(name, ...)	APIKeyInfo	Create a new API key
list_api_keys()	list[APIKeyInfo]	List all API keys
delete_api_key(key_id)	dict	Delete an API key

Async Client

AsyncGuruCloudClient

Identical API to GuruCloudClient, but all methods are async. Uses httpx.AsyncClient under the hood. Supports async with for automatic cleanup.

Python

import asyncio
from gurucloud_kb import AsyncGuruCloudClient

async def main():
    async with AsyncGuruCloudClient(api_key="kb_...") as client:
        kb = await client.get_kb("my-kb")
        results = await kb.search("query")
        print(results)

asyncio.run(main())

KnowledgeBank Object

Returned by client.get_kb() or client.create_kb(). All methods are pre-bound to the KB's UUID.

Properties

Property	Type	Description
id	str	KB UUID
name	str	Human-readable name
description	str	KB description
entry_count	int	Number of entries
total_queries	int	Total search queries
info	KBInfo	Full metadata dict

Methods

Method	Scope	Description
search(query, k=10, threshold=0.5)	read	Semantic search (string or multi-dim)
add_entry(entry)	write	Add a single entry
ingest(entries, deduplicate=True)	write	Batch ingest (max 100)
list_entries(limit, offset)	read	List entries with pagination
get_entry(entry_id)	read	Get single entry
update_entry(entry_id, updates)	write	Update entry dimensions
delete_entry(entry_id)	write	Delete an entry
get_schema()	read	Get dimension schema
update_schema(schema)	admin	Replace full schema
validate_schema(schema)	read	Validate without applying
add_dimension(dimension)	admin	Add a dimension
remove_dimension(name)	admin	Remove a dimension
get_mcp_server_definition()	read	Get MCP server URL and tools
generate_pat(token_name)	admin	Generate a PAT for MCP auth
get_mcp_config()	read	Get .mcp.json snippet
get_mcp_tools()	read	Get MCP tool definitions
get_stats()	read	Performance statistics
refresh()	read	Re-fetch KB info

Search

The search() method accepts either a simple string or a full multi-dimensional search request.

Simple string search

Python

results = kb.search("how does auth work?", k=5, threshold=0.7)
for r in results:
    print(r["content"], r["combined_score"])

Multi-dimensional search

Python

results = kb.search({
    "dimensions": {
        "content": {"query": "JWT tokens", "weight": 1.0},
        "useful_for": {"query": "debugging auth", "weight": 1.5},
    },
    "k": 5,
    "threshold": 0.6,
})

Schema Management

Each KB has a dimension schema that defines what search dimensions exist and how they're combined.

Python

# Get the current schema
schema = kb.get_schema()
print(schema["dimensions"])  # list of dimension configs

# Add a new dimension
kb.add_dimension({
    "name": "priority",
    "dimension_type": "single",
    "description": "Priority level",
    "searchable": True,
})

# Validate before applying
warnings = kb.validate_schema(new_schema)
if not warnings:
    kb.update_schema(new_schema)

Ingestion

Single entry

Python

kb.add_entry({
    "dimensions": {
        "content": "The API uses rate limiting at 1000 req/hr per key.",
        "useful_for": "Understanding API limits",
        "relevant_systems": ["api", "rate-limiting"],
    },
    "source": "docs",
    "relevant_file_paths": ["routes/api/kb_api.py"],
})

Batch ingest

Python

result = kb.ingest(
    entries=[
        {"dimensions": {"content": "Entry 1", "useful_for": "..."}},
        {"dimensions": {"content": "Entry 2"}},
        # ... up to 100 entries per call
    ],
    deduplicate=True,  # default: skip near-duplicates
)
print(f"Ingested: {result['ingested']}, Errors: {len(result['errors'])}")

Batch ingest uses partial failure handling — successful entries are ingested even if some fail. Check result["errors"] for details.

Events & Conflicts

The Knowledge Bank tracks every deduplication decision made during ingestion. Use events to audit how entries are being deduplicated, view merge reasoning, and inspect conflicts.

Deduplication Events

Each time an entry is ingested, the system compares it against existing entries. The result is one of five actions:

Action	Meaning
`new`	No duplicates found — entry added as-is
`redundant`	Near-exact duplicate — entry skipped
`update`	Partial overlap — existing entry updated with merged content
`conflict`	Contradicting information — requires review
`error`	Processing error occurred

Python

# List all dedup events (paginated)
events = kb.list_events(limit=50, offset=0)
print(f"Total events: {events['total']}")
print(f"Action breakdown: {events['action_counts']}")

# Filter by action type
conflicts = kb.list_events(action="conflict")
for evt in conflicts["events"]:
    print(f"  {evt['content_preview']} (score: {evt['max_similarity_score']})")

# Get full details of a specific event
detail = kb.get_event(conflicts["events"][0]["id"])
print(f"Reasoning: {detail['reasoning']}")
print(f"Merged content: {detail['merged_content']}")
print(f"Similar entries: {detail['similar_entries']}")

Entry Event Logs

For deeper debugging, entry event logs capture the full processing lifecycle of each entry: queuing, hash checks, deduplication search, LLM decisions, and action execution.

Python

# List all event logs for a KB
logs = kb.list_event_logs(limit=100)

# Filter by event type
dedup_logs = kb.list_event_logs(event_type="dedup")

# Filter by specific entry
entry_logs = kb.list_event_logs(entry_id="pending-entry-uuid")
for log in entry_logs["logs"]:
    status = "OK" if log["success"] else "FAIL"
    print(f"  [{status}] {log['event_type']}/{log['event_name']} ({log['duration_ms']}ms)")

Event logs are append-only and ordered by created_at descending. Use entry_id to reconstruct the complete processing history of a single entry.

MCP Integration

Get everything needed to inject your Knowledge Bank's MCP server into an AI agent. Returns the MCP URL, server name, and available tools. Use your KB API key directly as the Bearer token for MCP requests.

Python

mcp_def = kb.get_mcp_server_definition()

# Returns:
# {
#   "type": "http",
#   "url": "https://www.gurucloudai.com/mcp/srv-uuid/mcp",
#   "server_name": "my-kb",
#   "available_tools": ["query_knowledge_bank", "report_learning"],
#   "auth": {"type": "bearer", "note": "Use your KB API key..."},
# }

# Inject into your agent's MCP config using your API key:
agent_config = {
    "mcpServers": {
        mcp_def["server_name"]: {
            "type": mcp_def["type"],
            "url": mcp_def["url"],
            "headers": {
                "Authorization": f"Bearer {api_key}"
            }
        }
    }
}

# Or generate a dedicated PAT (requires admin scope):
pat_info = kb.generate_pat(token_name="My Agent")
# pat_info["token"] is a never-expiring PAT for this MCP server

Error Handling

Python

from gurucloud_kb import (
    GuruCloudClient,
    AuthenticationError,
    NotFoundError,
    RateLimitError,
    APIError,
)

try:
    kb = client.get_kb("nonexistent")
except AuthenticationError:
    print("Invalid API key")
except NotFoundError:
    print("KB not found")
except RateLimitError:
    print("Rate limit exceeded - slow down")
except APIError as e:
    print(f"API error {e.status_code}: {e.message}")

Exception	HTTP Status	When
AuthenticationError	401	Invalid or missing API key
PermissionError	403	Insufficient scope
NotFoundError	404	Resource not found
RateLimitError	429	Rate limit exceeded
APIError	*	Any other API error
ConnectionError	—	Network/timeout error

REST API: Knowledge Banks

GET /banks

List all Knowledge Banks. read

GET /banks/{kb_id}

Get a specific Knowledge Bank. read

POST /banks

Create a new Knowledge Bank. write

{
  "name": "My KB",
  "description": "Optional description",
  "dimension_schema": { ... }  // optional
}

PATCH /banks/{kb_id}

Update KB name/description. write

DELETE /banks/{kb_id}

Delete a KB and all resources. admin

REST API: Entries

GET /banks/{kb_id}/entries?limit=50&offset=0

List entries with pagination. read

POST /banks/{kb_id}/entries

Add a single entry. write

{
  "dimensions": {
    "content": "The main knowledge content",
    "useful_for": "What this is useful for"
  },
  "source": "optional-source-label",
  "relevant_file_paths": ["path/to/file.py"]
}

POST /banks/{kb_id}/entries/batch

Batch ingest up to 100 entries. write

{
  "entries": [
    {"dimensions": {"content": "..."}},
    {"dimensions": {"content": "..."}}
  ],
  "deduplicate": true
}

GET /banks/{kb_id}/entries/{entry_id}

Get a single entry. read

PATCH /banks/{kb_id}/entries/{entry_id}

Update an entry's dimensions. write

DELETE /banks/{kb_id}/entries/{entry_id}

Delete an entry. write

REST API: Search

POST /banks/{kb_id}/search

Multi-dimensional semantic search. read

{
  "dimensions": {
    "content": {"query": "search text", "weight": 1.0},
    "useful_for": {"query": "context", "weight": 0.5}
  },
  "k": 10,
  "threshold": 0.5
}

REST API: Schema

GET /banks/{kb_id}/schema

Get the dimension schema. read

PUT /banks/{kb_id}/schema

Replace the full schema. admin

POST /banks/{kb_id}/schema/validate

Validate a schema without applying. read

POST /banks/{kb_id}/schema/dimensions

Add a dimension. admin

DELETE /banks/{kb_id}/schema/dimensions/{name}

Remove a dimension. admin

REST API: MCP

GET /banks/{kb_id}/mcp-config

Get .mcp.json config snippet. read

GET /banks/{kb_id}/mcp-tools

Get MCP tool definitions. read

POST /banks/{kb_id}/mcp-server-definition

Get MCP server URL, name, and available tools. read

Use your KB API key directly as the Bearer token for MCP requests. Alternatively, generate a dedicated PAT via the endpoint below.

POST /banks/{kb_id}/generate-pat

Generate a Personal Access Token for this KB's MCP server. admin

Body (optional): {"token_name": "My Agent"}. Returns token, server_url, token_name, note. Tokens do not expire. Store securely.

GET /banks/{kb_id}/stats

Get performance statistics. read

REST API: Events

GET /banks/{kb_id}/events

List deduplication events. read

Query params: limit (default 50, max 200), offset (default 0), action (filter: new, redundant, update, conflict, error)

// Response
{
  "events": [
    {
      "id": "uuid",
      "source": "mcp_tools",
      "content_preview": "Auth uses JWT...",
      "max_similarity_score": 0.95,
      "llm_invoked": true,
      "action": "update",
      "created_at": "2026-03-01T00:00:00"
    }
  ],
  "total": 42,
  "action_counts": {"new": 30, "update": 10, "conflict": 2}
}

GET /banks/{kb_id}/events/{event_id}

Get full details of a specific event. read

Returns the complete event including reasoning, merged_content, similar_entries, and execution details.

GET /banks/{kb_id}/event-logs

List entry processing event logs. read

Query params: limit, offset, event_type (lifecycle, hash_check, dedup, action), entry_id (pending entry UUID)

REST API: API Keys

API key management endpoints use session authentication (login cookies), not Bearer token auth. These are for the dashboard UI.

POST /api-keys

Create a new API key. Returns the raw key once.

{
  "name": "My Key",
  "scopes": ["read", "write"],
  "rate_limit_per_hour": 1000,
  "expires_at": "2027-01-01T00:00:00Z"  // optional
}

GET /api-keys

List all API keys (masked).

DELETE /api-keys/{key_id}

Delete an API key.