ksapi — Knowledge Stack Python SDK

Focus on agents. We handle document intelligence.

Python client for the Knowledge Stack API.

Install

pip install ksapi
# or
uv pip install ksapi

Related repos

• ks-cookbook — 32 production-style agent demos using this SDK and the MCP server.
• ks-mcp — MCP server for agent frameworks.
• ks-sdk-ts — TypeScript counterpart (@knowledge-stack/ksapi on npm).
• ks-docs — central developer docs (Mintlify → docs.knowledgestack.ai).

For most agent use cases, talk to Knowledge Stack via the MCP server, not this SDK directly. This SDK is the right choice when you need admin / write operations or are building infrastructure (ingestion pipelines, schedulers, migration tools).

How this package is maintained

This package is generated from the Knowledge Stack OpenAPI specification by OpenAPI Generator. API-shape issues belong upstream in ks-backend; client-side bugs (import problems, wrapper behavior, release packaging) belong here.

Below this line is the auto-generated reference content.

---

ksapi — Python SDK Quickstart

The official Python client for the Knowledge Stack API. One client authenticates with an API key, ingests documents, asks the agent, and streams chat.

The full per-endpoint reference is generated alongside this package under ./docs/.

Install

pip install ksapi

Authenticate

Mint an API key (in the app under API Keys, or POST /v1/api-keys) and pass it as access_token. The client sends it as Authorization: Bearer <key> on every request:

import ksapi
from ksapi.api.documents_api import DocumentsApi

config = ksapi.Configuration(
    host="https://api.your-host.example",
    access_token="sk-user-...",  # your API key
)
client = ksapi.ApiClient(config)

# Smoke test — list documents you can access
print(DocumentsApi(client).list_documents(limit=10).total, "documents")

Browser/session callers authenticate with the ks_uat cookie instead: ksapi.Configuration(host=..., api_key={"cookieAuth": "<ks_uat value>"}).

Ingest a document and wait until it's searchable

Ingestion is asynchronous. Upload, then poll until the active version's pipeline status is completed — at that point the chunks are embedded and searchable:

import time
from ksapi.api.documents_api import DocumentsApi

docs = DocumentsApi(client)

with open("contract.pdf", "rb") as fh:
    created = docs.ingest_document(
        file=fh.read(),
        path_part_id="<parent-folder-path-part-id>",  # a FOLDER
        name="contract.pdf",
    )

while True:
    version = docs.get_document(created.document_id).active_version
    meta = version.system_metadata
    status = meta.pipeline_state.status if meta and meta.pipeline_state else None
    if status in ("completed", "failed"):
        break
    time.sleep(2)
print("ingestion:", status)

Ask the agent

from ksapi.api.agent_api import AgentApi
from ksapi.models.ask_request import AskRequest

answer = AgentApi(client).agent_ask(AskRequest(prompt="What are the renewal terms?"))
print(answer.text)

Citation-grounded, multi-turn chat runs through threads; each assistant message exposes content.citations. Stream those tokens live below.

Stream chat responses (SSE)

The agent streams over Server-Sent Events at GET /v1/threads/{thread_id}/stream. The generated stream_thread() method buffers the whole response, so consume the stream directly — the body is data: <json> lines terminated by data: [DONE]:

import httpx

def stream_thread(host: str, api_key: str, thread_id: str):
    headers = {"Authorization": f"Bearer {api_key}"}
    with httpx.Client(base_url=host, headers=headers, timeout=60) as http:
        with http.stream("GET", f"/v1/threads/{thread_id}/stream") as response:
            for line in response.iter_lines():
                if not line.startswith("data:"):
                    continue  # skip event:/id:/keepalive lines
                payload = line[len("data:"):].strip()
                if payload == "[DONE]":
                    return
                yield payload  # JSON event — parse with json.loads()

for event in stream_thread(config.host, "sk-user-...", "<thread-id>"):
    print(event)

Errors

Every error response keeps detail and adds a machine-readable code plus the request_id (also returned as the x-request-id header) — quote it to support:

from ksapi.exceptions import NotFoundException

try:
    DocumentsApi(client).get_document("00000000-0000-0000-0000-000000000000")
except NotFoundException as exc:
    body = exc.body  # {"detail": ..., "code": "not_found", "request_id": "..."}
    print(exc.status, body)

The client raises a typed exception per status: BadRequestException (400), UnauthorizedException (401), ForbiddenException (403), NotFoundException (404), ConflictException (409), UnprocessableEntityException (422), and ServiceException (5xx).