Docs / Get started

Quickstart

Developer preview

Three steps: get a token, run a search, ask a question. Endpoints marked “Available” work today; those marked “Planned” will ship with the open core.

1
Get a token
The token is in the data directory’s token file. It’s generated automatically on first start and readable only by the current user.
Terminal
cat <data-dir>/token
# Every /api/v1 endpoint requires it; listens on 127.0.0.1:8090 only by default
2
Search
Search your material in natural language. Results include the file, the location and the passage, and don’t depend on file names.
Terminal
curl http://127.0.0.1:8090/api/v1/search \
  -H "Authorization: Bearer $TOKEN" \
  -d '{"query": "payment terms in the supplier contract", "top_k": 5}'
3
Ask
Streamed back. With search on, it answers only from your files and cites its sources.
Terminal
curl -N http://127.0.0.1:8090/api/v1/chat/stream \
  -H "Authorization: Bearer $TOKEN" \
  -d '{"messages": [{"role": "user", "content": "What was decided at the meeting last week?"}],
       "search": true}'

API status

APIWhat it doesStatus
MCP · ownward-mcp
Lets MCP-capable agents use search, Q&A and the library directly; tools with side effects are not exposed.
Available
POST /api/v1/search
One search across documents, images, audio and video.
Available
POST /api/v1/search-image
Search images by image.
Available
POST /api/v1/search/video-clips
Locate the exact segment in a video.
Available
POST /api/v1/chat/stream
Q&A streamed over SSE, with sources.
Available
POST /api/v1/generate
Image and video generation. A length the machine can’t finish is refused on the spot, and the supported maximum is returned.
Available
GET /hub/index.json
Scenario catalog; verified by sha256 before saving.
Available
SDK · Python / JS / C++
Wraps the endpoints above as client libraries for each language.
Planned
OpenAI-compatible API
Existing tools connect by changing the endpoint address.
Planned
MCP connectors
Connect third-party tools and data sources through MCP.
Planned

Models

Every kind of task has a default choice that you can replace in settings. Cloud models are optional and off by default.

Search
Text & multimodal
Embeddings · reranking · sparse · document-page vision
Images
Image understanding
Image-text semantics · faces · OCR · captions
Audio
Audio understanding
Transcription · word alignment · speakers · events
Video
Video analysis
Shot detection · detection & tracking
Chat
Q&A and planning
Local chat models
Generate
Image & video
Images · video · speech synthesis
Docs
Layout understanding
Page structure · element-level understanding
Swappable
No lock-in
Switch by type in settings
Join the developer preview
Get the preview docs and samples; you’ll be among the first to hear when the SDK and connectors ship.
Sent only when there’s news.