MarkDB
Connect a client

OpenWebUI

Give an OpenWebUI model MarkDB memory recall as a Tool.

MarkDB ships an OpenWebUI Tool so a model in OpenWebUI can search your memory mid-chat. The tool is a small Python file (tools/openwebui-tool.py in the repo) that calls the Memory API.

What it does

The tool exposes two functions to the model:

  • memory_search(query, scope) -> POST /v1/memory/search -- returns summaries, key facts, decisions, and page ids.
  • memory_get_detail(page_id) -> GET /v1/memory/detail/{page_id} -- drills into a specific result.

(memory_recall is an alias for memory_search.)

Install

  1. In OpenWebUI, go to Workspace -> Tools -> + and paste the contents of tools/openwebui-tool.py.
  2. Configure its valves:
ValveMeaning
markdb_urlYour MarkDB API base URL.
tenant_id / app_id / agent_idThe identity to attribute recall to.
  1. Enable the tool on a model and ask it something you've discussed before.

Auth: self-hosted vs. cloud

As shipped, the tool authenticates with X-MarkDB-Tenant-ID / App-ID / Agent-ID headers, which works against a self-hosted MarkDB running in headers auth mode. MarkDB Cloud requires an API key -- edit _headers() in the tool to send Authorization: Bearer mk_live_... and point markdb_url at https://api.markdb.cloud. See Authentication.

Capture, too

This tool is recall-only. To also capture an OpenWebUI conversation into memory, point OpenWebUI's model connection at the MarkDB proxy (https://proxy.markdb.cloud/v1) with your key, exactly like any other OpenAI-compatible client.