DeepSeek V4 just launched — competitive with GPT-5, costs a fraction. Their API is OpenAI-compatible, so integration should be straightforward.
Currently DEVONthink has no DeepSeek option. Please add it as a native provider.
There are no current plans to add more native providers. This is why we implemented the OpenAI (Compatible) option.
When I use the OpenAI-compatible option, the ‘Model’ field remains greyed out and I cannot select or manually enter a model.
It also makes no difference whether I append the v1 base URL or not.
In addition to the chat model, Deepseek v4 is also unavailable in the other AI functions, such as ‘Summary’.
The API key is valid.
DEVONthink expects the URL of the completions endpoint, usually ending with /v1/chat/completions
1 Like
Thank you very much. I’m not sure if my feedback is appropriate here.
Here is my experience.
All files to be analysed are sent by DEVONthink to Deepseek (or other providers).
When dealing with sensitive documents (such as reviewing medical reports), you really have to want that or be prepared to accept it.
But I am not.
My alternative is to run a local Hermes Agent instance.
This does not stand out because it constantly improves through use as a self-learning system. Using a preflight script, I can also prevent sensitive data from being transmitted.
Hermes developed the script itself and registered it as a plugin.
It would, of course, be ideal if I could integrate my local Hermes Agent instance as an LLM provider in Devonthink.
Your Hermes agent can search, retrieve, and file Devonthink documents using Applescript.
That’s why I’m asking if it would be possible to integrate Hermes with DEVONthink.
Yes it can
1 - Hermes can write Applescript integrations
2 - An unofficial MCP server for DT exists - Hermes supports MCP servers
3 - DevonTech advises “Stay Tuned” for further developments
1 Like